Nexed

Most academic disciplines provide some sort of unique fantasy to their practitioners. Many philosophers seek to be virtuous, psychologists to be rational (and both aspire to be happy), and historians to predict the future. In each case, these academics wish to go beyond what their discipline legitimately allows. Sometimes this works, or at least it seems to work well enough, so that this sort of intellectual self-gratification is unlikely to die out.Y. N. Harari is a historian who found fame with his magnum opus Sapiens: A Brief History of Humankind (2014). Although such a project might be said to toy with grandiosity, it certainly teaches some important lessons, as do other works of this genre, such as Graeber and Wengrow's (2023) The Dawn of Everything: A New History of Humanity. Whereas Graeber and Wengrow's strategy is to make the reader question everything they thought they knew, Harari's strategy is to find patterns, tell stories, imagine dystopian outcomes, and to call for self-reflection and wisdom. In his new book Nexus: A Brief History of Information Networks From the Stone Age to AI, Harari acts much like generative artificial intelligence would (Hofstadter, 1995). Yet imagining sinister outcomes and calling for a human awakening and self-reform, he also casts himself in the role of a prophet in the tradition of the Hebrew Bible. The people have sinned by adoring false gods, he calls out. Unless they turn to the one true God (He who represents rationality, sanity, moderation, self-correction), they will be doomed.Harari seems a bit surprised by his global fame, and he is sensitive to the irony of being considered a thought leader on AI when he is merely a country historian. What makes his work so compelling? For one, Harari is a knowledgeable historian and a gifted storyteller. His narrative shows that he was enjoying himself writing it, so much so that the book is overwritten in parts, when Harari gets carried away with the joy of telling his tale. A work shortened by about 30% might have been just as effective. The main reason for the book's appeal, though, is the deft deployment of historical analogies, which, when heaped up one on top of the other, bring a pattern into focus. Time and again, Harari reports, humans have enhanced existing information networks or created entirely new ones: from the dawn of all-purpose languages to written scripts to telecommunication to the computer and now to the inevitable AI. Many of these revolutions of networked information happened independently in more than one place, which suggests they were “in the air.” It points to a deterministic causal ground and thereby questions the idea that humans could have acted differently. Nonetheless, Harari pleads with humanity to preserve and cultivate self-correcting institutions and practices—as still exist in science and in some well-tempered democracies—in order to avert the damages wrought by runaway technologies, technologies that threaten not only to become an alien and uncontrollable intelligence but also to dominate an enslaved humanity. Indeed, evidence for this development can readily be seen in the morning paper or the click on the news app. Nexus is written to offer a fluent and fluid reading experience. Observers from Aristotle to Unkelbach (2006) have found that the experience of easy perception puts a few hedons in your account (i.e., it feels good), but it also poses the risk of overestimating your comprehension of the material.By going beyond the historical record, Nexus raises the question of whether the flirtation with historicism goes too far. Historicism, Popper (1957) taught, is fatally flawed by its assumption that the study of history can reveal patterns that in turn reveal underlying laws of development. Once these laws are understood—per historicism—the future can be forecast and its arrival can be hastened. This error of false insight and fallacious agency was made by Hegel, and it was taken to tragic conclusions by Marx and his acolytes. Unlike Marx, however, Harari urges not the hastening of an inevitable future but the taming of an evitable one. As with Marx and other historicists, this is all meant to be in the interest of human welfare.What protects Harari's message from devolving into mere storytelling? The key operative concepts are information (and information networks) and intelligence. As to information, the reader looks in vain for a crisp definition. Harari dedicates Chapter 1 to laying the conceptual groundwork. He rejects the “naïve view” that information consists of symbols that represent reality and nothing else. He broadens the scope for information to include anything that can facilitate collective cognition by way of an acceptance of shared meaning (“order”). Stories and myths are information thanks to their intersubjectively achieved validity. They help hold a group or society together, and they enable further information to flow without having to internally represent anything existing externally. Before the advent of AI, state and institutional bureaucracies have been the principal bearers of rationalized mythologies. They bring order with the consent of those who are being ordered.The danger of this conceptual expansion lies in its looseness. Harari misses the opportunity to explain when data are no longer information. Because much depends on patterns and thus multiple data points, it is interesting to see how the arrival of a single pigeon in the fog of World War I is information about the enemy's intentions. The soldiers who received the pigeon knew enough to understand what it meant. They constructed meaning from background knowledge. The pigeon was just one piece of a puzzle, albeit a decisive one. The pigeon story is instructive with regard to both native and artificial intelligence. Human perception is “theoretical” in the sense that it is constitutes a “way of seeing,” a way that is shaped by memories, expectations, and desires (Felin & Holweg, 2024; Felin et al., 2021). Without the perceiver's input, a pigeon is just a pigeon and a cigar is just a cigar.Human perception is not always stable. The Necker cube and the duck–rabbit figure show that reality is not just received but co-created by the viewer (Attneave, 1971). Every good joke is a multistable figure; the experience of the perceptual flip makes us laugh. Consider the title of this review. Why might it make you smile? Because you see, I propose, the multiple allusions from Nexed to Next (as in prediction), to H. Miller's novel Nexus, to the colloquialism Nixed (as in cancellation). It is an informative title because it does not commit to one meaning. Perhaps AI will soon be able to appreciate a good quip, or at least credibly pretend that it can. Meanwhile, the psychological glory of information is that it is underdetermined. To come into bloom, it requires a receptive and sensitive mind. Harari could have made a better case for the role of information in the ongoing AI revolution had he defined it more clearly. What is lacking is a criterion of demarcation. When are data mere data but not information?Some classic theories of information are overlooked, or their implications for Harari's story are not probed. Shannon's (1948) theory remains an important guidepost today. Information is here conceptualized as the complement of uncertainty. A fair coin allows no advantageous betting; the more biased a coin is, the more certain is a win. This view is related to the concepts of entropy, noise, and unpredictability. An information gain results from any reduction in uncertainty. Parenthetically, we may consider the possibility that any change in uncertainty can be regarded as an information gain. To take a psychological example, it used to be thought that ordinary residents of New Haven could not be brought to electrocute a fellow citizen. When Milgram found that 60% of participants were willing to deliver lethal shocks, the old certainty was gone. Human behavior had become less, not more, predictable (Krueger & Massey, 2009). Yet we learned something, just as we did when contemplating this review's title.Creating and appreciating humor and wit is the mark of intelligence, the other key concept in play. Can AI do that? Perhaps it can, though without feeling. One of the cornerstones of Harari's exposition is referring to AI as an alien form of intelligence. AI is alien in that it is not like human intelligence in its inefficiency (energy and coolant consumption) and use of brute force processing (the need for massive training data), and it is intelligent by virtue of being an agent, a form or unity that makes decisions. Its decision making may be without self-consciousness, intention, reflection, or affect, but at the output level what AI does can look like intelligent behavior. It can beat a Go master, for example. Victories like that make AI look agentic.As with his incomplete definition of the concept of information, Harari again leaves readers wondering what to make of the concepts of “intelligence” and “agency.” To say that we all know what we mean is to beg the question. The two concepts seem tightly connected, one perhaps even entailing the other. The very term “AI” claims that intelligence has been achieved, not just a clever simulation thereof. But let's remember Douglas Hofstadter, the godfather of the term “AI” and a student of anthills. Consciousness arises, Hofstadter (1979) held, from the massive, distributed, but individually simple activity of neurons. The neurons are to the brain what the ants are to their hill. Although anthills and AI networks may lack consciousness, the latter can look like intelligent agents. Nature offers many examples of apparent intelligence. Sparrow murmuration is regarded as a type of swarm intelligence where the agents are the individual sparrows acting on simple rules (e.g., follow the sparrow ahead of you and leave a little distance) without knowing what the whole does or what the whole looks like to a sentient observer. If such natural phenomena are models for algorithms, aren't they subject to the same limitations (Chakraborty & Kar, 2017)? What does it mean to say that a computerized information network possesses intelligent agency?As to the consequences of the AI revolution, Harari admits to stressing the dystopia over the utopia. AI networks might spell the collapse of civilization and humanity at large, unless we wise up, that is. Others, perhaps in an effort to allay dystopic fears, have made a sport of showing how AI can be fooled. To name three examples, Gigerenzer (2022; reviewed in Krueger, 2024) notes that, being parasitically dependent on the data of the past, AI consumes exorbitant amounts of energy and cooling water to achieve brute force optimization. This works well in stable environments, but it is easily derailed by unpredictable change or deliberate mischief. Smith (2025), inspired by Hofstadter, reports that ChatGPT is so dumb, it doesn't understand that the game of chess would not change one bit if the bishop and the knight were to trade places and their permissible moves. Wagner et al. (2025) finds that ChatGPT is culturally inept, not having mastered the simple and commonsensical rules of good-faith communication laid out by Grice (1975).Since the scientific community has not figured out what thinking really is, it continually seeks refuge in analogy and metaphor (Krueger, 2007). The cognitive revolution cast the human mind/brain as a computer; now the AI revolutionaries want us to believe that artificial thinking will soon surpass native thinking on all tasks that matter, at which point the study of psychology might be declared moot were if not for a final trump card: How do we do it all with so much less energy and coolant? Meanwhile, the efforts of those who point to AI's embarrassing or amusing failures are Nietzschean projects. If to Nietzsche God was being squeezed into an ever-shrinking metaphysical gap by the advances of science, so has human intelligence become an intelligence of the gaps, now being defined by whatever the machines cannot do. Pragmatic and fluid types of intelligence still favor humans, but for how long?To boost his claim that information is the one decisive currency of the game, Harari notes en passant that some physicists now entertain the notion that information, not energy or matter, is the ground of the universe. Intriguing as this idea may be, it is probably metaphysical nonsense. There may be data at the ground of the universe, but there is no information unless a sentient perceiver takes a peek in search of meaning (recall the pigeon). If information is the stuff of the universe, who is the observer? Oh wait . . . Like all reviews, this commentary is a partial response to the author's work. Nexus might be regarded a postmodern product, floating, indeterminate, rich in hints and possibilities. Some readers may prefer firmer ground. Still, Nexus is thought-provoking in a way AI has not managed to be.