Nvidia CEO Jensen Huang recently reiterated his view that society is five years out from having artificial general intelligence, or AGI, technologies that are capable of many of the same sorts of intellectual tasks as humans. These prospective technologies, he noted at an economic forum held at Standford University, are expected to be capable of many of the same sorts of intellectual tasks as humans.
His comments follow the statements he made during a New York Times DealBook Summit in November where he claimed AGI advancements are just five years away. Huang said that if you were to pose some of the same intelligence tests as are given to humans, in five years' time an AI will likely "do well" on "every single test you can possibly imagine."
Right now AI systems can pass some standard human tests -- witness recent headline-grabbing efforts like OpenAI's ChatGPT GPT-4 passing bar exams -- but struggle with more specialized topics. So while Huang is holding back from saying we'll soon see the emergence of a fully intelligent machine system -- something like the operating system from the sci-fi movie Her, capable of human-like interactions and problem-solving -- he can see that the current crop of slightly clunky, hallucination-influenced, unreliable AIs will very quickly develop into much cleverer systems.
Huang also underlined that the world will need more chip fabrication factories to enable AI to advance. This chimes perfectly with news that his company's stock price at closing on Friday marked the first time Nvidia closed with a value above $2 trillion. It also supports, in a way, OpenAI CEO Sam Altman's calls for trillion dollar-scale investment into new chips designed to process AI algorithms. Huang's overall point about AIs becoming smarter is also supported by recent work by scientists who found that very large chatbot AI systems are exhibiting tiny glimmers of understanding.
Huang did temper his bold prediction about AI's growth by noting that scientists are still working hard to fully understand how human brains work, so defining what a full artificial intelligence is may well be something of a moveable feast. That's supported by one of the co-founders of Google's DeepMind AI division, who recently argued that standard tests for how smart AIs are don't work well. According to Mustafa Suleyman, a better test would be if an AI could be a CEO of its own company; he pegged 2030 as the date by which he was certain AIs would be smart enough for this task.
But what does all this mean for everyone who's busy incorporating AI into their daily home, school, and work life today -- pressing Microsoft's new AI keyboard key and quizzing AIs for answers that previously they'd have posed to search engines? Not much, not right now. The debate on whether or not AI is directly threatening many people's jobs is ongoing, and expert opinions on the matter differ -- even if AI may have played a role in some of the recent high-profile tech industry layoffs.
Cleverer AIs will almost certainly emerge over time, as Huang says, but this is a revolution happening over years, not months. Industries that are likely to benefit from AI will have time to evolve and to choose how they use AI systems alongside workers. For example, when Adobe last week revealed a prototype AI system that can create music, the company was careful to note that the future of music creation probably involves a musician working with AI systems as a force multiplier instead of simply being replaced by the machine.
No comments:
Post a Comment