Yann LeCun, a leading scholar in the field of artificial intelligence (AI) and a professor at New York University in the United States, as well as chief AI scientist at Meta, emphasized that rather than harboring vague fears about AI technology, the world should collaborate for better technological advancements.
On Nov. 11, Professor LeCun delivered a keynote speech at the ‘K-Science&Technology Global Forum’ held at the Fairmont Ambassador Hotel in Yeouido, Seoul, stating, “As large language models (LLMs) such as ChatGPT and Gemini emerge, they can pass the bar exam and even write poems, yet there are still no robots that genuinely assist our daily lives,” and added, “Current AI cannot even perform simple tasks that a 10-year-old child can do.”
Professor LeCun is referred to as one of the ‘four AI emperors.’ He is the scientist who first developed convolutional neural networks (CNN), significantly enhancing the performance of video data processing. In 2018, he won the ‘Turing Award,’ often considered the Nobel Prize of computer science, along with Yoshua Bengio and Geoffrey Hinton. He is also continuing his collaboration with Korea by co-directing the ‘AI Frontier Lab,’ an international cooperation project in the field of AI.
He explained that for AI to integrate into the lives of actual people, it must move beyond traditional LLMs that learn from human-curated text data. Instead, he advocates for an approach where AI can learn complex information based on visuals, similar to how newborns learn by observing the world.
Professor LeCun noted, “Current LLMs have learned from 2 trillion tokens (word data), but the information learned by a 4-year-old child amounts to 100 trillion.” He added, “Humans do not need to painstakingly enter the names of objects one by one and can grasp more complex relationships.”
Professor LeCun introduced an ‘objective driven AI architecture’ that specializes in learning video data to predict representations as images. For example, when showing an AI a 5-minute video, it is tasked with predicting the next scene.
He stated, “The algorithms used for LLMs cannot predict video data,” and added, “We are researching a model called JEPA, which has garnered attention from AI researchers in Korea and around the world.”
He expressed hope that ultimately the goal-driven AI architecture could realize a model known as AMI (Advanced Machine Intelligence) that benefits everyone globally. AMI is defined as a model capable of logical reasoning and planning, while still being safely controllable by humans. However, he emphasized that this requires global researchers to come together.
Professor LeCun mentioned, “Ultimately, we need to develop AI that accumulates all human knowledge and culture, integrating a universal value system,” and added, “All of this can only be achieved through collaboration based on open-source principles.”
He further noted, “Governments and scientists worldwide must collaborate to ensure that AI regulations do not hinder the spread of open-source platforms (core technologies),” and added, “We must also efficiently allocate computing resources to accumulate data that reflects each language and culture.”