Yann LeCun, chief researcher at Meta and a pioneer of deep learning, believes that current artificial intelligence systems will have to go through a development path lasting several decades before these AI configurations will get some semblance of sentience, having a core in the form of the common sense, expanding their capabilities beyond the creative generalization of huge arrays of textual materials.
This opinion on the prospects of digital intelligence in terms of strengthening its functional base does not coincide with the point of view of Nvidia CEO Jensen Huang on a similar issue, who believes that machine thinking models in less than five years will be competitive in the context of their comparison with the human mind and will be able to surpass humans in solving many tasks that require significant cognitive effort.
Yann LeCun, at an event dedicated to the 10th anniversary of the fundamental AI research team at Meta, commented on Mr. Huang’s position on the mentioned issue, noting that the company he heads can benefit a lot from the passion for artificial intelligence. Speaking in a figurative and symbolic format, the chief researcher of the technology giant under the leadership of Mark Zuckerberg said that there is currently a war on machine intelligence, and Nvidia is a weapon supplier. It is worth noting that against the background of rapid growth in the AI area, Jensen Huang’s company was able to significantly increase profits by manufacturing GPUs for advanced digital intelligence systems.
Mr. LeCun also stated that the belief in the demand for artificial general intelligence is a platform for increasing the level of demand for the mentioned processors. In this case, it means attempts to create such AI configurations that, in terms of thinking capabilities, are comparable to the power of the human mind. Currently, researchers from companies such as OpenAI are developing projects aimed at creating artificial general intelligence. This activity increases the need for Nvidia-manufactured chips.
Yann LeCun suggests that society is likely to receive machine intelligence corresponding in terms of the capabilities of the intelligence of cats or dogs, and only after years have passed, digital thinking systems will reach the level of human cognitive abilities. According to him, the current attention of the technology industry to large language models and text data is not enough to achieve this goal. For decades, many researchers have dreamed of so-called human-like artificial intelligence systems.
According to Mr. LeCun, the text is a very poor source of information. He says that it would probably take 20,000 years for a human to read the amount of text that was used to train modern large language models. According to him, an AI system trained on the equivalent of two tens of thousands of years of reading still cannot understand that if A coincides with B, then B coincides with A. Yann LeCun says that there are many elementary things that artificial intelligence configurations do not understand due to the current learning methodology.
The Meta AI team is actively exploring how the so-called transformer models used to create apps, such as, for example, ChatGPT, can be adapted to work with various data, including audio, video, and graphical information. There is an opinion that the more these configurations of machine intelligence can detect hidden correlations between different types of specified data, the higher their potential for achievements in the sphere of cognition of the surrounding space will be.
Some Meta studies involve working on creating software that helps people learn how to play tennis better using Project Aria augmented reality glasses that combine digital graphics and the real world. In this case, the specified device will display visual prompts during the game on how to hold the racket correctly and swing arms. The machine intelligence models required for the operation of this digital tennis assistant require a combination of three-dimensional visual data in addition to text and audio if the AI configuration deems it necessary to communicate with the user.
The specified multimodal variations of artificial intelligence are a new frontier in the advanced technology industry. In this case, the development process involves significant financial costs. Meta and Google are actively exploring advanced models of machine intelligence. Against this background, the demand for Nvidia products is growing.
Jensen Huang’s company has become the largest beneficiary of generative artificial intelligence. The expensive GPUs that Nvidia produces have become a standard tool used in training large language models. Meta applied 16,000 Nvidia A100 GPUs to train its Llama AI software.
Yann LeCun, answering a question from media representatives about the need to expand the number of hardware suppliers against the background of the creation of complex machine intelligence configurations by the technology industry, said that in this case there is no reason to talk about something critically important, but noted that it would be nice if there were more companies producing products for the development of advanced AI models. He also suggested that computer chips of the future might not be called GPUs. Moreover, Mr. LeCun expressed the hope that new microcircuits will appear, which he defines as neural accelerators of deep learning.
Meta’s chief scientist is skeptical about quantum computing, which is used by technology giants such as Microsoft, IBM, and Google, having spent significant funds on it. At the same time, there are experts with a different point of view. In their opinion, quantum computing machines can accelerate progress in those spheres where large amounts of data are needed, for example in the area of drug development. This point of view is based on the fact that such machines are capable of performing many calculations using so-called quantum bits, unlike conventional binary bits, which are applied in modern computing.
Yann LeCun says that problems solved with the help of quantum computing can be handled much more efficiently using classical computers. According to him, there is currently no clarity about the practical significance and possibility of creating quantum computers with real benefits.
Senior Meta researcher and former CTO Mike Schroepfer shares Jan LeCun’s point of view. According to him, useful quantum machines may appear at some point, but this will not happen in the foreseeable future and has nothing to do with what the technology industry is currently doing. He also stated that the artificial intelligence laboratory, created ten years ago and part of the Meta ownership structure, is the result of the realization that this technology will be commercialized.