Jensen Huang says AGI doesn’t really matter, and he may be right for the wrong reason
Skynet has missed another appointment. Depending on which Terminator timeline you prefer, the murderous artificial intelligence should have become self-aware in 1997, 2004, 2011, 2017, or perhaps last Tuesday. Each date has passed without a nuclear exchange. The machines have mostly occupied themselves by flattering users, mangling fingers in photos, and confidently inventing facts for gullible people to believe.
That has not stopped Silicon Valley from treating artificial general intelligence as both an imminent miracle and an approaching monster. AGI supposedly marks the moment a machine can match human intelligence across a broad range of tasks, though nobody can agree on which tasks, which humans, or what counts as intelligence.
Nvidia CEO Jensen Huang thinks it already happened and nobody noticed. “For many tasks, we could say that we’ve already achieved AGI,” he said during Nvidia’s latest earnings call. He added that such milestones are “kind of senseless at this point.” So much for Skynet.
Latest Videos FromTechRadar
But he is right that AGI has become close to useless as a measure of progress. Companies can redefine it according to whatever their newest model happens to do well. It’s a promotion, not a scientific standard.
Apocalyptic decoy
But AI isn’t harmless even if it’s not inclined to play with nuclear bombs. There are less theatrical dangers from AI fueled by old-fashioned human greed and malevolence.
Today’s AI can produce endless fodder for scams. It can clone voices and faces well enough to trick loved ones out of money, use phishing messages to hack secure servers, and flood social networks with synthetic accounts that never tire of arguing. None of this qualifies as AGI under any serious definition, but the distinction offers little comfort to the person hearing a fake recording of a relative begging for money.
The fixation on AGI may actually make life harder for regulators.
Profit isn’t a good yardstick
Huang argued that the concept of AGI should be replaced. We should consider if it is “doing productive and useful work” and “generating profitable tokens.” That makes sense for Nvidia, but to ordinary people worried about AI, profitability is not much of a safety test.
Scams are profitable. That’s why there are so many. Firing human employees for an AI model that is 30% as good at 10% the price is profitable (in the short term). Neither result is something to brag about when describing the result of your product.
Retiring AGI as the organizing obsession and fear of the industry makes sense, but that doesn’t mean AI companies can relax. Intelligence is too complex to serve as the sole trigger for concern. Power might be a better lens. We can measure who uses AI, and how, as well as who is expected to clean up when they fail. Then we’ll know how explosive an impact it’s had on our lives.
Conversations about liability, identity verification, ownership, and other legal details don’t have the visual excitement of mushroom clouds, but regulators should make AI companies discuss them instead. We need to keep an eye on the human hands on the keyboard, not the perpetually delayed Skynet.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
The best business laptops for all budgets
Skynet has missed another appointment. Depending on which Terminator timeline you prefer, the murderous artificial intelligence should have become self-aware in 1997, 2004, 2011, 2017, or perhaps last Tuesday. Each date has passed without a nuclear exchange. The machines have mostly occupied themselves by flattering users, mangling fingers in photos,…
