Yuval Noah Harari's "Nexus": A Stark Warning About AI's Impact on Humanity

Nexus by Yuval Noah Harari review - the AI apocalypse

The Sapiens author offers an oracular vision of the end of humanity As befits a writer whose breakout work, Sapiens, was a history of the entire human race, Yuval Noah Harari is a master of the sententious generalisation. "Human life," he writes here, "is a balancing act between endeavouring to improve ourselves and accepting who we were." Is it? Is that all it is? Elsewhere, one might be surprised to read: "The ancient Romans had a clear understanding of what democracy means." No doubt the Romans would have been happy to hear that they would, 2,000 years in the future, be given a gold star for their comprehension of eternally stable political concepts by Yuval Noah Harari. In his 2018 book, 21 Lessons for the 21st Century, Harari wrote: "Liberals don't understand how history deviated from its preordained course, and they lack an alternative prism through which to interpret reality. Disorientation causes them to think in apocalyptic terms." It seems that, in the intervening years, Harari has himself become a liberal, because this book is about the apocalyptic scenario of how the "computer network" - everything from digital surveillance capitalism to social feed algorithms and AI - might destroy civilisation and usher in "the end of human history". Take that, Fukuyama. Like Malcolm Gladwell, Harari has a passionate need to be seen to overturn received wisdom. Many people think, for example, that the printing press made a crucial contribution to the emergence of modern science. Not so, insists Harari: after all, printing equally enabled the dissemination of fake news, such as books about witches, and so Gutenberg is partly to blame for the gruesome torture and murder of those accused of witchcraft across Europe. Silly as that might sound, it also misses the fundamental point: because the scientific method is accretional, modern science could only come into being once the results of previous experimenters were widely available to those who followed them. Only via the ladder of print could early-modern scientists stand on the shoulders of giants. But perhaps I have fallen prey to what Harari dubs "the naive view of information", which subtly changes throughout the book as rhetorical circumstances demand until it is something of a straw Frankenstein's monster. The naive view of information encompasses the idea that "[it] is essentially a good thing, and the more we have of it, the better", which lots of people believe and is hard to argue with, but it also supposedly holds that sufficient information leads ineluctably to political wisdom and that the free flow of information inevitably leads to truth, propositions that almost no one believes. "Knowing that e=mc2 usually doesn't resolve political disagreements," Harari says, to no one. We know this already from the history of dictatorial and totalitarian governments and their attempts at information gathering and control, a history from which Harari draws dozens of colourfully chilling anecdotes to persuade the reader of the falsity of a patently ridiculous view. What, then, can modern computers do that should worry us so much? Harari is peculiarly credulous about the capabilities of what is now marketed as "AI". No one has yet seen a chatbot create new ideas, as Harari thinks they can, let alone generate art that is not simply a probabilistic recombination of patterns in its training data. ("Computers can make cultural innovations," he writes in one of many passages besides which I scrawled "citation needed".) At some point in the future, meanwhile, what Harari calls "our new AI overlords" are apparently sure to acquire scary godlike powers. In his crystal ball, an AI overlord could decide to engineer a new pandemic virus, or a new kind of money, while flooding the world's information networks with fake news or incitements to riot. The descent prophesied herein of a "Silicon Curtain", meanwhile, is not a problem of AI per se but of geopolitics: extrapolating from the Great Firewall of China, which prevents most Chinese citizens from accessing sites such as Google and Wikipedia, Harari supposes that, in time, Chinese and US computer systems might be completely prevented from interoperating or even communicating with one another, so "ending the idea of a single shared human reality". Worrying if true. Vladimir Putin's violent irredentism in Ukraine is, Harari notes, in part inspired by his belief in a partisan version of Russian history, which shows the danger of a lack of shared myths. This, however, is a feature of almost all wars since the dawn of time, so I am not sure we can blame the computers for it. So what can we do to save human civilisation and our shared reality? Simple, concludes Harari: subject algorithms and AI to strong official regulation, and focus on "building institutions with strong self-correcting mechanisms". So, carry on being liberal democracies? It's a wan sort of conclusion to a book that has struck such an end-of-days tone. The annoying thing is that Nexus also contains many too-brief but fascinating discussions on subjects ranging from the process by which the books that comprise the modern Bible were canonised, or the role of Facebook's "news feed" in fomenting the massacres in Myanmar of 2016-17, to the facial recognition system used by Iran to detect unveiled women. There are a brilliant few pages, in particular, on the plight of Jews in fascist Romania, including Harari's own grandfather, who in 1938 were forced to provide papers proving their right to citizenship that in many cases had been destroyed by municipal authorities. When Harari is not in his mode of oracular pontification, he can be a superb narrative writer. But oracular pontification is what, we must assume, his readers want.

Review | Yuval Noah Harari takes on AI

In his latest book, "Nexus," the historian examines the history of information networks and the challenges posed by the AI revolution. A curious blind spot limits the persuasiveness of many recent books warning of the threats and challenges of our AI revolution: They consistently fail to acknowledge the extent to which they are themselves already reflective of the automation of so much of contemporary life. Yuval Noah Harari's latest book is no exception, as indeed at several points it betrays what at least looks to be an assembly-line process behind its creation. I do not know how many humans, machines, or collectivities of both had a hand in bringing Harari's "Nexus: A Brief history of Information Networks From the Stone Age to AI" into the world in its final form. But it does not give the overall impression of having been produced by a singular mind, laboring away in seclusion to deliver to the world its original insights. That said, the elevator pitch version of the book is a good one: It is a history of human information networks, broadly conceived to include everything from text messages to smoke signals, markings on cave walls, or indeed just face-to-face speech or gesture. This history is related from a perspective sufficiently zoomed out as to reveal both the real continuities and the surprising discontinuities between, say, cuneiform clay tablets and iPhones. Harari ultimately comes around to laying out his concerns about the rise of AI, but it's never totally clear what worries him, even if he frets that whatever it is threatens to change the nature of human consciousness for the worse. This latest book builds, of course, on Harari's string of tremendous successes, and it surely is a reflection of this success that the latest offering has the air of a big-budget production with several hands contributing to its manufacture. Harari, a scholar of the Middle Ages by training, has largely cornered the market on popular explorations of "big history," a trend within academic history that does not limit itself to expertise and close reading of archival sources from a narrow period, but rather considers all of human history within its vastly larger context, which includes not just the prehistory of Homo sapiens, but also primate evolution, the origins of life and even the Big Bang. Harari's previous work, especially "Sapiens: A Brief History of Humankind" (2011), has excelled in delivering to a mass readership some idea of how much we share with human beings vastly removed from us in time. With "Nexus," Harari's purpose is to look at humanity's future prospects, which are, he believes, significantly compromised by our most recent information technologies. In the book's opening section, Harari sets out to refute what he calls the "naive view of information." This is the idea that when humans gather more information, they have a greater share of truth, which will in turn bring them both greater wisdom and greater power. To the contrary, Harari argues, there is no correlation at all between information, truth and the ability to hold sway over other people. What binds human information networks together is not primarily the truth, but stories. This is why religions have been so successful over the millennia, even if they seem ripe for disproving. Religions "work" not because they convey the truth, but because they keep a large and often widely dispersed network of people connected, both to one another and to the story they all tell. The history of previous information revolutions bears this thesis out. A naive information theorist might suppose that the Gutenberg revolution led to a sharp decrease in the circulation of misinformation. Yet it did no such thing, Harari writes. While it did enable the publication of scientific treatises, it also facilitated the spread of wild conspiracy theories, futile sectarian hairsplitting and such travesties as the early modern European witch hunts. This attempt at a grand history of narrative begins to break down, however, when Harari insists on understanding narrative in biological terms. Harari has been much criticized, especially in academic circles, for repeating the great sin of E.O. Wilson, Jared Diamond and others: In their eagerness to portray human individuals and human society as part of a natural world that includes, for example, ants, they understand human morality as little more than a secondary consequence of strictly natural forces at work at a deeper level. There is much in "Nexus" that seems to justify this association. In one particularly weak section, Harari cites animal precedents for many classic human myths and folk tales. It is true that the strongest of a batch of tiger shark fetuses will devour its siblings while still in utero -- adelphophagy, it's called. But Harari takes such scientific observations as a reason to conclude that myths such as the Ramayana that feature sibling rivalry may appropriately be described as "biological dramas." One might just as easily look at the phenomenon of "loon fallout," where migratory birds sometimes fall from the sky to their death as a result of challenging atmospheric conditions, and conclude that the Sandra Bullock film "Gravity" (2013), about a space vessel in orbit being pulled perilously down to Earth, is likewise both a "biological" and a "physical" drama. Indeed, pretty much anything that happens to, or is done by, a human being has analogues in the animal and broader natural world. But what a story is, and the reason we tell stories, cannot be captured by appeal to biology, or indeed to the physics of planetary gravitation. The difference between a tiger shark eating its siblings and a human committing fratricide is that the latter is inevitably followed by court proceedings, memoirs, conversations with kin. This is what makes human fratricide into "fratricide," while using that word in reference to tiger sharks is only a façon de parler. This is also, though Harari does not seem to see it, what matters in the stories that he calls "biological dramas," and it is also the same thing that matters in stories that have no obvious animal counterpart. Harari's view grows more muddled still when he takes up the question of artificial intelligence. For him, Assyrian clay tablets, Gutenberg's presses and all the other information technologies that have brought us to the present moment could only record and propagate the information that human beings wished to see recorded and propagated. By contrast, Harari maintains, AI itself "decides" what information circulates through our networks. The notion of "decision" is central to his argument. In a key chapter opening Part II, "The New Members: How Computers Are Different From Printing Presses," Harari uses "decision," in its noun or verb form, upward of 13 times, by my count, to describe what it is machines now do, as well as related terms such as "choose" and "act." This vocabulary implies that computers possess what philosophers call "intentionality" -- the capability to follow courses of action that emerge from within themselves rather than from external mechanical pressures or from algorithmic "if-then's." Eventually, seven pages into the chapter, Harari does acknowledge that some readers will have a problem with his decision to describe the way machines act as essentially identical to the way humans do. But he bites the bullet confidently, I suppose to his credit in a perverse way, and simply insists that any purported difference here is an illusion. After all, he notes, "Human soldiers are shaped by their genetic code and follow orders issued by executives, yet they can still make independent decisions." That's a pretty quick and easy way to dispatch the enormous problem of the differences between human agency and machine intelligence. No matter what your theory of where morality comes from, it is a fact that we regularly put on trial soldiers who commit massacres, and the language used to describe what they have done is laden with moral notions such as culpability. If an AI were to cause a massacre, it would, by contrast -- one hopes -- simply be decommissioned. At worst, the humans who developed it might face prosecution. Reading Harari, I sometimes thought of Henry Miller's very different book titled "Nexus." To a friend who had harshly criticized the trilogy of which "Nexus" is a part, Miller replied: "If it was not good, it was true." Miller has in mind here the sort of "truth" that is experienced as an inner feeling, and that attaches to the hard work of honestly exploring one's own depths. This is work that no machine has yet given any evidence of being able to perform. This great difference between human beings and machines -- that we can look within while they cannot -- is also what explains why soldiers who decide to commit massacres can be held legally and morally culpable, even if bloodthirstiness is written into our DNA. This is a difference that is of little interest to Harari, which leaves us uncertain as to what exactly he thinks is being threatened by the rise of the machines. If our consciousness really is ontologically on par with AI, if AI makes "decisions" just like we do, but better, then it is not really clear what is worth preserving about our particular sort of agency in the world. Miller's "Nexus," by contrast, gives us a fairly compelling account of what makes us special: an understanding that interest in truth can be a moral project. Harari has nothing of the kind to offer. Nothing, that is, besides a fuzzy batch of misguided animal fables and often unpersuasive reflections on technology. In the end, this is a book about AI that shares in the prevailing spirit of the era of ChatGPT. This is of course not an accusation that the author in fact turned to that new resource to generate his book for him. It is only an acknowledgment that a lucid reader cannot help but see just how reflective this book is of the same worrisome trends that Harari, along with the entire team sustaining the Harari industry, proposes to expose and critique. Justin Smith-Ruiu's most recent book is "The Internet Is Not What You Think It Is" (Princeton, 2022). He is the founding editor of the online magazine the Hinternet. Nexus A Brief History of Information Networks From the Stone Age to AI By Yuval Noah Harari Random House. 492 pp. $35