By Marcus Weldon
Newsweek Contributing Editor for AI and President Emeritus of Bell Labs
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
A conversation with Sir Stephen Fry is a whirlwind of eclectic and esoteric references across a staggering diversity of knowledge that is stochastically connected in his polymathic mind to produce answers that inform, entertain and seduce the audience in equal measure. In many ways, he is the human equivalent of a large language model (LLM), with its preternatural ability to generate insightful answers from an astounding compendium of harvested information. So, he is arguably uniquely equipped to compare and contrast human intelligence with its artificial “cousin.”
It is clear from the outset that he finds the capabilities of these models impressive in the way that any artisan of great ability respects those of a rival, and he sees great potential for the elevation of humankind by these tools due to their easy accessibility and seemingly infinite capacity, not only in the scope and scale of their resources but also in their patience and ability to adapt to the needs of each human user. But he is also clear-eyed in his assessment of the potential harms caused by the combination of intrinsic flaws and the extrinsic deceptive, duplicitous or devious motivations of companies, countries and criminals, who serve nefarious commercial or political masters. He argues that these forces have aided and abetted the creation of “counterfeit people” in the form of chatbots with incredibly human interfaces yet no real understanding of the human condition or the attendant diversity and complexity of our value systems.
Fry is undoubtedly one of the leading intellects and outstanding wits of the last fifty years. He attributes his endless cognitive quest to what he calls his “GI”—his general intelligence—and a curiosity that stems from a basic “greed” for knowledge. He muses that this may be the product of his childhood environment: “I could argue that it was because I grew up in the countryside, the remote countryside by British standards, to quote Sidney Smith, the great, early 19th century divine and wit: ‘I can best describe where I am as saying that I am simply miles from the nearest lemon’; all I had was books, and so I entered the world of books.” This book collection was also unusually diverse due to the different backgrounds and interests of his parents: “my parents were two people of different bents, if you like. My father: a physicist inventor, he was called the ‘Mad Inventor’ locally. He converted the stable block of the house I grew up in into a laboratory. And my mother read history and was passionate about literature. So, they had a terrific collection of books that I worked my way through. And I didn't distinguish between the scientific books and the books of literature…I was lucky enough to grow up to think that everything in human life, and indeed the physical world, was grist to my mill.”
Read More on Newsweek.AI
Remarkably, for those who know Fry as a quintessentially British icon and “national treasure,” he reveals that he “could almost have been American” as his father was offered a job at Princeton University when Fry’s mother was pregnant with him, but they decided that Britain was where they wanted to raise their children. So, “Steve Fry” never manifested. But he sees this little vignette as instructive, as it reminds him of the contextual nature of our existence and, by extrapolation, the pressing need for AI systems to interact with us as individuals, and with sufficient contextual awareness so as to be of genuine assistance, rather than reducing us to a statistical mean.
...
Exposure to a diversity of knowledge as a rural British child is only part of the Fry equation, and his innate intelligence clearly provides the capacity to process myriad sources of information and make seemingly effortless ontological connections. Indeed, in prior conversations he has clarified that these connections are made in real time, and the answers dynamically synthesized in response to the question—analogous to the instantaneous way in which an LLM responds to a user prompt.
Intelligent Thoughts on Not-So-Intelligent Models
However, there are various attributes to intelligence that Fry would not confer on these AI models. By referencing the Latin roots, interlegere, meaning to “select between” and intellego, which translates to “reading into or between,” Fry argues that intelligence “is the ability to read things into things. And it doesn't matter whether it's music, reading something into the music you're listening to, or whether it is a complex idea, or whether it is social.” In other words, intelligence is the multidimensional or multidomain interpretation of a reality. Fry agrees that humans have evolved a unique ability to create these rich representations of our experienced reality, based on our neocortical substrate and sensorimotor learning algorithm, as described by Jeff Hawkins in his book, A Thousand Brains, and that current LLMs cannot represent this richness as they lack the “perceptual bricks.”
Importantly, Fry also sees a connection between intelligence and wisdom, and cites an interview he saw as a child with Michael Ramsey, then archbishop of Canterbury and widely regarded as a very wise man. When asked about the origin of wisdom, Ramsey answered, “I think wisdom is the ability to cope.” This immediately struck the young Fry, “I thought, well, that's a surprising answer from an archbishop. [But] actually, that's about right? You know, whatever life throws at you, coping without complaining, managing, just getting on, forbearing, doing this is something machines don't have to bother to think about. They don't have to cope. They don't get stressed.” This is clearly consistent with Darwinian principles, and Darwin himself was well aware that life evolves to cope with new stresses and there is evidence to suggest that enhanced coping strategies lead to the development of the three essential forms of wisdom: cognitive, reflective and affective wisdom. In combination, it is thought that these three facets allow for the “positive reinterpretation” of events through rational analysis and empathy (to consider events from multiple points of view), allowing efficient learning from unexpected and novel scenarios. It is clear that current LLMs incorporate multiple perspectives—indeed, an unprecedented number of points of view are considered in computing an answer—but the physiological, physical, psychological and emotional understanding essential to human wisdom is not well-captured due, in large part, to the absence of the necessary models to describe these affects.
The last aspect of intelligence that Fry highlights is the critical pruning or elimination function, noting intelligence “is as much about inhibition as action,” in that the elimination of distractions and irrelevant information forms the key basis for insight and is one of the defining attributes of what we consider “expertise.” This selection process is somewhat akin to the gradient descent algorithm and the adaptation of weights in LLM training, or path selection during inference, but the LLM process is necessarily less rich than that employed by human brains, as the continuous learning algorithms we employ to allow cumulative perspective refinement without catastrophic forgetting are still not well understood. Fry also points out that a critical missing piece in current LLMs is the lack of a set of defining principles that describe the world as it should be. As he eloquently puts it, “specificity without principle is rather useless,” as it doesn’t allow reliable generalization, one of the key attributes of any intelligence.
In short, although Fry’s view of intelligence is aligned with the summary view expressed in this series, it is also enhanced by consideration of not just these functional attributes, but also the demonstration of three critical capabilities: the ability to read into things; the ability to cope with stress; and the ability to suppress irrelevant information and inhibit undesirable actions, across the entire panoply of human experience.
Creative Context and Connections
The conversation naturally turned to the role of AI in the creative arts and the role that LLMs can constructively play in creative expression, as well as the appropriate protections that should be in place in order not to “harm the market,” a critical criterion in intellectual property law.
Fry sees art as an essential lens through which we perceive the world. He references the German philosopher Arthur Schopenhauer and his observation that “Talent reaches things that we cannot reach. Genius reaches things that we cannot see,” to which Fry adds, “Art teaches us to see things that we hadn't seen.” He explains this by reference to another great thinker with whom Fry has often been associated, Oscar Wilde, that “nature imitates art,” meaning that we interpret our experiences through the lens or filter of the art we have been exposed to. He recalls Wilde viewing a painting by the English Romantic painter J.M.W. Turner, and declaring, “I cannot now look at a sunset without seeing how Turner has changed it for me. I never saw it like that before, but he made me see it in a new way, and that's how sunsets now look and will for humanity ever since Turner did this.”
The written word undoubtedly has equivalent power. Fry observes, “we changed because of Shakespeare, if you like, because he saw what no one else had seen. He changed the way we are.” Echoing Harold Bloom and his book Shakespeare: The Invention of the Human, he continues: “Shakespeare had the ability to ‘fuse the magisteria’ all the time in the most astonishing way that you just wouldn't imagine. And it's the secret of poetry too, just wrenching, yoking, absolutely resonant sentiments expressed, colliding [elements] that shouldn't collide, creating ironies filled with ironies, with things from the wrong category, co-existing, which is the essence of humor, but also the essence of metaphorical power.”
Fry resonates with the view expressed by neuroscientist and author, David Eagleman, that one of the fundamental requirements of a reader of a piece of literature is that they felt “the beating heart” of the author coming through the page, because art is intrinsically about a relationship between two humans, “there is the Bard, and there is the receiver...there's the mouth and the ear. There's the eye and the page or whatever the form is, there's the paint on the canvas and the person beholding and encountering.”
It is arguably this ability of language and art to encode, redefine (and even “invent”) the human experience that results in the overt power of LLMs to influence and impress us, but also to the sense of betrayal or deception when the perspective provided doesn’t comport with our reality, or represent our experience. Fry cites the example of AI-created music that is offensive to the listener, and agrees with the sentiment expressed by author and illustrator, Rob Biddulph, that “art is about translating something you feel internally into something that exists externally, whatever form it takes,” and, by extension, since machines don’t feel in a human way, they cannot make human-equivalent art, only “machine art.” As Fry puts it: “music itself is a technique, and it is a mathematical thing as we know. It's Pythagorean originally. It goes a long way back, and it's something you learn, and it's a sequence of waveforms…but songwriting is about the experience. Yes, the process is technical; you might learn all those little techniques for songwriting for the classic three and a half minute song, but [AI tools] wouldn't ever help write something that came from inside you and the experience you've had since you were a baby, growing up as a child into this state of adolescence that you're in now, and that's where your song should come from, not from a corporate [process] that made an AI. That is the insult to the human spirit, that is the destruction of the soul of the world,” to which Nick Cave referred in his letter to two young songwriters contemplating the use of AI to compose a song. Similarly, Fry observes that even though LLMs have been used to create poetry, it is a “Vogon gift for poetry,” referring to the lowest form of composition in the Hitchhiker's Guide to the Galaxy, that caused physical pain to the reader*.*
But Fry sees an equivalent role for AI in art as it is beginning to play for science and other areas of human creativity, “the G in GPT is generative,” which explicitly indicates the creative potential. He gives the example of composing a piece of fiction, “the three pillars of fictional art and prose are character, story and language. [AI]’s not bad at character and story…if you ask, I want to write a TV series, a crime series, I want four characters who work on the legal side and the police side and four who work on the criminal side. And give me some variety. [You think] that's pretty good. Give me some stories. Oh, that's pretty good, and then you start writing and you think, ‘Oh, shit!’” when the output is underwhelming. Fry points out that the reason is that these models are only useful up to a point: “it's a language model, but it’s not a character model or a plot model.” This is similar to the view expressed recently by Ben Affleck in a podcast, in which he describes the written output as “really shitty, because by its nature it goes to the mean—the average.”
Therefore, Fry finds the nexus of real value not in the emulation of human experience, but rather the exploration of alternative points of view, “difference yields meaning, as any semiologist will tell you…so what if AIs have no other value than they make us examine ourselves in ways that we've never properly examined.” This point of view is of course rooted in an ancient tradition going back to Socrates, who famously proclaimed that “a life unexamined is not worth living.” This ability to assist with human introspection and extrapolation is also reminiscent of the view espoused by Rick Carter that GenAI allows the exploration of strange new internal and external worlds, in ways that were previously impossible or prohibitively expensive.
Fry sees these creative excursions and explorations as more than just nice to have, they are in fact essential, “It's the things we don't need that make life worth living. If we only had protein pills and put our helmets on…things that kept us alive through absolute, accurate nutrition, and this is all a human needs to be alive, this much water, this much nutrition, this much sunlight, we'd say ‘no’; to be alive, you need wine, and don't just need sex, you need love. You need all these extra things that are in another magisterium entirely.”
In summary, Fry sees great potential for AI to assist or augment the human creative experience, but that this must be done with full transparency, so as not to deceive the receiver or deprive the artist of a means of making a living. He sees this as particularly relevant to emerging artists who do not yet have a catalog of material and a relationship to an audience that is unlikely to be impaired by copycat derivatives, “You have to care about the fact that there are also…young people coming, they could be the next Beethoven, the next Bob Dylan, whoever you choose as your hero, your staggering outlier—your once in a generation, once in ten generations’ genius. Humans are much more likely to pull one of these out of the hat…than a machine is at the moment, and possibly forever. It's just one of the freaks of evolution that you get these astonishing outliers in so many different fields and [we can’t] risk their not entering the field because the field no longer exists, or it's so small there isn't room or it isn't sustainable economically.”
But Fry is also refreshingly practical in his view of how to proceed with using AI in the creative realm. He is one of the leading signatories of an open letter to the U.K. government that contributed to the abandonment of a permissive “opt out” licensing proposal for AI model usage of IP. However, he sees the key is creative transparency and this manifests differently for different situations. For example, he highlights the case of a graphic designer using AI as a tool as part of a larger creative process, when there is no need to declare this usage explicitly, but human copyright should be acknowledged and respected. In contrast, when AI is used to create the majority of the finished work, we need not only to respect relevant copyright but also to “monitor and make any product of those [AI] machines declare itself with a signature,” akin to the way a human artist would, and “that it's watermarked in a way that it cannot remove.”
Taking Education “Back to the Future”
Fry sees another domain in which AI has a pivotal role to play: education. He again references the Latin origin of the word: “educare means drawing out, it's not putting in, it's [about] taking out, and that's not what education does. And that's the exciting thing that AI can do. It can inspire and help find out what the child is good at, instead of doing what education does now: find out what they're not good at and judging them, because they didn't answer the question.” Similar to Lila Ibrahim of DeepMind, Fry believes that AI has the power to tease out and enhance the unique abilities of each student or learner, and to do so in a way that is more globally accessible than the rarified atmosphere of conventional academia, with its increasingly competitive and elusive admission, and escalating price tag.
This compelling “back to the future” view of education builds on the ideas outlined for creative human endeavors, namely that AI can augment human capabilities—in this case learning—by helping explore the strangeness of each individual’s internal world and expanding their potential. He pithily summarizes the traditional academic system and its limited gravitational pull and orbital paths, observing that “A university of nerds and scholars is not a universe-ity.”
Characteristically, Fry summarizes the potential he sees for AI in creative augmentation and learning enhancement by reference to a classic piece of literature, Thomas Gray's poem, Elegy Written in a Country Churchyard , in which the author stands in a churchyard and “wonders, how many mute, inglorious Miltons are there who never got a chance, were never taught even to read and write.” His view is that AI tools that are accessible to all would clearly help fulfill this unrequited human potential.
A Risky Business
Despite the positive human potential, Fry is also deeply concerned about the risks attendant to these opportunities. He cautions that we must be wary of what he calls “the three C’s”: countries, companies and criminals. He observes that we live in an age when “countries are surveying the citizens, controlling people, companies are making money recklessly and criminals are stealing everything from us.” Moreover, he sees a particular responsibility lies with the current tech companies who are building the GenAI models, and with their leaders, who need to be held accountable. As he puts it, “Let's also concentrate on making sure that [we limit] the untrammeled power of the ‘flying blind’ tech people who are creating these models…they want to develop and develop and develop and develop without really caring, because they have never cared what it does to humanity and that isn't good enough. All the power is theirs at the moment, but all the power in reality is with the human population. And if we say loudly enough, ‘you must behave yourselves,’ then they will have to.”
Fry’s argument that the tech leaders—particularly the owners of social media platforms—need to be held accountable by the public, rather than relying on self-policing, stems from the observation that, as Yuval Noah Harari points out in his book, Nexus, the leaders of these companies chose an algorithm that maximized engagement, based on greed and without sufficient regard for the fact that “you'll be maximizing hatred, envy, resentment, lust, disgust, fear, all the Pandora's Box, all the negative things, not puppies, kittens and nice sayings from Confucius, they engage [people] for a second, but the other ones, people really engage, and they will drive them forward. And we know what happened as a result, and that the people behind it didn't give a f**k.”
But his concerns go beyond the further pollution of social media by AI. Fry cites the influential philosopher and cognitive scientist Dan Dennett, and the ideas expressed in his essay, The Problem with Counterfeit People, in which he states, “Today, for the first time in history, thanks to artificial intelligence, it is possible for anybody to make counterfeit people who can pass for real in many of the new digital environments we have created. These counterfeit people are the most dangerous artifacts in human history, capable of destroying not just economies but human freedom itself.” Dennett describes what he sees as a coming pandemic that will attack our brains using these counterfeit people to “impair our capacity to reason with one another by distracting and confusing us and by exploiting our most irresistible fears and anxieties and will talk us into adopting policies and convictions that will make us vulnerable to still more manipulation. Or we will simply turn off our attention and become passive and ignorant pawns. This is a terrifying prospect.” Fry agrees with Dennett that we need to treat the counterfeiting of people similarly to the counterfeiting of money, with criminal consequences for those involved in the creation, propagation or utilization of this alien form.
Fry also thinks there is a non-negligible existential risk to humanity posed by the unfettered deployment of AI. He agrees with the findings of Phil Tetlock’s existential risk prediction forecasting study and points out that “even the optimists are saying it's a 5 percent or 7.5 percent chance that humans will be wiped out,” on the long time horizon of 100 years. He makes the salient observation that if one compares this probability with that of winning the lottery—a chance that a significant proportion of people are willing to entertain, people must “wake up” to this reality: “I believe in the inherent goodness of community behavior, but it has to be awoken. AI is like the opposite of mushrooms; It doesn't flourish in the dark, it needs light.” Fry was notably one of the signatories of the recent open letter to the U.K. government that calls for a prohibition on the development of superintelligence, and that this should not be lifted before there is both broad scientific consensus that it will be done safely and controllably, and strong public buy-in.
He is sufficiently alarmed that he is contemplating a more active role in making people aware of these risks, using his connections in the entertainment industry, “I am trying to prepare a film called P doom! P doom!,” a play on P(doom), the probability of an existentially catastrophic scenario caused by AI. He playfully suggests that he will “see if I can get Kylie [Minogue] to re-record her song, ‘Padam, Padam,’” and create a thought-piece equivalent to Al Gore's An Inconvenient Truth about the existential environmental risk, with the goal, as Fry puts it, “Basically to say…wake the f*** up.”
Fry again makes a simple analogy with a prior era when a seismic shift was occurring that required overt caution to be employed; he sees the current need as akin to the person who walked in front of the car in the early days of automotives, waving a flag that indicated to all that they should proceed with caution. He argues that an equivalent system is required for AI today so that, although “it may well be that the [analogous] motorways will be built and the motels and the whole landscape will be transformed,” we will have proceeded with appropriate human-augmenting-not-replacing caution. He argues that this caution is called for as it is impossible to align the interests of artificial forms with human needs and desires—the so-called “The Alignment Problem.” Fry sees that the problem with aligning values is not new, “this has existed before, the genie problem, it used to be called. You know, you ask a genie to stop all suffering in the world, and instantly all life forms are ended. ‘Oh, I didn't say that,’ you say, to which the genie responds ‘You asked me and your wish is my command, Master. I have ended all suffering’.”
Final Fry-Up
Summarizing the complex but compelling view on AI that Fry espouses comes down to the description of a dramatic and dynamic tension between two forces. On one hand he sees the potential for AI to enfranchise, educate and enliven our lives in ways that has, until now, been the stuff of utopian science fiction. As he puts it, “everything AIs do is unimaginable. That's why we want them.” This desire is likely rooted in our innate status as possessing what cognitive scientist, David Chalmers, calls “extended minds” that combine the brain and the associated neurological and physiological systems and allow us to also become "natural-born cyborgs" with hybrid thinking systems. Philosopher Andy Clark eloquently extends this view saying that this results “not in simple ‘offloadings’ of work, so much as the creation of delicately interwoven new wholes—brain, body, world tapestries in which what the brain does, what the body does, and what the loops via external media and apps provide are all in continuous flux, each adapting (in its own way, and at its own timescales) to what the rest has to offer.” Fry sees this human-extending role of AI not only extending artistic creation, but also scientific endeavor, and cites medicine as one area of particular promise and that AI “make help keep us alive” by helping discover solutions to problems of previously impenetrable complexity, like the remarkable success of DeepMind’s AlphaFold in predicting the folding of proteins and the attachment to other biologically important molecules.
However, this very ability to extend human existence and abilities gives rise to competing force: the manifest risk associated with these systems being corrupted or acting nefariously. Fry sees a clear and present danger from companies, countries and criminals behaving in immoral or amoral ways, driven by myopic self-interest rather than communal human values, and producing illusory, deceptive counterfeit people and poisonous—or just plain inhumane—perspectives. Indeed, this existential risk has been the topic of much recent conversation, for example in Dario Amodei’s essay, The Adolescence of Technology, and comments from Demis Hassabis at the recent AI Summit in India. Fry identifies the risk of governments using AI to build autonomous weapons and carry out assassinations as the most pressing (and depressing) doom scenario, and this has been chillingly borne out in recent conflicts, including during the latest U.S. and Israel attack on Iran, when Anthropic’s Claude models were reportedly used for targeting and intelligence mining.
Looking at the interplay of these two forces, Fry describes the current status of AI as both “thrilling and chilling.” There is also the intermediate risk that jobs will be replaced, but on this front he is more optimistic that in many cases the jobs replaced will be those that “insulted the human spirit” and are a form a “wage slavery” due to their mindlessness. He references the invention of the steam shovel in the 1830s that displaced the Irish “navvies” who built early railways and dug the canals, as an example of a soul-destroying and backbreaking role that was better performed by a machine, so this replacement represented real societal progress with undesirable jobs morphing into other less laborious, more human-valuing work, over time.
Fry is also positive about the potential for humans to rediscover the value of human-made things and “the touch and feel, the texture, the ‘whatness’ of things” by comparison with artifacts produced by machines, in much the same way that artisanal production and analog audio experiences like playing a record have been revalued in recent years. In many ways, these trends may be bellwether signs for the future—a future in which, on balance, Fry sees, if due care is taken, “ultimately the threat from AI will drive us to learn to value our humanity.” And that sentiment is one that is truly thrilling and around which there should be no alignment problem.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines