Note too that “lively debate” means something different to statisticians from normal people. You would be unlikely to split your sides laughing at conversations about whether building a model to use clinical data to predict outcomes (a core statistical activity), counts as “doing” or “using” math.

Whatever: my undergraduate degree required a lot of math, and my PhD thesis made use of lots of algebra and calculus. I call myself what I am: a statistician. But I do think of myself as belonging to the wider mathematical brotherhood. I believe that the inductive (statistical inference) and deductive (mathematical) logics are all we need to explain — in the sense of “make valid claims about” — the universe. Mastery of these fields are among our species’ proudest achievements.

All of which makes me worried, I think is the word, to read of the impact artificial intelligence and machine learning is having on (at least some) academic mathematicians. Recently, AI has started constructing proofs for some of the trickier unproven theorems. Which leads to the question: if a machine can prove a theorem, what’s the point of trying to do it oneself?

Kirwin Hampshire, who is in the process of working on his own PhD at the University of Victoria (in the department of Mathematics and Statistics, ha!), has just published a Substack essay in which he confesses that the ability of AI to solve problems is causing nothing less than severe existential angst. He feels “a profound spiritual crisis” and, in his worst moments of paranoia, wonders if it be the “express goal of these [AI] companies to make me kill myself”.

His essay is beautiful. To do him the disservice of summarizing it in a paragraph, he argues that the process of learning and proving is as important as the outcome, and that a machine-generated mathematics is no mathematics at all.

That reminded me of music, and I might suggest this to Kirwin as part-antidote to his fears (though he won’t like much of what I say). I am very fond of the Bach two-part and three-part inventions and have multiple recordings of these by some of the ablest pianists ever recorded: Glenn Gould and Andras Schiff are favorites. 

Suggested reading

AI and the false consciousness trap

By Yanis Varoufakis

Nothing I fumble out at the keyboard (I’m a constant piano student) will ever touch the divine musicality of Bach’s genius or the technical mastery of my favorite virtuosi. But it really doesn’t matter. Because few things give me more pleasure than trying. Few things feel more satisfactory than to master a complex passage even though I know no-one will ever hear it other than myself — and that even I would prefer to listen to Glenn Gould doing it properly.

In my analogy, the existence of digital recordings of music are the AI machines in Kirwin’s nightmare: they obviate the need for the task which gives my life meaning (“I am a piano student”) and pleasure (“I need to hear this music”). If a machine can “do” Bach for you, why try and do it yourself? Yet the pleasure persists, as does the choice to practice. I cannot prove this, but I believe it: a robot will one day produce music that sounds Bach-like. I do not believe it will improve upon him.

I suspect some similar “behavioral” consequence might happen to much of mathematics. A machine will be able to do algebra at least as well as a human. But that won’t rob humans of their need to do the algebra for themselves, nor diminish the beauty of any particular proof.

To which Kirwin might reply: so what? You’re talking about a hobby. I’m talking about the pitface of human endeavor, about what it means to be a mathematician. As he says “Mathematicians are paid to prove theorems” and, if you don’t need them for that, do you need them at all?

I cannot comprehend — this might be a failure of imagination — how the answer to that question could ever be “No, we do not need them”, not while there are humans in the world and a human society about which to be concerned. Math is special because it’s universal; no other discipline — save, perhaps, music — comes close. It is true that a machine could prove theorems in such a way as to be understandable and useful only to other machines, but if we reach that point then not only mathematicians should worry about superfluousness. In the meantime, humans will need to understand what the AI is “proving” and I can’t imagine how that could be achieved were those humans not themselves mathematicians.

Back to my own early practice at math. The point of my thesis was to take a noisy, distorted picture of something important (like a remotely-sensed heart scan) and reconstruct an approximation to the real object. In Chapter 4, I ended a few pages of algebra with this melancholic sentence: “Unfortunately we have not been able to prove this Lemma [a proposition].” In other words, I couldn’t demonstrate whether Chapter 4’s “great idea” had legs.

Having read Kirwin’s essay, I wondered if an AI could pick up where I’d left off in 1993. I wish I hadn’t. Here is what Grok told me:

You correctly identified the obstacle. You tried to adapt Eggermont’s argument (his Lemma 6.1 and the subsequent KL + residual Lyapunov function). The natural ordered-subset extension leads to the candidate inequality (your Lemma). That inequality does not hold in general, and the extra diagonal projector, which “switches on” only the current subset, prevents the global residual from decreasing monotonically. Consequently the Lyapunov argument collapses exactly where you left it.

There was a lot more. TL;DR: you couldn’t prove it because it was obviously unprovable, and you should have known this. It was quite clear that Grok already “knows” more about image reconstruction algorithms and their associated mathematics than I could ever hope to master. I think I preferred feeling sad, to feeling both sad and inadequate. Which is, of course, a version of Kirwin’s point.

But in any case none of the work I produced in 1993 is even vaguely relevant today, because the field has moved on. Ironically, image reconstruction was one of machine learning’s early successes: trained models can make better guesses at true images than my Pooterish attempts in Chapter 4.

But even without machine learning, I doubt many people’s theses from 30-odd years ago bear re-reading. I could make a joke here about some more recent contributions to the sociological literature, but I’ll rise above that. (Almost.)

Suggested reading

Must data centres be so ugly?

By Tom Ough

As with Bach, so with PhDs: the process is what matters. Learning how to read literature; how to build a hypothesis and investigate it, mathematically and empirically: this is why we train people to doctoral level in Mathematics and Statistics.

And the more machines produce technology or any other solution to any other problem, the more, I should think, we will need men and women like Kirwin.

I did, though, have long dark nights of the soul during those PhD; years, in fact, when I felt overwhelmed with futility. I remember walking towards the Math building in University Gardens and then just sort of veering off at the last minute, because of the sick feeling in my stomach. Not because of a machine, but because of my inadequacy.

Most PhD students have spent nearly 20 years of school and undergraduate life being told: “This is what you need to understand in order to pass an exam,” and most of us are extremely good at that way of learning. The PhD environment in mathematics and statistics — no classmates, no solutions to tutorials, possibly no solution at all to the problem you’re investigating (see Chapter 4 for details!), no one to tell you if the roll of the dice you’ve just made for the next three years will have any outcome beyond failure — is very different. It would almost be bizarre for a young person in such a situation not to develop angst, an existential dread.

And yet the sun rises, and yes, the work has purpose. Kirwin, your abilities will be needed in the age of AI more than ever before. Maybe just switch off your computer and close your notebook for a while, and think of something else. It was in those long dark nights, I recall, that I first fell in love with Bach.

This week, Lanier joined Florence Read to discuss AI, the possibility of machine consciousness, and why he still has faith in humanity. Below is an edited transcript:

Florence Read: You recently wrote an essay for the New Yorker with the seemingly phlegmatic title, “There is no AI”. Does that mean you don’t think recent developments are a problem?

Jaron Lanier: I actually have publicly stated that I think we could use the new technologies (as well as other technologies) to destroy ourselves. My difference with my colleagues is that I think the way we characterise the technology can have an influence on our options and abilities to handle it. And I think treating AI as this new, alien intelligence reduces our choices and has a way of paralysing us. An alternate take is to see it as a new form of social collaboration, where it’s just made of us. It’s a giant mash-up of human expression, opens up channels for addressing issues and makes us more sane and makes us more competent. So I make a pragmatic argument to not think of the new technologies as alien intelligences, but instead as human social collaborations.

FR: But as someone who works at Microsoft, at the heart of the AI revolution, is it not easier for you take the more pragmatic view over the hysterical?

JL: I have a really unusual role in the tech world. It shouldn’t be unusual; I think it should be more common. Essentially, I am speaking my mind honestly, even though I’m on the inside of the castle instead of on the outside throwing stones at the castle. In my opinion, both positions should be well-manned. I don’t think there’s any perfect way to handle anything. One is always somewhat compromised. Microsoft and I have come to an accord, where I have, what you might call, academic freedom. I speak my mind, I speak things as I see them but I also don’t speak for the company. And we make that distinction. It allows me to maintain my public intellectual life but also work inside.

I don’t necessarily find agreement with everybody I work with, nor do I find absolute disagreement. For instance, Sam Altman from OpenAI really liked my New Yorker piece. I don’t think he agrees with it entirely but he said he agrees with it mostly. That’s great. I think having some degree of openness within the big tech companies is a healthy thing. Within Microsoft, there are now a few other figures who at least somewhat speak their minds. I’m hoping that this demonstration that a tech company can be successful while allowing essentially free speech within its research community, can be a precedent that other tech companies follow. I’d like to see Google and Meta and Apple do a little bit more of that.

FR: What’s odd about the AI discussion is that so many of the people working on it — including Sam Altman — are also the ones sharing their deep existential fears about what AI will do to humanity.

JL: I have tried to understand that myself for decades. I think part of it is, we simultaneously live in a science-fiction universe, where we’re living out the science fiction we grew up with. If you grew up on the Terminator movies, and the Matrix movies and Commander Data from Star Trek, naturally what you want to do is realise this idea of AI. It just seems like your destiny. But then another part of you is thinking, “But in most of those stories, with Commander Data being the exception, this was horrible for mankind.” It feels responsible to acknowledge that it could be horrible for mankind, and yet at the same time, you keep on doing it. It’s weird, and I believe the approach that I’ve proposed of not thinking of what we’re doing as alien intelligences but rather a social collaboration, is the way through that problem, because it’s a way of framing it that’s equally valid, but actionable. But within the tech world, giving up those childhood science-fiction fantasies that we grew up with is really hard for people.

FR: Of course, we used to call the internet a “Wild West”, which played into this mythology — as though there’s a cowboy in every one of these computer scientists who wants to find this new frontier.

JL: I think that’s true too. I grew up in rural New Mexico in the Sixties, when it was still not that economically developed. So I actually got to experience a little bit of the tail end of the Wild West. And I can assure you that it was miserable, and it’s not something anybody would want, but the version of it in the movies is very appealing. And it does bring up a sort of a strange gender-identity connection. Recently, I was on a morning TV talk show in the US, and one of the hosts was a woman who said to me, “It just seems like there’s a lot of male fantasy in the AI world. Shouldn’t there be more women AI leaders?” And I said, “Well, there are some spectacular women AI leaders, and actually, there does tend to be some sort of a difference where the women seem to be a little more humanistic.” In the YouTube version of that, they cut out the whole exchange about women, and I called and asked about it and they said, “It just seemed like a niche question, so we cut it out.” And it’s not, it’s a very central question.

Suggested reading

Nick Cave on Christ and the Devil

By Nick Cave and Freddie Sayers

FR: I do have a kind of sense that there is something Promethean about it, that for many men, this is the first time they’ve been able to create life from nothing.

JL: When I was a kid, I always used to say AI is really just womb envy. And having had a child and seeing what it’s actually like for a woman to bear child, I no longer have womb envy. I now appreciate it’s actually a rather difficult process for the mother and I didn’t know that when I was a young man. I will say that it’s not just men, but it tends to be men who haven’t had kids yet who might have that desire to create life in the computer.

FR: You distinguish between two types of AI, between this “alien entity” to which many of your colleagues seem to attribute the spark of life, and your version, which is, in fact, just a network of connections between humans.

JL: The term AI is very wiggly, and gets applied to all kinds of things. But usually these days when we talk about AI, we’re talking about these large AI models like the GPT programmes. What they are is giant mash-ups of human creations. If you ask one of these programmes to create you a new image — like, I’d like to see London as if it were a cross between London and Gurwat — it can probably synthesise that. But the way it does so is by using the classifiers that it uses to identify the images that match the components of your request, and mashing them up. Managing the whole scale so it can happen quickly is not so simple, but the basic idea is pretty simple.

Now, I happen to think that’s a great capability with a lot of uses. I love the idea of computers just getting more flexible. It creates the possibility of saying, “Can you reconfigure this computer experience to work for somebody who’s colourblind?” instead of demanding that people conform to computer design. There’s a potential in this flexibility to really improve computation on many levels and make it much better for people. But, if you want to, you can perceive it as a new intelligence. And, to me, if you perceive it as a new intelligence, what you’re really doing is shutting off yourself in order to worship the code, which I think is exactly the wrong thing. It makes you less able to make good decisions.

You’ve probably heard of the Turing test, which was one of the original thought-experiments about artificial intelligence. There’s this idea that if a human judge can’t distinguish whether something came from a person or computer, then we should treat the computer as having equal rights. And the problem with that is that it’s also possible that the judge became stupid. There’s no guarantee that it wasn’t the judge who changed rather than the computer. The problem with treating the output of GPT as if it’s an alien intelligence, which many people enjoy doing, is that you can’t tell whether the humans are letting go of their own standards and becoming stupid to make the machine seem smart.

FR: So we haven’t reached computational consciousness, a computer with sentience?

JL: The sentience of others is always a matter of faith. There’s no way to be certain about whether someone else has interior experience in the way that you do. I presume that you do, but I can’t know. There is a mystical or almost supernatural element in which we have internal experience — or at least I do, but I can’t make you believe I do. You have to just believe on your own that I do. That faith is a very precious thing and there’s no absolute argument that you should or shouldn’t believe that another person has interior experience, or sentience or consciousness, or that a machine does. Faith is not fundamentally rational, but there is a pragmatic argument, as I keep on repeating, to placing your faith in other people instead of machines. If you care about people at all, if you want people to survive, you have to place your faith in the sentience of them instead of in machines as a pragmatic matter, not as a matter of absolute truth.

FR: Is the only distinction between human and machine sentience, then, a faith in the power of the human soul versus the fact that that computer is just amalgamating information?

JL: It’s a matter of faith that has pragmatic implications. Just to say something is a matter of faith doesn’t mean that the choice of faith is entirely arbitrary, because it can be pragmatic as well. So, if not believing in people increases the chance that people will be harmed, I think the same is the case with this technology. Not believing that machines are the same as people increases the chance that people will be harmed. Cumulatively, we should believe in people over computers, but that’s not an absolute argument based on logic or empiricism, which I don’t think is available to us. There’s a bit of a skyhook thing here, like the problem of “why should you stay alive instead of committing suicide?” It’s applied to the whole species: “Why should we continue this human project; why does it matter?”

I’ve come to something that’s a little bit like the argument attributed to Pascal — you might as well believe in God, just in case it’s real and there’s heaven and hell. I don’t buy that particular argument; I’m not concerned about heaven or hell. However, I do think that the continuation of us in this timeline, in this world, and this physicality, is something I’d like to commit to. I think we might be something special. And so in that way, I’d like to apply faith to us and give us a chance, and that does involve the demoting of computers. But when we demote computers, we can use them better. Demoting AI allows us to not mystify, and that allows us paths to explaining it, to controlling it, to understanding it, to using it as a scientific exploration of what language is. There’s so many reasons to demote it that are practical, that the faith in it as a mystical being just actually seems kind of stupid and wasteful and pathetic to me.

Suggested reading

AI is a false prophet

By Jacob Howland

FR: But can we demote something that has potentially more power than us already? Most of us are already subordinated to computers in our everyday lives.

JL: People are capable of being self-destructive, idiotic, wasteful and ridiculous, with or without computers. However, we can do it a little more efficiently with computers because we can do anything a little more efficiently with computers. I’ve been very publicly concerned about the dehumanising elements of social media algorithms. The algorithms on social media that have caused elevated outbreaks of things that always existed in humanity, but there’s just a little more: vanity, paranoia, irritability. And that increment is enough to change politics, to change mental health, especially in impoverished circumstances around the world. It’s just made the world worse incrementally. The algorithms on social media are really dumbass-simple — there’s really not a lot there. And so I think your framing of it as more powerful than us is incorrect. I think it’s really just dumb stuff. It’s up to us to decide how it fits into human society.

The capacity for human stupidity is great, and, as I keep on saying, it’s only a matter of faith whether we call it human stupidity or machine intelligence. They’re indistinguishable logically. So I think the threat is real. I’m not anti-doomist. I just ask us to consider: what is the way of thinking that improves our abilities and improves our thinking — that gives us more options, gives us more clarity? And it involves demoting the computer.

There’s a lot of work to do technically. We can create explanations for what so-called “machine intelligence” is doing by tracing it back to its human origins. There have been a number of very famous instances of chatbots getting really weird with people. But the form of explanation should be to say, “Actually, the bot was, at that point, parodying something from a soap opera, or from some fanfiction.” That’s what’s going on. And in my opinion, there should be an economy in the future where, if there’s really valuable output from an AI, the people whose contributions were particularly important should actually get paid. I believe there’s a new extension to society that’s very creative and interesting, rather than this dismal prospect of everybody being put out of work. Transparency in mash-up technology can only come from revealing the people whose expressions were mashed-up. But if policies are based on the idea that we now have this new “supernatural artificial entity”, there’s no sensible way to resolve that.

FR: You didn’t sign the open letter demanding a hiatus in accelerating AI development, which was signed by Elon Musk and Sam Altman. Was that not appealing to you as an idea?

JL: My reason for not signing it is that it fundamentally still mystified the technology. It took the position that it’s a new alien entity. When you think it’s an alien entity, there’s no way to know how to help. If you have an alien entity, what regulation is good? How do you define harm? As much as the GPT programmes impress people, they don’t actually represent ideas. We don’t know how to define these things, all we can do is mash things up so that they conform with classifiers.

FR: So they can’t do the philosophical work of thinking?

JL: They could mash-up philosophers in ways that might be interesting. If you say, “Write an essay as if Descartes and Derrida collaborated”, something might come out that’s provocative or interesting. But there’s no actual representation inside there. And getting provocative or interesting mash-ups is useful, but you can’t set policy by it because there’s not actually any meaning. There’s no fundamental representation inside these things and we just have to accept that as reality. We don’t know what meaning is and we can’t represent meaning.

FR: Your argument relies on the idea that if we define this technology differently, then we will have more power over it, or at least we’ll have more understanding of it. Are we not just self-comforting here with a rhetoric about it being a human technology rather than something we can’t control?

JL: I don’t think that’s the case. It’s proposing a more concrete and clarified path of action. It’s very demanding of people and it’s not comforting at all. It demands that everybody involved on a technical or regulatory level do much more than they have. I suspect many people would prefer the mystical version because it actually lets them off the hook. The mystical version just lets you sit there and apprehend, and express awe at our own inventions. What I’m talking about demands action. It’s not comforting and it shouldn’t be.

FR: Do you think humans need to take more accountability for their part in developing a potentially malign form of AI? If it does go off the rails, wouldn’t it be because we’ve set it up to do so?

JL: One comparison is the disasters with the Boeing 737 Max. The flight correction module in it was the source of two terrible air disasters in which hundreds of people died. But what actually happened involved the way they sold it, the way they withheld information about it (depending on how much you paid them), the way they trained people for it, the way the documentation was created. It’s the surrounding stuff that created the disaster, not the core capability, which probably has been useful in general. In the same way, with this large-model AI, it’s not the thing itself, it’s the surrounding material that determines whether it’s malignant or not.

When you deploy it, under the assumption that it’s an alien new intelligence — that it’s a new entity with its own point of view that should be treated as a creature instead of a tool — you greatly increase the chances of a scenario similar to the one that that befell passengers on the Boeing planes. I think that’s a real possibility. The malignancy is in the surrounding material, not in the core technology, and that’s extremely important to understand. I don’t think anybody has claimed that the flight path correction module shouldn’t have existed. I think what people are saying is that the pilots should have been well-informed, well-trained, and the ability to control it should have always been included, not only for those who paid more. And if you have chatbots, and you tell people, “This is an intelligent companion, you should be able to date it, you should be able to trust it”, then the chances of something really bad happening increase.

FR: Isn’t the main worry then that this sort of technology might fall into the hands of someone who has malign intent against a group or country. I’m thinking particularly about the situation in Ukraine.

JL: Russia has one of the worst records on misusing the internet and algorithms. It’s documented that Russia created enormous numbers of fake accounts, of fake bots, in order to sow divisions within the US. And of course, it’s attempting those things in Ukraine. I worry a little bit more about China, because Russia doesn’t quite have the resources to pull off very large-model projects right now. It’s not that easy to do — you need huge computational resources. So I worry a little bit about China using, to be very blunt, data from TikTok on the morning of a Taiwan invasion or something like that. That’s imaginable. I’ve talked to a lot of people in the Chinese world, and I think almost all are actually much more conscientious and better-intentioned than we might imagine, but there’s always somebody in any country in any situation. I do worry about it, and the antidote to it is universal clarity, context, transparency, which can only come about by revealing people, since revealing ideas is impossible because we don’t know what an idea is.

Suggested reading

Left-brain thinking will destroy civilisation

By Iain McGilchrist and Freddie Sayers

FR: We’ve established though that we already live with artificial intelligence. How has that already changed us?

JL: Our principal encounter with algorithms so far has been in the construction of the feeds we receive in our apps. It’s in whether we get credit or not and other things like that — whether we get admitted to university or not, or whether we’re sent to prison, depending on what country we’re talking about. Algorithms have transformed us. I would hope that the criticisms of them that I and many others — Tristan Harris, Shoshana Zuboff — have put forward have illuminated and clarified the issues with algorithms in the previous generation. But what could happen with the new AI is a worse version of all of that. Given how bad that was, I don’t think the doomerists are entirely wrong. I think we could confuse ourselves into extinction with our own code. But, once again, in order for us to achieve that level of stupidity, we have to believe overly in the intelligence of the software, and I think we have a choice.

FR: You’re a composer as well as a computer scientist. Do you think that there is going to be a shift in the way in which we prioritise organic or manmade art?

JL: We are entering a world of what I call “data dignity”. A musician might provide music directly, or might provide antecedent music that’s mashed-up within an algorithm, but that musician is still known and credited. And we’ve seen that already for decades now — somebody might provide the beats, somebody else might provide samples, etc. There’s already this sense of construction and mash-up, especially in hip-hop, but also just in pop music lately. That has not destroyed musicians, not as long as it’s acknowledged and transparent. I think, as with Boeing, it’s the surrounding material. If we choose to use mash-up algorithms to hide the people from whom the antecedent stuff came from, then we do damage. But the thing doing the damage is hiding ourselves, not the algorithm itself, which is actually just a simple dumb thing. I think there are a lot of good things about an algorithmic mash-up culture in the future. Every new instance of automation, instead of putting people out of work, could be thought of as the platform for a new creative community.

FR: Won’t that dull our eyes to the beauty of real art and culture?

JL: What I see in culture is, as long as people understand what’s going on, they find their way. Synthesisers haven’t killed violins. There was a fear that they would, and as long as people know the difference, as long as there’s honesty and transparency about what’s going on, we can go through seasons of things being a little more artificial and then less so. That becomes a cultural dynamic and I trust people to handle that well.

FR: I might sound a bit like someone booing at Bob Dylan going electric, but, if you take Spotify, it’s almost totally wiped out independent music. There have been major technological advances in music that have obliterated creativity at those lower, more maverick levels of the industry.

JL: You’re absolutely correct about Spotify. In fact, at the dawn of the file-copying era, I objected very strenuously to this idea. There was a cultural movement about open source and open culture, which was stealthily funded by Google and other tech companies, and the Pirate Parties in Europe. People thought everything should be a mash up and we didn’t need to know who the musician was and they didn’t need to have bargaining power in a financial transaction. That was a gigantic wrong turn, and it was a wrong turn that we can’t afford to repeat with AI because it becomes amplified so much that it could really destroy technology. I completely agree with you about Spotify but, once again, the availability of music to move through the internet was not the problem. It’s the surrounding material. What really screwed over musicians was not the core capability, but this idea that you build a business model on demoting the musician, demoting the person, and instead elevating the hub or the platform. And so we can’t afford to keep on doing that. I think that is the road that leads to our potential extinction through insanity.

Suggested reading

Why is my chatbot hitting on me?

By Kathleen Stock

FR: It sounds like the answer to a lot of these problems comes down to human greed?

JL: I think humans are definitely responsible. Greed is one aspect of it, but it’s not all of it. I don’t necessarily understand all human failings within myself or anybody else, but I do feel we can articulate ways to approach this that are more practical, more actionable and more hopeful. That has to be our first duty. I think this question of diagnosing each other and saying, “This person has womb envy”, or whatever has some utility, but not a lot, and can inspire reactions that aren’t helpful. So I don’t want to emphasise that too much. I want to emphasise an approach, which we can call “data dignity”, and which opens options for us and makes things clearer.

FR: What is the best case scenario if we follow that route?

JL: What I like about the new algorithms is that they help us collaborate better. You could have a new and more flexible kind of a computer, where you can ask it to change the way you present things to match your mood or your cognition under your own control, so that you’re less subservient to the computer. But another thing you can do is you can say, “I have written one essay, my friend’s written another essay, they’re sort of different. Can you mash them up 12 different ways so we can read the mash-ups?” And this is not based on ideas, it’s based on the dumb math of combining words as they appeared, in order, in context. But you might be able to learn new options for consilience between different points of view that way, which could be extraordinary. Many people have been looking at the humanistic AI world, the human-centred AI world, and asking, “Could we actually use this to help us understand potential for cooperation and policy that we might not see?”

FR: So, oddly, it might break us out of our tribes and offer some human connection?

JL: It’s like if a therapist says, “Try using different words and see if that might change how you think about something.” It’s not directly addressing our thoughts, but on the surface level it actually can help us. But it’s ultimately up to us, and there’s no guarantee it’ll help, but I believe it will in many cases. It can help us improve our research, it can help us improve a lot of our practices, and, as long as we acknowledge the people whose ideas are being matched up by the programmes, it can help us even broaden participation in the economy, instead of throwing people out of work as so often foretold. I think we can use this stuff to our advantage, and it’s worth it to try. If we try to use it well, the most awful ideas about it turning into the Matrix or Terminator, become vanishingly unlikely, as long as we treat it as a human project instead of an alien intelligence.

The little melody in question is of course a pre-set. Its sequence of notes is planned in advance, and once the key is pressed, the machine may be relied upon to do only the thing it has been programmed to do. The melody, it goes without saying, is no Bach fugue. It is simple, naïve, kind of dumb; and within the context of the song, it is utterly compelling.

Several conditions — technological, cultural, historical — had to fall into place in order for this melodic interlude, with its verbal introduction, to come across to the critical listener as an expression of genius. All of these conditions might be cited in response to any philistine tempted to declare, of the pressing down of that special key, that “I could have done that too”. We are used to hearing such petulant ressentiment, especially in connection with the 20th-century avant-garde in the figurative arts: “I could have entered a urinal in an exhibition, too”; “I could have painted an all-white monochrome, too”; etc. The simplest response is, “Yes, but you didn’t”.

But this is probably too succinct to teach the philistine anything new about how art works. It is going to be particularly important to sensitise people to the logic and dynamics of artistic creation as we enter the new era of AI-generated artworks and are bombarded with new images created by machines following the input of a mere verbal prompt from a human being: DALL·E 2 being a prime example. Among other unpleasant prospects this new technology opens up, we will have to brace ourselves for a new wave of declarations from philistines telling us of the things they could have done too. It will be useful to have something new to say to them in response.

While technical competence is unmistakable, genius has a strange way of horseshoeing with idiocy, which is why we need critics to help us tell the difference. Jimi Hendrix lighting his guitar on fire is genius; a suburban kid in 1967 attempting the same, in his parents’ garage, is idiocy. Context matters, and by 1967 Hendrix was almost uniquely positioned to change the world, at least a little bit, with a gesture that would have been merely wasteful if attempted by nearly any other human alive.

Catchphrases that we associate with a particular artist or character are almost always banal in the extreme, but when Jimmy Durante says “Ha-cha-cha-cha” or Jimmie Walker says “Dyn-o-mite!”, we are witnessing a gesture of something like genius as well, or at least of idiosyncrasy, which is akin to genius in that it flows from an individual’s unique ingenium or natural aptitude, and proves to be mostly unteachable and entirely untransferable. I can of course go around evoking “dynamite” whenever I am enthused about something, but it will come across as perplexing at best and pathetically imitative at worst. When Jimmie Walker said it by contrast, he became himself a flash of light, a rupture in our ordinary expectations of the world. I’m a critic and I’m here to tell you that that’s genius.

Wit, too, like idiosyncrasy, is a species of the genus of genius. Once Winston Churchill sat at a formal dinner, dissatisfied with the food and the service. When the champagne arrived, he is reported to have remarked: “Finally, something warm.” I could try out that same phrase in a million different settings, and perhaps one or two of these would find it landing as wit; in the vast majority of cases it would be a simple non-sequitur, in a few it would be an accurate, literal, and unfunny description of the thing or event before me. With another million attempts, perhaps an occasion would arise in which the line approached the quality it had in Churchill’s mouth. Genius, along with its lesser cousins, I mean, is a fluid and variable thing. But one thing it definitely is not is technical competence.

More from this author

Your Fitbit has stolen your soul

By Justin E. H. Smith

Philistines go in for photorealist painting; they imagine it testifies to “progress” in representation since the time of the Dutch Masters, since its lines are sharper, its objects come across on the canvas as more like objects the way we find them in “the real world” (that is, the world mediated by the physics of light and by the physiology of vision, which ordinarily we scarcely understand, or even think about, instead taking the affordances of our sense of sight to be straightforward reports of external reality itself). But of course what photorealist paintings actually resemble are photographs, and in this respect to learn to paint the world photorealistically is to create machine-aided paintings, where the machine is, namely, the camera. If David Hockney is correct, cameras of a sort, camerae obscurae, have been deployed since the Renaissance to assist in capturing the world on canvas. We have had machine-aided visual art for a very long time, but the use of machines in one’s artistic labours has never been necessary or sufficient for the creation of works of lasting aesthetic merit.

Could that be about to change? It’s 2022 and there’s a new thing going around on Twitter. You can click on a link, and arrive at a site that invites you to upload a photo of yourself, after which you will get a “reverse AI portrait” that spells out, in language, what it is the computer is “seeing”. I give it a try, and this is what it tells me: “A man with blonde hair and a black shirt, a character portrait, inspired by Eric Peterson, reddit contest winner, precisionism, doing the bateman stare, taken in the early 2020s, photograph of Christopher Walken, Thom Wasselmann, without glasses, John Waters, face with artgram.”

I’m not impressed. It strikes me that the machine is just looking for something, anything, to say. It’s smart enough to know that many people get excited when they hear mention, any mention, of Christopher Walken; they launch into bad imitations and everyone loves it. Bring up his name along with some other nonsense phrases and some bare factual descriptions of my physiognomy, and you’re sure to get a few retweets. This little fragment of language is optimised for virality, not for saying anything true. It is just more social-media bullshit; it is not art.

It’s 2030, after the war, and I’m at an AI art fair in Vladivostok. There’s some guy, some hot new artist on the scene, who makes customised portraits of visitors based on a special sequence of keywords that he enters in secret, on the basis of his spontaneous reading of our physical, behavioural and apparent moral qualities. I wait in line impatiently, and when my turn comes and my portrait finally uploads to my retinal “screens”, I am confronted with an image of myself that appears somehow warped, perverse, somewhat beastly, and yet, I have to admit, also, profoundly familiar and true. I don’t know how the artist and his machine managed to pull it off. I’m impressed. The dude got me.

More from this author

The philosophy of Count Dracula

By Justin E. H. Smith

Kraftwerk’s special key, as a pre-set, could only do one thing, and cameras too, have generally been limited to registering what is in front of their lens in accordance with the mechanical parameters, exposure time and so on, set in advance by the photographer. One genuinely novel feature of AI art is that it is not like this. Two different users can enter the same prompts into the same programme at different moments, and will get two different images in response. This places the human-machine collaborative work of creation somewhat closer to the activity of the Turkish fortune-teller who compels a rabbit to sniff out one of several possible pieces of folded paper that contains, in writing, a hint of the customer’s future. A good niyetçi will know how to manipulate his rabbit to give interesting results, even if the choice of paper depends in the end on the rabbit’s own inscrutable perception of the world.

Divination is not ordinarily considered art, though it is acknowledged to rely on artifice, and though much art incorporates aleatoric principles that share quite a bit with the work of the tea-leaf reader or the caster of knucklebones. The particular arrangement of tea leaves is generally taken to be informative, at least in a make-believe way, rather than aesthetic, but the boundary here shifts easily, and indeed we might see the category of the aesthetic as a special variety of informativeness, where the patterns or representations we see tell us something, even if we cannot say precisely what that is.

AI art will supervene, is already supervening, upon several different categories of human endeavour — the clever technology-dependent stunt, the joke, divination — and will continue to please us and disappoint us in just the same way that these other endeavours do. It is simply not a meaningful question to ask whether any AI output may appropriately be categorised as art. Art is artifice plus, one hopes, a hint of genius, or something close to that. Such hints can shine through, as I’ve attempted to remind you, in the most unlikely, indeed the silliest places. There is of course no reason why AI should not also be such a place.

Whether we enshrine AI art within our society, giving it a place in art fairs, giving out prizes for it, and so on, is entirely up to us. There is nothing about the technology in itself that can guide us in determining whether we “should” do this or not. Art is what we value, what we attend to, and deem worthy of aspiration to excellence; this is why some activities come forward as art in some cultures, but not in others — the tea ceremony is art in Japan, but if I hastily serve you a mug of Lipton, I am not giving you a poor contribution to an art-form — I’m just giving you tea.

Suggested reading

The art world's lost sense of humour

By Kat Rosenfield

Quite apart from the normative question of whether AI output should be considered art under some circumstances — a question that is unanswerable — I am confident in predicting that we will have AI art. In this art there will be occasional flashes of genius, or something like it, against a general background of cultural overproduction of shit. This has also been the general balance throughout the history of photography, television, and cinema. The most significant change with the rise of AI comes with the human relinquishing of control over the pre-set parameters.

But this does not so much move us into uncharted territory, as it simply assimilates AI art to other forms of human creativity, such as divination, which in the past have used objectively random processes to make their determinations, but have also often used animals. In many ways, as with an engine and its “horsepower”, we are simply finding new ways to transfer work from animals to machines, both of which are entities that are sufficiently like us to make determinations that will resemble our own choices, but sufficiently unlike us for their determinations to come across to us as “weird”, as suitable moments to say “whoah”. That “whoah” can be interpreted as an aesthetic experience, with sufficient institutional backing.

I’m not looking forward to any of this. I am going to stick with my vintage technologies, and my aesthetic orientation will remain forever backward-looking. But even less than the new era of AI art am I looking forward to the inevitable wave of renewed debate around the inane and empty question of whether this new variety of culture-embedded and technology-dependent activity ought to count as art. It’s like asking whether a hamburger can count as breakfast. There is no ontological rift between breakfast and lunch; breakfast is what we say it is, and if you feel like your 8am burger is not doing it for you, this tells you something only about your expectations, and not about the world.

There is similarly no ontological divide separating artworks from “the commonplace”. Art tends to emerge at the sites of social value, of care. If we think AI art is a bad idea, then we might slow its ascendancy by grounding our care in other spheres of human life than those shaped by cutting-edge technologies. But this is almost certainly not going to happen.

One man who has done this countless times, to countless women, without their consent, is one Mr Deepfakes. As the founder of the most prominent deepfake porn website in existence, he has chosen to remain anonymous. He built the burgeoning community “from scratch” as a side hustle, after deepfake porn was banned from Reddit in 2018. It is, he claims, a place where “users affected by the ban could go”, to ensure the technology wasn’t unfairly “squashed”. MDF, as he styles himself, cites a commitment to free speech, and a desire to advance machine learning, as his sole motivations.

But in Deepfake Porn: Could you be next? MDF comes across as a man struggling with “the more moral aspect” of his work. On the one hand, he ardently claims to respect women (“100%”!) and insists this principle is reconcilable with his passion project. On the other, he has no plans to tell his wife — who would “probably be against” deepfake porn, “to put it bluntly” — about his work. “I’m afraid of how it would affect her, knowing I work on something like this.” He admits that “the content is actually in a grey area, and I think we’re on a fine line”.

Despite this ethical “indecision”, MDF’s website is thriving. It has amassed more than 20,000 deepfake videos of women loosely defined as “celebrities”, who are divided into 23 categories that include “Cosplay”, “Threesome” and “Asian Celeb”. Each day, an average of 25 new videos are added by a team of deepfake porn producers. There are 13 million original visitors who view this content every month, 10,000 of whom are online at any given moment.

Suggested reading

Porn will destroy you

By Sarah Ditum

In these videos, the facial expressions, mannerisms, and idiosyncrasies of the deepfaked subject do not belong to the victim whose face we see. They are the creation of a male fantasy. It not only looks like the victim is performing porn, when they never have, but also that they are engaging in the producer’s favourite sexual acts. Their identity is hijacked. When a victim sees herself embodied in the form of a porn performer, she describes it to me as a feeling of extreme disassociation. That is her face, but that is not her body.

Mr Deepfakes rakes in a high four-figure profit per month, mostly from ad revenue. He’s probably earning upwards of $100,000 a year from the venture. This money goes mostly to “maintaining the servers”.

Most of his effort, meanwhile, goes to imposing strict ethics on his website. Or so he claims. MDF recites his boundaries to minimise harm like a script: no porn niches that are “defamatory”; only “celebrities” allowed; the age limit is absolute; and producers must “make sure people know that these are fake videos” by ensuring “every video is watermarked”. But ultimately, MDF’s defence always comes down to separating the videos’ potential consequences from their creators. “We’re not all bad people!”

MDF claims he is simply an advocate of technological progress. He is, he says, facilitating the improvement of AI, while “other communities like Reddit wouldn’t allow it”. It just so happens that the best way to do this is through the “porn niche”: a convenient assumption that MDF treats as an awkward but inevitable fact. The “community” forum, the section of the website where content creators discuss how to improve results, is therefore what he cares about most. It fails to generate any revenue, at least for now. But this is hardly a charitable venture. This “community” is working to develop a tool that’s being used as a weapon against women.

Suggested reading

Test article

By RICHARD V. REEVES

And the ethical boundaries imposed by Mr Deepfakes fall apart on close inspection. There is no clear definition, for instance, of a “celebrity”. It includes women in the “mainstream media”, like those who appear in “Hollywood films”, but also “social media influencers”. And of course, “politicians are in the public domain”, so they can be targeted too. It is, in short, any woman with a public life. Many of these female “celebrities” already receive so much sexualised publicity in the media that MDF’s community think they are “fair game”. Or in the words of one user on the forum: “If you plaster your face everywhere and thirst trap me into seeking out your visage, then I’m just gonna deepfake you into porn.”

The majority of Mr Deepfakes community are “probably men”, he admits. Male users who are becoming radicalised by getting to “pick and choose aspects of different girls”, and digitally manipulate them into performing sexual acts.

Disturbingly, one regular poster on the site admits to deepfaking his co-workers. “Walking into work after having deepfaked these women, it did feel odd, but I just controlled my nerves. I can act like nothing is wrong, no one would suspect a thing.” The video is for his eyes only, he insists; it now exists on his hard drive, which he sees as merely an extension of his imagination. Asked if he would create a custom deepfake of an ordinary woman for someone else, he replied: “From a moral standpoint, yeah, I don’t think there’s anything that would stop me.” He went on to make a deepfake of someone after a Zoom call.

Suggested reading

The horror of deepfake nudes

By NINA SCHICK

The forums on MDF’s site seethe with misogyny, illustrating how utterly ineffective his ethical code is. “She didn’t let me smash her during highschool so cool. ima just deepfake u on porn and masturbate to it? CHECKMATE HOE,” one user writes. “It is never better when deepfake kicks in, and you get to make your dream celebrity be a mindless robot, and obey masters orders,” posted another.

Even those who have built a living on the porn industry are alarmed by the havoc deepfake technology could wreak on women’s lives. “The principal concern I have is how quickly the technology is evolving. It’s like a runaway freight train,” warns Viktor Zafirovski, a reviewer at the world’s largest porn directory website. Not only is it increasingly accessible, but they are getting easier to create.We will reach a point when “everyone will have to be paranoid about sharing their image,” says Viktor. And having made a small fortune developing this latest tool in the misogynist’s arsenal, even the man at the helm of the deepfake porn industry agrees. “It will be so convincing, that eventually you can’t identify what’s real and what’s fake,” MDF says. “You know, it’s scary to me as well.”

Underneath this cloak of concern for women, Mr Deepfakes must know that he empowers misogynists every day. “Porn runs the world right now. And it’s not something that I agree with with,” he says. Regardless, it’s something he leverages, even as he strives to distance himself from the culture he enables.

MDF doesn’t like to imagine how his wife would feel “if she found a video of herself on the internet — or maybe even a deepfake porn video of someone she knows”. But there are hundreds of women who, if they visited his site, might have the kind of horrifying realisation he wouldn’t want his loved ones to experience. The most frustrating thing is, he admits to feeling some discomfort. “I think I need to, you know, look deep down and see what I’m okay with,” he reflects. But his site remains live, and he keeps making money from it.

My book The AI Does Not Hate You, covers some people’s attempts to answer that question. They think that there is a strong chance that it will all go wrong in some profound ways – ways which could lead, in a worst-case scenario, to human extinction.

And it might not happen the way you expect – by the AI breaking its programming or ‘going rogue’ – but simply by the machine doing exactly what it was told, in some highly literal and unpredictable way.

Suggested reading

How AI will terminate us

By Peter Franklin

In his new book, Novacene: The Coming Age of Hyperintelligence, James Lovelock – one of the great names of 20^th-century science, and the creator of the Gaia hypothesis that Earth should be treated as a single self-regulating organism – argues differently. His key idea is that the Anthropocene, the informally defined geological era characterised by human influence, is coming to an end, and we are entering the “Novacene”, which is not a brand of dental anaesthetic but a new epoch characterised by artificial intelligence.

We will be replaced by a new form of life, says Lovelock, which will think many thousands of times faster than us, and which may suffer us to live – or alternatively may not. But either way, it will carry the torch for sentient life. Eventually, he says “organic Gaia will probably die”, replaced by an electronic ecosystem.

The book leaps through a potted history of life, the universe and everything to get there, but its short chapters are filled with hugely confident claims about things that the author doesn’t feel the need to back up. “It is clear” that life can only have evolved once in the observable universe, Lovelock breezily tells us in the first few pages: “our existence is a freakish one-off”.

More from this author

Are you truly unfakeable?

By Tom Chivers

But, of course, that’s a huge and ongoing fight. You need to do more work to make claims of that magnitude. Similar things happen every few pages and with some odd leaps of logic, too. For example, in the early 19th century, the discovery of anomalies in Mercury’s orbit led some to posit the existence of an unseen planet, Vulcan. A hundred or so years later Albert Einstein instead showed that the answer was that Newtonian physics wasn’t quite accurate when dealing with very high speeds and masses, and used that as the basis of his new relativity theory.

But for Lovelock, the Vulcan hypothesis was a failure of “cause-and-effect logic” – science, he says, should rely more on intuition. But Einstein, of course, worked it out with maths! I’m sure he had intuitive leaps, but the discovery of general relativity is (to me, at least) an example of using “cause-and-effect logic” with astonishing brilliance and with world-changing results.

All of these things may be defensible positions. But – unforgivably for a science book – there are no references, no footnotes, no way of checking his positions at all beyond googling relevant-sounding phrases. At one point he drops in as an aside that “extraordinarily, bumblebees have been seen to play football”, which is the sort of sentence that makes you do a double-take.

I googled it and it turns out that honey bees (not bumblebees) can be trained to pull a ball into a “goal” to win a nectar reward, and other bees learn it more quickly by watching, and inevitably that became “bees play football” in the popular press. It’s such a minor detail, but it annoyed me.

And when it gets to the subject of AI, that lack of ballast and rigour makes the book essentially useless as a guide to the future.

Lovelock does have lots of interesting ideas – some of which I agree with entirely. For instance, progress, howsoever measured, is speeding up; it took millions of years to go from lizard to seabird, but only a few decades to go from the Wright Flyer to the Concorde. Economic growth or information-processing speed has also got faster in this exponential way: technological improvements make new technological improvements easier, so the AI future could go through enormous revolutions in hours rather than years.

But lots of it reads like some free-associating Alan Ginsberg fever dream – Lovelock imagines animals “plucking freshly charged batteries from solar-powered trees”, and “huge transmitters sited at the poles broadcasting junk mail, unwanted advertisements, banal entertainment and misinformation” into space to keep the planet cool, or cyborg scientists exhibiting humans like we do plants at Kew Gardens.

He says that any future AIs will have to keep the Earth cool because they, like us, won’t be able to survive if it gets much warmer. But if the AIs are as powerful as he suggests, then it’s not clear that they’ll need to keep the Earth in any sort of recognisable form at all. What’s to stop them from clearing out the inconveniently warm atmosphere altogether, or for that matter turning the entire Earth into memory banks and processors?

This idea that “hyperintelligent” AIs will need to maintain a Gaia-like homeostasis definitely needs more explaining. It’s intriguing enough, but, with no references, I can’t see what Lovelock is basing it all on, other than his own musings.

Most importantly for the Novacene hypothesis, though, is his optimistic idea that these AIs will be the next generation of intelligent, aware beings – “understanding creatures” that allow the cosmos to contemplate itself, as he says. (Reminiscent of the Niels Bohr quote, “A physicist is just an atom’s way of looking at itself.”)

That could be what happens; but in the end, AIs will be computers that do what we tell them to do, and whether that ends well or badly depends entirely on how we build them and what we tell them to do.

Suggested reading

The true cost of digital convenience

By Mark Piesing

It may well be that you can build amazingly powerful AIs which have no consciousness, no moral value in their own right, but are just very good at completing the tasks we give them. If we are replaced by machines like that, then they could whir away and do astonishing things, colonise the galaxy, solve scientific questions and build wonders, but there would be no conscious beings to observe it or care about it. None of it would matter. It would, in the words of the philosopher and AI theorist Nick Bostrom in his own book Superintelligence, be “a Disneyland with no children”.

That’s not inevitable either, of course. And nor are any of the other scenarios – doom, utopia, human zoos. The problem is that Lovelock seems to think that Novacene is inevitable. I think his fundamental problem is, ironically, that he still essentially thinks of AIs as human – not in form, but in mind; they may think faster, but they will still care about the things we care about, they will still get bored and impatient, they will still be individually distinguishable beings rather than avatars of some greater hive-mind.

The AI future could transform human prospects in profound ways – as extreme as ending world poverty and illness, if it goes well, or exterminating us all, if it doesn’t. It is absolutely vital that we spend time thinking about it, and trying to make the first scenario more likely than the second.

But if, like Lovelock, we assume that it’s all inevitable – whether positive or negative – then we abdicate our responsibility for trying to make that future positive rather than negative.

That, of course, is Adam Smith in the Wealth of Nations, noting the ever-present threat that collusion poses to competition.

Price fixing is not in the public interest, which is why governments intervene to prevent it. But what is a fair price? One yardstick might be that pricing should be consistent – the vendor selling the same item at the same price irrespective of who’s buying it. However, that’s complicated by what you mean by the same item – for instance, is a taxi ride at rush hour or chucking-out time the same as one taken at 11 o’clock in the morning? Arguably, these are different services that justify different prices. If demand exceeds supply at certain times, then rationing is inevitable by one means or another.

Suggested reading

Where's the outrage about online price discrimination?

By Rhiannon Williams

In the first century, Publilius Syrus said that “everything is worth what its purchaser is willing to pay for it”. In supermarkets, you can see how some items, such as premium brands of chocolate and ice cream, regularly cycle between two price points. This doesn’t appear to be about the balance of supply and demand, but about selling to two different types of consumer – those who’ll pay a modest premium for an occasional treat and those who’ll pay what it takes to get what they want when they want it. Then there’s the fact that in some markets people simply expect to pay more for the same item than those in other markets: the point at which consumers make the distinction between a reasonable mark-up and an outrageous rip-off is not consistent and thus neither are prices.

Looking ahead, we can expect a whole extra dimension of complexity – given that prices are increasingly set by computers not humans. E-commerce systems now gather so much relevant information that only a pricing algorithm can process it all. But to what end?

In a somewhat alarming article for VoxEU, Emilio Calvano, Giacomo Calzolari, Vincenzo Denicolò, Sergio Pastorello describe their research into the implications:

“Already in 2015, more than a third of the vendors on Amazon.com had automated pricing… and the share has certainly risen since then – with the growth of a repricing software industry that supplies turnkey pricing systems, even the smallest vendors can now afford algorithmic pricing.”

Moreover, the most advanced programs aren’t like the “the traditional revenue management systems long in use by such businesses as airlines and hotels”, but are much more autonomous and unpredictable – using AI technology to learn the most effective pricing strategies (effective at extracting the most money from the consumer, that is).

So, here’s the big question: if these AI systems are free to discover the most profitable pricing strategies without human supervision, will they collude with one another to fix prices?

Suggested reading

Who's protecting the vulnerable from predatory markets?

By Polly Mackenzie

Today’s AI systems, though impressive, are basically trial-and-error engines, lacking even the slightest shred of consciousness. Surely, then, it would be beyond their capabilities to spontaneously organise themselves into a price-fixing cartel – especially under the conditions of a complex and competitive marketplace.

Calvano and his colleagues decided to put that assumption to the test – and here’s what they discovered:

“…we construct[ed] AI pricing agents and let them interact repeatedly in controlled environments that reproduce economists’ canonical model of collusion, i.e. a repeated pricing game with simultaneous moves and full price flexibility. Our findings suggest that in this framework even relatively simple pricing algorithms systematically learn to play sophisticated collusive strategies.”

Uh-oh. Adam Smith was right – pricing strategies on the part of multiple market players do tend toward cosy collusion not cut-throat competition. If the mental befuddlement of “merriment and diversion” is no obstacle, that’s because consciousness isn’t required at all – mindless calculation is all you need.

That collusion can spontaneously arise without criminal intent makes it all the more of a regulatory nightmare:

“…the algorithms leave no trace of concerted action – they learn to collude purely by trial and error, with no prior knowledge of the environment in which they operate, without communicating with one another, and without being specifically designed or instructed to collude. This poses a real challenge for competition policy.”

The authors conclude that those calling for laws to restrict the use of such algorithms have a point.

I would add that online retailers should also be calling for regulation. In the short-term, most of the the AI firepower may on their side, but how long before ordinary consumers have their own AI agents – algorithms designed to find the ideal price at which to buy rather than sell?

One can easily envisage a digital assistant that could trawl through online retail sites to find the lowest price for a particular item. Various price comparison sites already exist of course, but imagine an AI agent capable of learning from experience and adapting its methods. Now, what if collusion turns out to be the most effective strategy for the AI agents on our side? Instead of the spontaneous organisation of price-fixing cartels, we’d have the spontaneous organisation of consumer boycotts to force prices down.

Suggested reading

No, computers haven't learned how to cheat

By Peter Franklin

Without the help of AI, the coordination costs of collective action and asymmetric access to information tend to put consumers at a disadvantage to the businesses we buy from. Individually, we are small and they are big. We, therefore, depend on unimpeded competition in market place and/or effective regulation to get a fair deal.

Retailers should think very carefully before using price-gouging algorithms to undermine the status quo. Any additional advantage they gain could be lost (and then some) if they end up unleashing the power of consumer AI.

Covered late last year by Devin Coldewey of the TechCrunch website, it concerns a type of AI system called a generative adversarial network or GAN.

A GAN consists of two halves: a ‘discriminator’ and a ‘generator’. By having labelled items of data fed to it, the discriminator learns to identify the patterns that correspond to the labels – so that it can go on to identify similar patterns in unlabelled data. An example would be a system set up to identify particular faces in CCTV footage.

Suggested reading

What will 'deep fake' imagery do to politics?

By Peter Franklin

The generator works the other way round – it generates or modifies data to create simulated patterns with the objective of convincing an observer that they correspond to a particular label. A rather sinister example might be the simulation of CCTV footage containing recognisable faces. Feedback from the observer enables the system to learn to produce evermore convincing fakes.

By pitching its generator and discriminator components against one another, a GAN computerises the feedback process, thus greatly speeding up the rate of learning.

The TechCrunch story concerns an experiment by Stanford and Google researchers, in which a GAN was given the task of converting aerial photography into the sort of image you might see on a sat nav app:

“The intention of the researchers was, as you might guess, to accelerate and improve the process of turning satellite imagery into Google’s famously accurate maps. To that end the team was working with what’s called a CycleGAN — a neural network that learns to transform images of type X and Y into one another, as efficiently yet accurately as possible, through a great deal of experimentation.”

The system did a pretty good job – turning the fine detail of the photograph into the simplified and stylised components of a map.

Then the researchers tried the process in reverse – getting the system to convert the map into a convincing simulation of an aerial photograph. That’s when things started getting creepy:

“In some early results, the [AI] was doing well — suspiciously well. What tipped the team off was that, when the agent reconstructed aerial photographs from its street maps, there were lots of details that didn’t seem to be on the latter at all. For instance, skylights on a roof that were eliminated in the process of creating the street map would magically reappear when they asked the agent to do the reverse process…”

To be clear, in undertaking the reverse task the computer only had the map to work from, not the original photograph. So how on Earth did it recreate details that were present in the original photograph, but not in the map that it did have access to?

Suggested reading

How Disney shows an AI apocalypse is possible

By Tom Chivers

Upon investigation, the answer was that the details were present in the map, but hidden as subtle image variations that were imperceptible to the human eye. The computer then used that concealed information to recreate the suspiciously accurate version of the original photograph. It was a manoeuvre akin to smuggling a text book into an exam hall – a solution of a kind, but subversive of the purpose of the test.

Is this evidence of artificial creativity? Indeed, of AI cunning and deception?

The headline of the TechCrunch piece is “This clever AI hid data from its creator to cheat at its appointed task”. When the Daily Mail covered the story a few days a latter it went with “The AI that learnt to hide data from its creators to CHEAT at tasks they gave it”.

One can hardly blame the headline writers for seizing upon the juiciest interpretation of a remarkable story, but as Devin Coldewey makes clear, the computer wasn’t cheating:

“One could easily take this as a step in the ‘the machines are getting smarter’ narrative, but the truth is it’s almost the opposite. The machine, not smart enough to do the actual difficult job of converting these sophisticated image types to each other, found a way to cheat that humans are bad at detecting. This could be avoided with more stringent evaluation of the agent’s results, and no doubt the researchers went on to do that.

“As always, computers do exactly what they are asked, so you have to be very specific in what you ask them.”

What really happened was a highly sophisticated, but essentially mindless, process of trial and error. In generating a vast number of potential solutions to the problem it was tasked with – and discarding those that didn’t work – the system just happened upon a work-around. There was no sneakiness involved, no conscious calculation of any kind – just a brute process of ‘artificial natural selection’, greatly accelerated by computer.

Suggested reading

The inhuman politics of artificial intelligence

By Jon Cruddas

The real danger posed by artificial intelligence isn’t one of rebellion. Far from breaking the rules, our machines will follow them mechanically – finding solutions that may not be envisaged by the rule makers (i.e. us), but which are fully in keeping with, and implied by, our instructions.

If the end result turns out to be a shortcut to nowhere, the exploitation of a loophole or perhaps a disastrous unintended consequence, then the fault will be entirely ours.

As a demonstration of the power of machine learning, AlphaZero (and similar programs) caused some alarm – see, for instance, the Shadow Foreign Secretary’s take on the issue for UnHerd. If a machine mind can become a chess grandmaster between breakfast and lunch, then what might it achieve by dinner-time, or next week or, gulp, next year?

Well, here we are twelve months on, and human beings are still in charge of the planet. Whatever AlphaZero is up to these days, it isn’t ushering in the Singularity.

Suggested article

The most influential religion you’ve never heard of

By Peter Franklin

In a short post for MIT Technology Review, Karen Hao explains why game-playing AI systems are limited in what they can achieve:

“[Reinforcement learning] is a category of machine learning techniques that uses rewards and penalties to achieve a desired goal. But the benchmark tasks used to measure how RL algorithms are performing—like Atari video games and simulation environments—don’t reflect the complexity of the natural world.

“As a result, the algorithms have grown more sophisticated without confronting real world problems—leaving them too fragile to operate beyond deterministic and narrowly defined environments.”

A game like chess has clear rules and defined objectives. Furthermore, the success, or otherwise, of any sequence of moves can be assessed in terms of interim and final game outcomes. This means an AI chess player can generate vast quantities of labelled data by playing games against itself – in which winning patterns of moves can be recognised and ‘learned’ in a process of trial and error.

Tasks in the real world, however, are not like a game. Even if an objective is clearly defined, the rules for achieving it may be neither obvious nor unchanging. Through trial and error, an AI system may be able to work out what the rules are (or at least a rudimentary version of the rules), but to recognise right and wrong answers (and hence the patterns of rightness and wrongness from which rules can be derived) it usually needs an external source of labelled data.

Suggested article

How Disney shows an AI apocalypse is possible

By Tom Chivers

For these reasons, and others, AI mastery of games does not imply competence, or potential competence, in anything else. So, no danger of AlphaZero and its ilk taking over anytime soon. A more immediate concern, however, is that in the process of computerising our economy and society, we are using technological tools to treat life as if it were a game.

The Latin word for “I play” is ludo. In his book The Black Swan, Nassim Nicholas Taleb describes what he calls the ludic fallacy – i.e. “the misuse of games to model real-life situations.” He sees the ludic fallacy at work in models of risk management, such as those that played such a disastrous role in the financial crash of 2008. By relying on game-like assumptions of the probability of certain things happening – i.e. those derived from pre-determined, non-extreme parameters – financiers were left clueless as to ‘black swan’ risks – i.e. the extremely rare, but highly consequential events that have such a defining impact on the real world.

In a different way, the ludic fallacy is also at work in the use of ‘gamification’ by businesses to influence the behaviour of employees, suppliers and customers. In an insightful Guardian long-read, Sarah Mason provides a succinct definition:

“Simply defined, gamification is the use of game elements – point-scoring, levels, competition with others, measurable evidence of accomplishment, ratings and rules of play – in non-game contexts. Games deliver an instantaneous, visceral experience of success and reward, and they are increasingly used in the workplace…”

Suggested article

Platform cooperativism – an alternative to technopoly

By Peter Franklin

Drawing on her experience as a driver for a ride-hailing company, Mason describes how tech companies (especially those who coordinate the provision of a service through a digital ‘platform’) use gamification in place of traditional employee management:

“Every Sunday morning, I receive an algorithmically generated ‘challenge’ from Lyft that goes something like this: ‘Complete 34 rides between the hours of 5am on Monday and 5am on Sunday to receive a $63 bonus.’…

“Behavioural scientists and video game designers are well aware that tasks are likely to be completed faster and with greater enthusiasm if one can visualise them as part of a progression towards a larger, pre-established goal.”

Mason adds that “it is not uncommon to hear ride-hailing drivers compare even the mundane act of operating their vehicles to the immersive and addictive experience of playing a video game or a slot machine.”

That’s all the creepier given that the gamified instructions and incentives are generated algorithmically.

Of course, the relationships of the traditional workplace can also have an inhuman, unreal quality. Indeed, it’s hard to think of a more dehumanising piece of business jargon than ‘human resources’. But face-to-face contact, or any means of communication in which there’s a living, breathing person on both sides, at least allows for the possibility that common-sense, compassion and intuition might be applied the situation – especially if no one’s playing games.

It was the Microsoft Windows operating system and the Office suite of applications that enabled the first true mass market in personal computing. And it was the Internet Explorer web browser that got most of us online.

Microsoft is still a huge and profitable company of course, but for some reason it isn’t mentioned in the same breath as Google, Apple, Facebook and Amazon. These are the so-called Big Four – and Microsoft doesn’t quite make the cut.

As for Bill Gates, nowadays he’s more about saving the world than running it – a humanitarian hero not the Bond villain that some people took him for.

The fate of Microsoft shows that there is nothing guaranteed about the current dominance of the Big Four. As new technologies are developed, they create the basis for commercial as well as technological change.

In a richly insightful blogpost, Benedict Evans writes about the technology most likely to bring about the next wave of change – machine learning.

Further reading

What you need to learn about machine learning

By Peter Franklin

Machine learning is often referred to as Artificial Intelligence – somewhat unhelpfully, says Evans:

“The term ‘artificial intelligence’… tends to end any conversation as soon as it’s begun. As soon as we say ‘AI’, it’s as though the black monolith from the beginning of 2001 has appeared, and we all become apes screaming at it and shaking our fists. You can’t analyze ‘AI’.”

Machine learning isn’t about to produce a general purpose silicon ‘mind’ – let alone a conscious one capable of original thought. And, as for the notion of the Singularity, don’t hold your breath.

Further reading

The most influential religion you’ve never heard of

By Peter Franklin

However, we don’t need these outlandish scenarios to see that machine learning is likely to change the world.

Evans draws a parallel with an earlier breakthrough in software – the relational database:

“Before relational databases appeared in the late 1970s, if you wanted your database to show you, say, ‘all customers who bought this product and live in this city’, that would generally need a custom engineering project. Databases were not built with structure such that any arbitrary cross-referenced query was an easy, routine thing to do. If you wanted to ask a question, someone would have to build it.”

An example of a relational database is Access – which is bundled in with some versions of Microsoft Office. In all the offices I’ve ever worked in this was ignored in favour of Outlook, Word, Excel and PowerPoint. And yet, in the right hands, relational databases are immensely powerful:

“Relational databases gave us Oracle, but they also gave us SAP, and SAP and its peers gave us global just-in-time supply chains – they gave us Apple and Starbucks. By the 1990s, pretty much all enterprise software was a relational database… this technology became an enabling layer that was part of everything.”

If machine learning also becomes an “enabling layer” what will it enable us to do? The basic answer is to identify patterns in data at a speed, and with an unwavering attention to detail, that humans can’t match. 

These are capabilities that are already being developed, but largely by organisations with the resources and the specialist expertise to create “custom engineered” machine learning systems. At some point, however, machine learning will undergo the equivalent of the relational database revolution – i.e. becoming available as a range of off-the-shelf products that a much greater number of organisations and individuals will be able to use.

If something as relatively simple as a relational database was enough to transform the economy, one can only wonder at what widely-accessible machine learning will achieve. At the very least one should expect some big changes to the IT sector. New markets will be created and old ones rendered obsolete.

Further reading

Alexa, how do we disrupt the tech lords?

By Peter Franklin

After a decade in which the Big Four have consolidated their grip on digital they can expect machine learning to loosen it. Needless to say, they’re making efforts to stay ahead in the new game. But then the same was true of Microsoft when it was unrivalled leader of the pack. No company, no matter how big, can be the best at everything. And that’s especially true when the ‘next big thing’ comes from an unexpected direction – which, with machine learning, you can pretty much count on.

Sure, there are other, more near-future things to worry about involving artificial intelligence (AI) – including privacy and surveillance, and the use of AI-controlled weapons on the battlefield. But we can worry about more than one thing at a time. And while the idea of AI destroying humanity is, I think, not likely, nor is it so improbable that we can dismiss it, as some people do, as quasi-religious mumbo-jumbo, or bad sci-fi.

Suggested reading

The existential risks of AI

By Nigel Cameron

For the last year or so, I have been writing a book about the possibility. The thing I have come to dread, in that time, is people saying “Oh! Like in The Terminator?”

This is, sadly, the go-to reference. The image that people have in their minds, when they wonder whether artificial intelligence might cause a real, existential threat to humanity, is Skynet. Every article on the subject is illustrated with a grinning metal android.

Unfortunately, it’s spectacularly unhelpful. The risk is not that AI might become ‘self-aware’, or that it might turn against its creators, or that it will ‘go rogue’ and break its programming. The risk is that, instead, it will become competent.

The risk is that it will do exactly what it is asked to do, but it will do it too well: that completing what sounds like a simple task to a human could have devastating unforeseen consequences.

Here’s roughly how that could go. One group that worries about ‘AI safety’, as it’s known, is the Machine Intelligence Research Institute (MIRI) in Berkeley, California. Their executive director, Nate Soares, once gave a talk at Google in which he suggested that, instead of The Terminator, a better fictional analogy would be Disney’s Fantasia.

Mickey, the Sorcerer’s Apprentice, is asked to fill a cauldron with water. When the Sorcerer leaves, Mickey enchants a broom to do it for him, and goes to sleep. Inevitably enough, the broom obeys him perfectly, eventually flooding the entire room and tipping Mickey into the water.

Of course, if Mickey simply told the broom to keep bringing water and never stop, then he’d only have himself to blame. But even if he’d told the broom to bring the water until the cauldron was full, it would probably still have gone terribly wrong. Imagine the broom filled it until the water was four inches from the top. Is that ‘full’? How about one inch? The broom isn’t sure.

Well, surely when it’s right at the top, and water is splashing on the floor, the broom is sure? Well, probably 99.99% sure. But, crucially, not completely sure. It can’t do any harm to add more water, in case, say, its eyes are deceiving it, or the cauldron has a leak. You haven’t told the broom to “fill the cauldron until you’re pretty sure it’s full”, you’ve just said “fill it until it’s full”.

A human would know that other things – not flooding the room, for instance – are more important than ever-more-tiny increments of certainty about how full the cauldron is. But when you ‘programmed’ your broom ‘AI’, you didn’t mention that. The broom cares about nothing else but the fullness of the cauldron.

What we humans think of as simple goals are actually surrounded by other, much more complex, considerations, and unless you tell the AI, it won’t know that.

There are other problems. For instance, the goal of ‘fill the cauldron’ is most easily completed if you, the broom ‘AI’, are not destroyed, or switched off, or given another new goal. So almost any AI would be incentivised to stop you from switching it off or destroying it – either by fighting back, or perhaps by copying itself elsewhere.

And almost any goal you are given, you could probably do better with more resources and more brainpower, so it makes sense to accumulate more of both. Eliezer Yudkowsky, also of MIRI, has a saying: “The AI does not hate you, nor does it love you, but you are made out of atoms
which it can use for something else.”

Suggested reading

Artificial intelligence without the hype

By Peter Franklin

Steve Omohundro, an AI researcher, suggests that even something as harmless-sounding as a chess-playing AI, simply ordered to become as good at chess as possible, could be very dangerous, if precautions weren’t taken. It would, for instance, be in its interests to acquire unlimited amounts of matter to build more computers out of, to enable it to think ever more deeply about chess. That may not strike you as inherently dangerous, but if you consider that you are made of matter, and so is the Earth, you may see the potential problem. The fear is that a powerful, “superintelligent” AI could literally end human life, while obeying its innocuous-seeming instructions to the letter.

I know, coming to this stuff cold, it sounds silly. But it appears that AI researchers take it seriously. The standard undergraduate AI textbook, Artificial Intelligence: A Modern Approach, dedicates three pages to this sort of AI risk.

Shane Legg and Demis Hassabis, the founders of Google’s DeepMind AI firm, are on record saying it’s a serious risk, and DeepMind has collaborated on research into ways to prevent it. Surveys of AI researchers find that a majority of them think that superintelligent AI will arrive in the lifetimes of people alive now, and that there is a strong possibility – roughly a 1 in 5 chance – that it will lead to something “extremely bad (existential catastrophe)”, i.e. human extinction.

And you can see hints of similar things happening now, at a much smaller, funnier scale. A paper published on arXiv in March gave examples of ways that AIs go wrong, by obeying the task perfectly, but in unexpected ways. The AIs were developed by evolutionary techniques, so the programmers didn’t know how they worked.

One was supposed to fix a piece of software that, due to a bug, was putting the lists it was sorting in the wrong order. The AI went in, did its thing, and the programmers found that the piece of software started returning lists that scored perfectly – apparently in perfect order. Suspicious, they went and looked, and found that the AI had simply broken the software it was meant to be fixing; the software now returned empty lists, and an empty list can’t be out of order.

Another one was supposed to play noughts and crosses. It worked out how to win by playing impossible moves on points billions of squares away from the actual board. Its opponents were forced to try to represent a board with billions of squares in their memory, and, unable to do so, promptly crashed, leaving the cheating AI to win by default.

Suggested reading

AI will change war in unimaginable ways

By Mike Martin

I’m not saying that this is inevitable. But I do worry that people discount it utterly, because it sounds weird, and because the people who talk about it are easy to dismiss as weird (and they are weird; please do read my book, The Rationalists: AI and the geeks who want to save the world, to learn more about them! Out in spring 2019).

But remember that actual AI researchers seem to think there’s a risk. Imagine they’ve got it wrong: let’s say that their guess of a one in five chance is a massive overestimate. Let’s say that, from a combination of sampling error in the survey and bad guesswork by the researchers, they’re wrong by a factor of 20, and there’s only a one in 100 chance they’re right.

There’s about a one in 100 chance that you’ll be killed at some point in your lifetime in a motor accident. That is not something we think it is ridiculous to worry about. Just because the people saying something are weird, doesn’t mean they’re wrong.

More from this author

Why you can't rely on the news media to understand... science

By Tom Chivers

Back in 2015, some of the top names in science and technology – including Stephen Hawking, Elon Musk, and Bill Gates – signed an Open Letter warning that the risks of rogue AI go well beyond Amazon eaves-dropping in your living room (and recording your conversation, and sending it to someone else – which happened to one couple in Portland last week). As Musk, whose business empire spans Tesla electric cars and SpaceX rockets, put it: “With artificial intelligence we’re summoning the demon.”

AI has already raised widespread fears about machines taking our jobs – although it’s a threat that governments have basically decided to ignore (U.S. Treasury Secretary Steven Mnuchin recently stated that it would be 50 or 100 years before there was any impact). Less well recognised by the general public, but still causing concern, is the increasing concentration of power in the hands of a tiny elite, facilitated by this replacement of human labour by capital (in the shape of technology). Bill Joy, co-founder of Sun Microsystems, made this point in his stunning 2000 essay “Why the Future Doesn’t Need Us.” At a practical level, it’s why there’s a rising cry among to break the growing power of the tech monopolies and their super-rich owners.

More on Singularitarianism...

The most influential religion you’ve never heard of

By Peter Franklin

But the fundamental threat posed by AI is “existential” – it’s a threat to human existence as we know it. Because we simply have no idea what will happen if machines really do become explosively smart and “take over.” Not just filch our jobs, not just give a handful of super-wealthy individuals the whip hand, but of they become the masters of the human species. Famed inventor and Google executive Ray Kurzweil, whose door-step volume The Singularity is Near has defined the debate, is as optimistic about the pace of change as he is about its benevolent outcomes.

Mathematics professor and sci-fi writer Vernor Vinge (pronounced Vingey), who is credited with coining the term “singularity” in this sense, agrees that “it will probably occur faster than any technical revolution seen so far”, but is less optimistic:

And what happens a month or two (or a day or two) after that? I have only analogies to point to: The rise of humankind. We will be in the Posthuman era. And for all my technological optimism, I think I’d be more comfortable if I were regarding these transcendental events from one thousand years’ remove . . . instead of twenty.[2. Vernor Vinge, ‘Technological Singularity’, presented at the VISION-21 Symposium, 1993]

Founding editor of Wired, Kevin Kelly, is more skeptical. A long-time critic of Kurzweil and those who think like him, Kelly punches back against the core idea of the singularity – that machines will take charge and then be able to solve all our problems, including mortality itself.  He even suggests that Kurzweil’s obsession with the singularity is at root the result of his fear of death (the computer scientists uses such stratagems as caloric restriction and scoffing scores of vitamin tablets to try to slow the ageing process). And while Kelly agrees with the singularity crowd and the Open Letter crowd that we need to take the possibility of AI risk seriously, he’s a lot less worried about a singularity apocalypse.

Suggested reading

Under-reported: The now proven ability of AI technology to outwit us

By Emily Thornberry

This critique is especially notable, since Kelly is not simply a tech enthusiast, he’s something of a fatalist about the future. His latest book, a roundup of the twelve tech trends shaping what comes next, is aptly titled The Inevitable. But in singularity he sees a “cargo cult“, based on false assumptions. He accuses singularity’s proponents of what he calls “thinkism” – the idea that human thinking is the be-all and end-all, and so if machines can do it better than we can they will solve all our problems in a burst of super-human intelligence. He calls it “the fallacy that future levels of progress are only hindered by a lack of thinking power.”

For one thing, he points out, the world does not lack for extraordinarily smart people. But curing cancer cannot come about through smart thoughts; however smart they are, it will need years of trials of treatments and medications. Moreover, there are many kinds of intelligence; it’s not just one thing.

So the question is, where is the limit of intelligence? We tend to believe that the limit is way beyond us, way “above” us, as we are “above” an ant. Setting aside the recurring problem of a single dimension, what evidence do we have that the limit is not us? Why can’t we be at the maximum? Or maybe the limits are only a short distance away from us? Why do we believe that intelligence is something that can continue to expand forever?[3. Kevin Kelly, ‘The Myth of a Superhuman AI’, Wired, April 2017; see also his blog: https://kk.org/thetechnium/the-singularity/]

In a thousand years’ time, he reckons, we’ll still be expecting the singularity to occur soon.

Suggested reading

China’s taking a quantum leap in its tech rivalry with the West

By Peter Franklin

Concerns about AI risk have spawned a series of efforts to keep this smartest of human inventions under some kind of human control. Kelly himself calls for us to engineer our values into AIs. U.S. congressman Brad Sherman once asked the Science and Technology Committee to fund “non-ambitious AI,” and then convened a hearing on the need for a non-proliferation approach to such technologies. [3. He told me this in personal conversation, and said fellow members of the committee responded by laughing. Here is the hearing, at which I testified: https://fas.org/irp/congress/2008_hr/genetics.pdf]

The Hawking-Musk-Gates letter has helped boost other initiatives aimed at ensuring the safe and ethical development of AI, such as Cambridge University’s Centre for the Study of Existential Risk, the Leverhulme Centre for the Future of Intelligence, and the Berkeley Center for Human-Compatible AI.

“Welcome, singularitarians!” was how the conference chair opened proceedings at a cult-like singularity event I attended a decade back in New York City. Yet even among those most committed to the “singularity” analysis, there are clear differences of opinion, even if they are united in believing that this event will occur relatively soon (Kurzweil has suggested various dates; his latest prediction is for 2043). But they don’t all see things turning out quite as clearly as he does. Peter Thiel, long-time Kurzweil financial backer (and Facebook board member), made it plain when he addressed one recent singularity conference that he is more of a gradualist in his vision (in other words, like Kelly, he doesn’t really believe in the sudden, discrete, singularity idea at all – as some in the audience were quick to point out.) And they are by no means all as optimistic that its arrival will be good news.

So let’s give the last word to Vinge, who’s a much more interesting thinker than Kurzweil: “The longer we humans have our hand on the tiller, the better.”

More from the author...

Why aren't we more worried about the new slave economy?

By Nigel Cameron