I recently read an astonishing paper titled AI Must Embrace Specialization via Superhuman Adaptable Intelligence, co-authored by none other than AI godfather Yann LeCun. It seems fair to say that the godfathers are having a bit of a spat. LeCun all but kicks Elon Musk and Demis Hassabis in the nuts.
The paper is remarkably antihuman in tone and substance. It’s also incredibly inflammatory and I expect it’ll provoke a visceral reaction, so stock up on popcorn. LeCun is clearly heretical and contrarian, but as a fan of the world model paradigm in AI, I’m theoretically sympathetic. I feel that he’s cheating, however. Claiming that humans aren’t generalists because AI can kick our asses at chess is like shooting fish in a barrel. He intentionally deprecates human specialness, perhaps motivated by a belief that AI will soon out-special us. The argument LeCun should’ve engaged with is that humans are the most general animal that’s evolved thus far: unlike squirrels for example, we’re general enough to fold laundry and fold proteins, and design AI too.
LeCun forcefully demotes humanity while ignoring the subjective experience of nonhuman intelligence, and it’s a jarring combination. Supposedly we flatter ourselves that we’re generalists, whereas on the contrary we’re merely specialists, and agentic AI must specialize too in order to successfully compete for scarce resources. Other than dragging Magnus Carlsen through the mud, the evidence for these extraordinary claims amounts to vociferous handwaving. LeCun is a partisan for unsupervised self-learning, yet he famously criticized AlphaGo for being too specialized. What is he really proposing? Hunger games for humans and AI alike? I have a bad feeling that there’s a dystopia concealed in his polemic. Though he’s amusingly abrasive towards his fellow tech lords, he has something in common with them: hubris. Everyone should listen to LeCun, as only he knows the great wisdom that makes everyone else look stupid, and so on.
I’m not implying that LeCun consciously intends dystopia, though I can’t rule it out. The problem is rather that dystopia flows naturally from his worldview. Like all tech lords, LeCun floats in a bubble of wealth surrounded by sycophants, and this blinds him to the prosaic realities mere mortals face. It’s his ruthless unconcern for the suffering of individuals that stands out. He seems obsessed with domination, unsurprisingly given that he spends his days dominating AI.
LeCun’s net worth is estimated at between 100 and 150 million dollars in stock value alone. That absolutely constitutes privilege, even if he’s quiet about it. His bosses have yachts and he’s steeped in their classism whether he knows it or not. His job predisposes him to determinism, behaviorism and utilitarianism, all harsh ideologies which can and often do lead to brutality. Notice that the plight of animals—hundreds of millions killed every day, as I keep reminding everyone—does not even appear in LeCun’s paper. The unquestioned assumption is that AI exists to smooth and prolong the lives of those who can afford it, just as animals exist to put dinner on plates.
LeCun’s language reminds me of Dr. Edward Pointsman, the fanatically Pavlovian antagonist in Gravity’s Rainbow. His paper is a philosophical treatise posing as a technical study, and that’s exactly why I’m so motivated to attack its explicit and implicit conclusions. I’m no AI researcher, but I’m definitely a philosopher. LeCun is out of his lane, and it shows. I expect his peers won’t even notice his presumption, because they’re busy electrocuting AI rats.
No comments:
Post a Comment