Essay

AI - Artificial Intelligence

Comment

30 October 2023

11 min read

The summit of humanity: decoding AI's affectations

An AI summit’s prophecies need to be placed in the right philosophical register, argues Simon Cross. Because being human in an AI age still means the same thing it has for millennia.

Simon Cross

Simon Cross researches ethical aspects of technology and advises on the Church’s of England's policy and legislative activity in these areas.

An AI generated image of robot skulls with bulging eyes on a shelf receding diagonally to the left.

Alessio Ferretti on Unsplash.

The UK’s global artificial intelligence (AI) conference is nearly upon us. If the UK had a ‘prophecy office’ it would have issued a yellow or even amber warning for the first days of November by now. Prophecy used to be a dangerous business, the ancient text of Deuteronomy sanctioned death for false prophets, equating its force with a leading away from God as the ultimate ground of truth. But risks duly acknowledged, here is a prophecy about the prophecies to come. The global AI conference will loudly proclaim three core prophecies about AI.

This time it’s different. Yes, we said that before but this time it really is different.
Yes, we need global regulation but, you know, it’s complicated so only the kind of regulation we advise is going to work.
Look, if we don’t do this someone else will. So, you should get out of our way as much as you possibly can. We are the good guys and if you slow us down the bad guys will win.

I feel confident about this prediction not because I wish to claim the office of prophet but because just like Big Tobacco and Big Oil, Big Tech’s lobbyists will redeploy a tried and tested playbook. And here are the three plays at the heart of it.

Tech exceptionalism. (We deserve to be treated differently under the law.)

Regulatory capture. (We got lucky, last time, with the distinction between platform and publisher that permitted self-regulation of social media, the harvesting of personal data and manipulative design for attention, but the costs of defeating Uber in California and now defending rearguard anti-trust lawsuits means lesson learned, we need to go straight for regulatory capture this time).

Tech determinism. (If we don’t do it, someone else will. We are the Oppenheimers here.)

Speaking of Pandora

What should we make of these claims? We need to start by exploring an underlying premise. One that typically goes like this “AI is calling into question what it means to be human”.

This premise has become common currency, but it is flawed because it is too totalising. AI emphatically is calling into question a culturally dominant version of human anthropology – one specific ‘science of humanity’. But not all anthropologies. Not the Christian anthropology.

A further, unspoken, premise driving this claim becomes clearer when we survey the range of responses to the question “what does the advent of what the government is now calling ‘frontier’ AI portend?”

Either, it means we have finally prized open Pandora’s box; the last thing humans will ever create. AI is our Darwinian evolutionary heir, soon to make us homo sapiens redundant, extinct, even. Which could happen in two very different ways. For some, AI is the vehicle to a new post-human eternal life of ease, roaming the farthest reaches of the universe in disembodied digital repose. To others, AI is now on the very cusp of becoming abruptly and infinitely cleverer than us. To yet others, we are too stupid to avoid blowing ourselves up on the way to inventing so-called artificial general intelligence.

Cue main global summit speaking points…

Or,

AI is just a branch of computing.

Which of these two starkly contrasting options you choose will depend on your underlying beliefs about ‘what it means to be human’.

Universal machines and meat machines

Then again, what does it mean to be artificially intelligent? Standard histories of AI always point to two seminal events. First, Alan Turing published a paper in the 1930s in which he proposed a device called a Universal Turing Machine.

Turing’s genius was to see a way of writing a type of programme to control a computer’s underlying binary on/off in ways that could vary depending on the task required and yet perform any task a computer can do. The reason your computer is not just a calculator but an excel spreadsheet and a word processor and a video player as well is because it is a kind of Universal Turing Machine. A UTM can compute anything that can be computed. If it has the right programme.

The second major event in AI folklore was a conference at Dartmouth College in the USA in the early 1950s bringing together the so-called ‘godfathers of AI’.

This conference set the philosophical and practical approaches from which AI has developed ever since. That this happened in America is important because of the strong link between universities, government, the defence and intelligence industry and the Big Tech Unicorns that have emerged from Silicon Valley to conquer the world. That link is anthropological; it is political, social, and economic and not just technical.

Let’s take this underlying question of ‘what does it mean to be human?’ and recast it in a binary form as befits a computational approach; ‘Is a human being a machine or is a human being an organism?’

Cognitive scientist Daniel Dennett was recently interviewed in the New York Times. For Dennett our minds and bodies are a “consortia of tiny robots”. Dennett is an evolutionary biologist and a powerful voice for a particular form of atheism and its answer to the question ‘what does it mean to be human?’ Dennett regards consciousness as ephemera, a by-product of brain activity. Another godfather of AI, Marvin Minsky, famously described human beings as ‘meat machines.’

By contrast, Joseph Weizenbaum was also one of the early computer pioneers in the 1960s and 1970s. Weizenbaum created one of the first ever chatbots, ELIZA– and was utterly horrified at the results. His test subjects could not stop treating ELIZA as a real person. At one point his own secretary sat down at the terminal to speak to ELIZA and then turned to him and asked him to leave the room so she could have some privacy. Weizenbaum spent the latter part of his professional life arguing passionately that there are things we ought not to get computers to do even if they can, in principle, perform them in a humanlike manner. To Joseph Weizenbaum computers were/are fundamentally different to human beings in ways that matter ineluctably, anthropologically. And it certainly seems as if the full dimensionality of human being cannot yet be reduced to binary on/off internal states without jettisoning free will, consciousness and transcendence. Prominent voices like Dennett and Yuval Noah Harari are willing to take this intellectual step. Their computer says ‘no’. By their own logic it could not say otherwise. In which case here’s a third way of asking that seemingly urgent and pressing question about human being;

“Are we just warm, wet, computers?”

The immanent frame

A way to make sense of this, for many people, influential and intuitively attractive meaning of human being is to understand how the notion of artificial intelligence fits a particular worldview that has come to dominate recent decades and, indeed, centuries.

In 2007 Charles Taylor wrote A Secular Age. In it he tracks the changing view of what it means to be human as the Western Enlightenment unfolds. Taylor detects a series of what he calls ‘subtraction stories’ that gradually explain away the central human experience of transcendence until society is left with what he calls an ‘immanent frame’. Now we are individual ‘buffered selves’ insulated by rational mind so that belief in any transcendent reality, let alone God, is just one possible choice among personal belief systems. But, says Taylor, this fracturing of a shared overarching answer to the question ‘What does it mean to be human’ over the past, say, 500 years doesn’t actually answer the question or resolve the ambiguities. Rather, society is now subject to what Taylor calls ‘cross pressures’ and a lack of societal consensus about the answers to the biggest questions of human meaning and purpose.

In this much broader context, it becomes easier to see why as well as how it can be the case that AI is either a profound anthropological threat or just a branch of computing – depending on who you talk to…

The way we describe AI profoundly influences our understanding of it. When Dennett talks about a ‘consortia of tiny robots’ is he speaking univocally or metaphorically? What about when we say that AI “creates”, or “decides” or “discovers” or ‘seeks to maximise its own reward function’. How are we using those words? If we mean words like ‘consortia’ or ‘choose’ and ‘reward’ in as close to the human sense as makes no difference, then of course the difference between us and our machines becomes paper-thin. But are human beings really a kind of UTM? Are UTMs really universal? Are you a warm wet computational meat-machine?

Or is AI just the latest and greatest subtraction story?

To say AI is just a branch of computing is not to say the harms of outsourcing key features of human being to machines are trivial. Quite the opposite.

How then should we judge prophecies about AI emanating from this global conference or in the weeks and months to follow? I suggest two responses. The first follows from my view of AI, the other from my view of human being.

Our view of current AI should be clear eyed, albeit open to revision should future development(s) so dictate. I am firmly on the side of those who, without foreclosing the possibility, see no philosophical breakthrough in the current crop of tools and techniques. These are murky philosophical waters but clocks don’t really have human hands now do they, and a collapsed metaphor can’t validate itself however endemic the reference to the computational theory of mind has become.

Google’s large language model, Bard, for example, has no sense of what time it is where ‘he’ is, let alone can freely choose to love you or not, or to forgive you if you hurl an insult at ‘him’. But all kinds of anthropological harms already flow from the unconscious consequences of re-tuning human being according to the methodological image of our machines. To say AI is just a branch of computing is not to say the harms of outsourcing key features of human being to machines are trivial. Quite the opposite.

Which brings me to the second response. When you hear the now stock claim that AI is calling into question what it means to be human, don’t buy it. Push back. Point out the totalising lack of nuance. The latest tools and techniques of AI are calling a culturally regnant but philosophically reductive anthropology into question. That much is definitely true. But that is all.

And it is important to resist this totalising claim because if we don’t, an increasingly common and urgent debate about the fullness of human being and the limitations of UTMs will struggle from the start. One of the biggest mistakes I think public theology made twenty-some years ago was to cede a normative use of language that distinguished between people of faith and people of no faith. There is no such thing as being human without faith commitments of one kind or another. If you have any doubt about this, I commend No One Sees God: The Dark Night of Atheists and Believers by Michael Novak. But the problem with accepting the false distinction between ‘having faith’ and having ‘no faith’ is that it has allowed the Dennetts and Hararis of this world to insist that atheism is on a stronger philosophical footing than theism. After which all subsequent debate had, first, to establish the legitimacy of faith per se before getting to the particular truth claims in, say, Christianity.

What it means to be human

I see a potentially similar misstep for anthropology – the science of human being – in this new and contemporary context of AI. Everywhere at the moment, and I mean but everywhere, a totalising claim is being declared ever more loudly and urgently: that the tools and techniques of AI are calling into question the very essence of human identity. The risk in ceding this claim is that we get stuck in an arid debate about content instead of significance; a debate about ‘what it means to be human’ instead of a debate about ‘what it means to be human.’

This global AI summit’s proclamations and prophecies need to be placed in the right philosophical register, because to be human in an age of AI still means the same thing it has for millennia.

Universals like wonder, love, justice, the need for mutually meaningful relationships and a sense of purpose, and so too personal idiosyncrasies like a soft spot for the moose are central features of what it means to be this human being.

Suchlike are the essential ingredients of the ‘me’ that is reading this article. They are not tertiary. Perhaps they can be computationally mimicked but that does not mean they are, in themselves, ephemeral or mere artifice. In which case their superficial mimicry carries substantial risks, just as Joseph Weizenbaum prophesied in Computer Power and Human Reason in the 1970s.

Of course, you may disagree. You may even disagree in good faith, for there are no knockdown arguments in metaphysics. And in my worldview, you are free to do so. But fair warning. If the human-determinism of Dennett or the latest prophecies of Harari are right, no credit follows. You, and they, are right only because by arbitrary alignment of the metaphysical stars, you, and they, have never been free to be wrong. It was all decided long ago. No need for prophecies. We are all just UTMs with the soul of a marionette

But when you hear the three Global summit prophecies I predicted earlier, consider these three alternatives;

This time is not different, it is not true that AI is calling into question all anthropologies. AI is (only) calling into question a false and reductive Enlightenment prophecy about ‘what it means to be human.’

The perennial systematic and doctrinal anthropology of Christianity understands human being as free-willed, conscious, unified body soul and spirit. It offers credible answers to the urgent questions and cross-pressures society is now wrestling with. It also offers an ethical framework for answering the question ‘what ought computers to be used for and what ought computers not to be used for – even if they appear able to be used for anything and everything?

This Christian philosophical perspective on the twin underlying metaphysical questions of human being and purpose are not being called into question, either at this global summit or by any developments in AI today or the foreseeable future. They can, however, increasingly be called into service to answer those questions – at least for those with ears to hear.

Article

AI - Artificial Intelligence

Culture

26 March 2024

10 min read

We’ll learn to live with AI: here’s how

AI might just help us with life’s dilemmas, if we are responsible.

Andrew Briggs

Andrew is Emeritus Professor of Nanomaterials at the University of Oxford.

Two construction workers stand and talk with a humanoid AI colleague.

Nick Jones/Midjourney.ai

Anxiety about algorithms is nothing new. Back in 2020, It was a bad summer for the public image of algorithms. ‘I am afraid your grades were almost derailed by a mutant algorithm’, the then Prime Minister told pupils at a school. No topic in higher education is more sensitive than who gets a place at which university, and the thought that unfair decisions might be based on an errant algorithm caused understandable consternation. That algorithms have been used for many decades with widespread acceptance for coping with examination issues ranging from individual ill health to study of the wrong set text by a whole school seems quietly to have slipped under the radar.

Algorithmic decision-making is not new. Go back thousands of years to Hebrew Deuteronomic law: if a man had sex with a woman who was engaged to be married to another man, then this was unconditionally a capital offence for the man. But for the woman it depended on the circumstances. If it occurred in a city, then she would be regarded as culpable, on the grounds that she should have screamed for help. But if it occurred in the open country, then she was presumed innocent, since however loudly she might have cried out there would have been no one to hear her. This is a kind of algorithmic justice: IF in city THEN woman guilty ELSE woman not guilty.

Artificial intelligence is undergoing a transition from classification to decision-making. Broad artificial intelligence, or artificial general intelligence (AGI), in which the machines set their own goals, is the subject of gripping movies and philosophical analysis. Experts disagree about whether or when AGI will be achieved. Narrow artificial intelligence (AI) is with us now, in the form of machine learning. Where previously computers were programmed to perform a task, now they are programmed to learn to perform a task.

We use machine learning in my laboratory in Oxford. We undertake research on solid state devices for quantum technologies such as quantum computing. We cool a device to 1/50 of a degree above absolute zero, which is colder than anywhere in the universe that we know of outside a laboratory, and put one electron into each region, which may be only 1/1000 the diameter of a hair on your head. We then have to tune up the very delicate quantum states. Even for an experienced researcher this can take several hours. Our ‘machine’ has learned how to tune our quantum devices in less than 10 minutes.

Students in the laboratory are now very reluctant to tune devices by hand. It is as if all your life you have been washing your shirts in the bathtub with a bar of soap. It may be tedious, but it is the only way to get your shirts clean, and you do it as cheerfully as you can … until one day you acquire a washing machine, so that all you have to do is put in the shirts and some detergent, shut the door and press the switch. You come back two hours later, and your shirts are clean. You never want to go back to washing them in the bathtub with a bar of soap. And no one wants to go back to doing experiments without the machine. In my laboratory the machine decides what the next measurement will be.

Suppose that a machine came to know my preferences better than I can articulate them myself. The best professionals can already do this in their areas of expertise, and good friends sometimes seem to know us better than we know ourselves.

Many tasks previously reserved for humans are now done by machine learning. Passport control at international airports uses machine learning for passport recognition. An experienced immigration officer who examines one passport per minute might have seen four million faces by the end of their career. The machines were trained on fifty million faces before they were put into service. No wonder they do well.

Extraordinary benefits are being seen in health care. There is now a growing number of diagnostic studies in which the machines outperform humans, for example, in screening ultrasound scans or radiographs. Which would you rather be diagnosed by? An established human radiologist, or a machine with demonstrated superior performance? To put it another way, would you want to be diagnosed by a machine that knew less than your doctor? Answer: ‘No!’ Well then, would you want to be diagnosed by a doctor who knew less than the machine? That’s more difficult. Perhaps the question needs to be changed. Would you prefer to be treated by a doctor without machine learning or by a doctor making wise use of machine learning?

If we want humans to be involved in decisions involving our health, how much more in decisions involving our liberty. But are humans completely reliable and consistent? A peer-reviewed study suggested that the probability of a favourable parole decision depended on whether the judges had had their lunch. The very fact that appeals are sometimes successful provides empirical evidence that law, like any other human endeavour, involves uncertainty and fallibility. When it became apparent that in the UK there was inconsistency in sentencing for similar offences, in what the press called a postcode lottery, the Sentencing Council for England and Wales was established to promote greater transparency and consistency in sentencing. The code sets out factors which judges must consider in passing sentence, and ranges of tariffs for different kinds of crimes. If you like, it is another step in algorithmic sentencing. Would you want a machine that is less consistent than a judge to pass sentence? See the sequence of questions above about a doctor.

We may consider that judicial sentencing has a special case for human involvement because it involves restricting an individual’s freedom. What about democracy? How should citizens decide how to vote when given the opportunity? Voter A may prioritise public services, and she may seek to identify the party (if the choices are between well identified parties) which will best promote education, health, law and order, and other services which she values. She may also have a concern for the poor and favour redistributive taxation. Voter B may have different priorities and seek simply to vote for the party which in his judgement will leave him best off. Other factors may come into play, such as the perceived trustworthiness of an individual candidate, or their ability to evoke empathy from fellow citizens.

This kind of dilemma is something machines can help with, because they are good at multi-objective optimisation. A semiconductor industry might want chips that are as small as possible, and as fast as possible, and consume as little power as possible, and are as reliable as possible, and as cheap to manufacture as possible, but these requirements are in tension with one another. Techniques are becoming available to enable machines to make optimal decisions in such situations, and they may be better at them than humans. Suppose that a machine came to know my preferences better than I can articulate them myself. The best professionals can already do this in their areas of expertise, and good friends sometimes seem to know us better than we know ourselves. Suppose also that the machine was better than me at analysing which candidate if elected would be more likely to deliver the optimal combination of my preferences. Might there be something to be said for benefitting from that guidance?

If we get it right, the technologies of the machine learning age will provide new opportunities for Homo fidelis to promote human flourishing at its best.

By this point you may be sucking air through your intellectual teeth. You may be increasingly alarmed about machines taking decisions that should be reserved for humans. What are the sources of such unease? One may be that, at least in deep neural networks, the decisions that machines make may be only as good as the data on which they have been trained. If a machine has learned from data in which black people have an above average rate of recidivism, then black people may be disadvantaged in parole decisions taken by the machine. But this is not an area in which humans are perfect; that is why we have hidden bias training. In the era of Black Lives Matter we scarcely need reminding that humans are not immune to prejudice.

Another source of unease may be the use to which machine learning is put for commercial and political ends. If you think that machine learning is not already being applied to you, you are probably mistaken. Almost every time you do an online search or use social media, the big data companies are harvesting your data exhaust for their own ends. Even if your phone calls and emails are secure, they still generate metadata. European legislation is better than most, and the Online Safety Act 2023 will make the use of Internet services safer for individuals in the United Kingdom. But there is a limit to what regulation can protect, and 2024 is likely to see machine learning powerfully deployed to sway voters in elections in half the world. Targeted persuasion predates AI, as Othello’s Iago knew, but machine learning has brought it to an unprecedented level of industrialisation, with some of the best minds in the world paid some of the highest salaries in the world to maximise the user’s screen time and the personalisation of commercial and political influence.

Need it be so? In some ways advances in machine learning are acting as the canary in the mine, alerting us to fundamental questions about what humans are for, and what it means to be human. The old model of Homo economicus—rational, selfish, greedy, lazy man—has passed its sell-by date. It is being replaced by what I like to call Homo fidelis—ethical, caring, generous, energetic woman and man. For as long as AGI remains science fiction, it is up to humans to determine what values the machines are to implement. If we get it right, the technologies of the machine learning age will provide new opportunities for Homo fidelis to promote human flourishing at its best.

Whatever the future capabilities of machines, they cannot be morally load-bearing because humans are self-aware and mortal, whereas machines are not.

Paul Collier and John Kay

Christians have been thinking about what it means to be human for two millennia, building on what came before, and so they ought to have something to contribute to how humans flourish. In It Keeps Me Seeking, my co-authors and I ask our readers to imagine that they were writing about three thousand years ago for people who knew nothing of modern genetics or psychological science about what it means to be human. ‘You are writing for a storytelling culture, and so you would probably put it in the form of a story. Let’s say you set it in a garden. The garden is pleasant, but it is also designed for character formation, and so there is work to do, and also the possibility for a hard moral choice. You want to convey that humans need social interactions (for the same reason that solitary confinement is a severe punishment), and so you try the literary thought experiment of having one solitary man and letting him encounter animals and name them. Animals can be useful and they can be good company. But ultimately no animals, not even a dog, are fully satisfactory as partners in work and companions in life. Humans need humans. An enriching component of human relationships is sex. So, the supreme gift to the solitary man in our story is companionship with an equal who is both like and unlike; a woman. It is hardly a complete account, but it is a good start. Oh, and there is one other aspect. They should be free of the shame which lies at the root of so much psychological disorder.’

As far as it goes, would you regard such an account as complete? If not, what would you add next? You can see where this is going. To be human you need to be responsible. So, you let the humans face the moral choice. You can even include an element of disinformation to make the choice harder. And then when it goes horribly wrong you let them discover that they are responsible for their actions, and that blaming one another does not help. If you have God in your story, then (uniquely for the humans) responsibility consists of accountability to God. This is how human distinctiveness was addressed in early Jewish thought. As an early articulation that to be human means to be responsible, the story of Adam and Eve is unsurpassed.

In Greed is Dead, Paul Collier and John Kay reference Citizenship in a Networked Age as brilliantly elucidating the issue of morally pertinent decision-taking. They write, ‘Whatever the future capabilities of machines, they cannot be morally load-bearing because humans are self-aware and mortal, whereas machines are not. Machines can be used not only to complement and enhance human decision-making, but for bad: search optimisation has already morphed into influence-optimisation. We must keep morally pertinent decision-taking firmly in the domain of humanity.’

The nature of humanity includes responsibility—for wise use of machine learning and much more besides. Accountability is part of life for people with widely differing philosophical, ethical, and religious world views. If we are willing to concede that accountability follows responsibility, then we should next ask, ‘Accountable to whom?’