Yann LeCun, Chief Scientist of Meta and a pioneer of deep learning, said that he believes that it will take decades for current artificial intelligence systems to reach some sense of perception, and that artificial intelligence systems with common sense can surpass the ability to just summarize large amounts of text in a creative way.
His views are diametrically opposed to Nvidia's
Nvidia CEO Jensen Huang recently said that artificial intelligence will be "quite competitive" with humans in less than five years, outperforming humans at many brain-intensive tasks.
"I know Huang," LeCun said recently at an event commemorating the 10th anniversary of the founding of the basic artificial intelligence research team at Facebook parent company Meta. LeCun said Nvidia's CEO has a lot to gain from the artificial intelligence boom. "This is an artificial intelligence 'war' and he is providing 'weapons'."
"(If) you think artificial intelligence is popular, you have to buy more GPUs," LeCun said of technologists trying to develop artificial general intelligence, which is on par with human-level intelligence. As long as researchers at companies like OpenAI continue to pursue AGI, they will need more Nvidia computer chips.
LeCun said society is more likely to have "cat-level" or "dog-level" AI years before human-level AI. The tech industry's current focus on language models and text data is insufficient to create the kind of advanced human-like artificial intelligence systems that researchers have been dreaming of for decades.
"Text is a very poor source of information," LeCun said, explaining that it would probably take humans 20,000 years to read the amount of text used to train modern language models. "Train a system with 20,000 years' worth of reading material and they still don't understand that if A is the same as B, then B is the same as A."
"There's a lot of very basic stuff in the world that they just don't get through that training," LeCun said.
So LeCun and other MetaAI executives have been working hard on how to tailor the so-called converter models used to create applications like ChatGPT to handle all kinds of data, including audio, image and video information. They believe that the better these AI systems can discover the billions of hidden correlations that may exist between these different types of data, the more miraculous feats they will be able to achieve.
Some of Meta's research includes software that helps people play tennis better while wearing the company's Project Aria augmented reality glasses, which blend digital graphics into the real world. Executives showed a demonstration in which a person playing tennis wearing AR glasses was able to see visual cues teaching them how to properly hold a tennis racket and swing their arms in the perfect manner. The artificial intelligence models needed to power such digital tennis assistants would need to be mixed with three-dimensional visual data in addition to text and audio, in case the digital assistant needs to speak.
These so-called multimodal AI systems represent the next frontier, but their development doesn't come cheap. As more companies like Meta and Google parent Alphabet work on more advanced AI models, Nvidia may gain a greater advantage, especially if no other competitor emerges.
The future of artificial intelligence hardware
Nvidia has been the biggest benefactor of generative AI, and its expensive graphics processing units have become a standard tool for training large-scale language models. Meta relies on 16,000 Nvidia A100 GPUs to train its LlamaAI software.
Some media asked whether the technology industry needs more hardware suppliers as Meta and other researchers continue to develop such complex artificial intelligence models.
"It doesn't need it, but it would be nice," LeCun responded, adding that GPU technology remains the gold standard for artificial intelligence.
However, he said, future computer chips may not be called GPUs.
LeCun is also skeptical of quantum computing, and technology giants such as Microsoft, IBM and Google have invested significant resources in it. Many researchers outside Meta believe that quantum computing machines could make huge advances in data-intensive fields such as drug discovery because of their ability to perform multiple calculations using so-called qubits, rather than the traditional binary bits used in modern computing.
But LeCun was skeptical.
"The number of problems you solve with quantum computing you can also solve more efficiently with classical computers," LeCun said.
"Quantum computing is a fascinating scientific topic," LeCun said. It's less clear "the practical implications and the possibility of making a truly useful quantum computer."
Mike Schroepfer, a senior fellow at Meta and a former technology director, agrees. He evaluates quantum technology every few years and believes that useful quantum machines "may appear at some point, but it will be too long a time horizon to be relevant to what we are doing."
“The reason we started the Artificial Intelligence Lab a decade ago was that it was clear that this technology would be commercialized within the next few years’ time frame,” Schroepf said.