Yann LeCun, Meta’s chief AI scientist, has said that generative AI systems are still not even as intelligent as dogs because they don’t have an understanding of the underlying reality of the real world, CNBC reported. LeCun said that AI could surpass human intelligence someday and if it does, it should not be seen as a threat.
Speaking at the Viva Tech conference in Paris yesterday, LeCun said the trouble with generative AI we use today is that it’s very limited and purely trained on text. In contrast, a lot of human knowledge has nothing to do with language, he said. Therefore, part of the human experience is not captured by AI.
He also said that while generative AI can pass a Bar exam in the US - a test required to become an attorney - it cannot load a dishwasher, something that a kid could pick up in five minutes. “What it tells you [is that] we are missing something really big … to reach not just human-level intelligence, but even dog intelligence.”
In another example of an intuition humans develop that hasn't been replicated in AI yet; Yann LeCun said that a five-month-old baby would see an object floating and not make too much of it but just four months later, the baby would realize that the item shouldn’t be floating and would express surprise.
He admitted that researchers still have no idea how to replicate this type of thing in AI and while it eludes the researchers, significant progress towards animal and human intelligence won’t be made.
Since generative AI burst onto the scene last year with ChatGPT, two fears have come to the fore; will it replace human workers and will it pose a danger to humanity? LeCun thinks not.
He imagines that in the future everyone will have their own AI assistant that is smarter than them and will assist you in your daily life. He said the AI will be designed so that it's controllable and subservient to humans and that there’s no correlation between being smart and wanting to take over the world.
7 Comments - Add comment