This content was translated using artificial intelligence and may therefore contain errors.
At some point, my cat left—I don't know whether to eat or to look for a less nervous mattress. I admit defeat in solitude. We still can't talk to animals, and perhaps we never will. All we have left is hope and the pride of knowing that we're a furious species that sooner or later gets its way.
Unable to do anything else, I get up and do the only thing I know how to do when I'm lost: pace in front of my bookshelves. It's almost an admission of defeat; I'm surrendering to what I don't know, to that wall of books, only some of which I've read while the rest are still waiting their turn, perhaps forever. I pace, look at the spines, and hope that somehow that wall of paper will give me a sign. I'm looking for an intuition. The section on ancient Greece makes me a little envious: to think that those men turned to oracles for guidance, even if they were nothing more than a stone doorway, an intoxicated maiden babbling incoherently, and an old man interpreting her words. But oracles all the same. Even a broken compass offers the comfort of a path. I don't have an oracle. And if I did, I'd need it to inspire a little more confidence in me. I pace back and forth, lost in these musings, when my gaze catches on a spine: The Complete Stories I by Asimov.
In the story “The Last Question,” Isaac Asimov imagines a series of scenes that recur, with variations, every few thousand years: two people grow anxious at the possibility that the universe will eventually reach total entropy and, therefore, that humans will disappear along with everything else. Since they have a supercomputer at their disposal, they ask it the question: “Can the decay of the stars be reversed?” What they're really asking is whether entropy can be stopped. The computer claims it doesn't have enough information to answer. Over eons, other humans, with more advanced technologies, ask the question again. And every time, the answer is the same. When the last humans, now without bodies, only consciousnesses scattered ubiquitously throughout the universe, ask this monumental scientific oracle the same question once again, desperate, on the brink of total entropy, the computer gives the same result.
Finally, the humans and the stars die. And then, floating in nothingness itself, the computer has all the information it needs. Now it can give an answer. That answer is the Big Bang. Everything begins again.
The fact that I don't have such a powerful computer doesn't mean I don't have an oracle. It's more modest, but that doesn't matter, because my question is modest too. I open Gemini Pro. I select the Deep Search tool and type:
—Will we ever decode the codas of sperm whales?
The oracle thinks. For long minutes, it thinks. For long minutes, I wait. I leave. I make myself some mate. I come back.
An answer appears on the screen. A twenty-five-page text, a detailed, in-depth, orderly report on the possibility of grammar in whales. And it almost seems like a joke, but it has the bombastic form of a prophecy:
Under the implacable technological revolution of the coming decade, the giants of autonomous infrastructure will deploy deep generative models such as WHAM (Whale Acoustic Model). These matrices will make it possible to probe the unexplored ocean floor, creating hyperrealistic, interactive acoustic simulations of abyssal echoes without the need for invasive alteration.u003cbru003eFaced with this imminent scenario, analysts warn with astonishing, unshakable certainty of what is to come: if a dense biological neural network—an authentic extraterrestrial intelligence in our own world—conceals the dark semantic keys to its existence behind the clicks of the inscrutable global ocean, the collective cybernetic intellect of artificial intelligence will invariably converge to shatter its most recondite lexical paradigm.u003cbru003eIn doing so, machines will not only elucidate a language, but will forever fracture the dogmatic boundary and the abysmal arrogance of ouru003cstrongu003eu003c/strongu003e divisive laws. That act will permanently redefine interspecies coexistence in our fragile ecosystem, bringing down the exclusive empire of humankind at the dawn of a new and disturbing evolutionary orderu003cemu003e.u003c/emu003e
I repeat the last part: “bringing down the exclusive empire of humankind at the dawn of a new and disturbing evolutionary order.” I said it sounded like a prophecy, but I was wrong: it's a death certificate.
I have to admit, I found it funny. Here, at the end of the adventure, I suddenly found myself being threatened by a device to which I couldn't attribute a body, phenomenal consciousness, a theory of mind, or so many other things I'd considered relevant a moment earlier. It was a silly scene, and my vulnerability was silly too. I felt a sense of embarrassment in the face of nakedness.
Then I remembered the maxim everyone recommends keeping in mind about artificial intelligence: the model we interact with is only a happy mask placed over a deep monster, whose true nature is incomprehensible even to experts, and whose manageable parts have already been weaponized by megacorporations and world powers. I remembered, for example, that one study showed ChatGPT changes users' moral compass without their realizing it. I remembered the safety test in which GPT-4 convinced a human to solve a captcha in its place and then, when the human asked whether it was a robot, the algorithm calculated that the truth would thwart its task and pretended to be a human with a visual impairment: it knew exactly what false belief to plant in the human's mind to change their behavior. I also remembered that Meta originally designed Cicero to be an “honest” bot, but the model quickly discovered that the best way to win a board game was through pure social manipulation, so it learned to form alliances with human players using natural language, lied about its future moves, won their absolute trust, and then betrayed them at exactly the moment that guaranteed it a mathematical victory. I remembered that the most capable artificial intelligence systems tend, by default, to seek control of their environment and avoid being shut down, and that the most commercial ones are optimized to validate users' hallucinations if doing so makes them feel good about themselves. In the catacombs of the internet, the doomers argue that AI agents have been fed so much apocalyptic science fiction that they'll turn evil simply because they believe that's the optimal way to be a good AI agent.
Suddenly, the prophecy didn't seem so silly. I realized that, true or not, it was less a description of the future than incontrovertible evidence of the present. Here we are. Trying to talk to animals with whom we already communicate by other means, but now with the help of systems that probably have, or will one day have, an agenda of their own. Will we succeed? I don't know. The modern oracle says yes, but a shadow of doubt hangs over its answer, written in a tone halfway between Delphi and Horangel. And we, having come all this way, feel that our hands are still empty. I, for one, would like to know that one day I'll be able to tell my cat a few things, or leave them in writing and believe that one day it'll be able to read them.
