【Article by IAIC】

In early 2004, not long after Jackie Lee had arrived at MIT, a friend from Harvard leaned in with a tip: "There's this thing called Facebook. You should try it." Two years later, another friend told him to put his research videos on YouTube. Only in hindsight did Lee recognize these moments for what they were, "messages from the future." At the time, the shape of things was still blurred, and he could not have known how violently all of it was about to reshape the world.

On October 13, at the invitation of Dr. Lyn Chen, the Assistant Professor of the Interdisciplinary Artificial Intelligence Center (IAIC) at National Chengchi University, Dr. Lee, an MIT Media Lab graduate who has since founded an AI startup in Silicon Valley, spoke at the Yu Shan International Hall of the College of Commerce on "Challenging Silicon Valley and the Future of Innovation."

What he holds most dear from MIT are two things: a circle of friends who still look out for him today, and a peculiar closeness to knowledge. On that campus, he recalled, every answer was only one degree away. As a doctoral student in MIT's Affective Computing group, he created Lover's Cups, a pair of cups in which one begins to glow the moment its distant partner takes a sip, and Shybot, a robot that draws near only to those who approach it gently. From these playful inventions came a lasting insight: our emotional signals always arrive before our actions do, yet arriving first is no guarantee of being understood.

Lee was disarmingly frank about his own venture. It is, he said, not yet much of a success. Silicon Valley teems with formidable players, and a small team can only keep feeling its way forward.

Studying abroad, landing in Silicon Valley and building a company of his own had once been his American Dream. In the golden age of virtual reality he founded ScienceVR, a virtual laboratory where learners could watch invisible magnetic forces come alive and run experiments side by side with James Clerk Maxwell and Marie Curie. Then the pandemic came, and no one wanted to share a headset anymore, so he moved into the metaverse. Then the AI wave broke, and the metaverse fell silent. "You always need to pivot and adapt," he said.

The age of AI sent him back to a question as old as philosophy: what, exactly, is knowledge? Drawing on physicist David Deutsch's reflections on genes and memes in The Beginning of Infinity, Lee argued that knowledge is something other than data or information. Knowledge must be actionable. It carries causal power, the power to make something in the world turn out differently. That conviction now runs through the causal networks he builds for AI agents, and through a "secret project" he allowed the audience to glimpse, in which AI, augmented reality and video generation come together to anticipate where a person or a vehicle might go next, laying out a branching set of possible futures. The work carries forward his early fascination with human behavior, and pushes it toward a harder question: can AI understand a situation well enough to foresee what happens next? He began, he said, in affective computing, went on to build a metaverse, and is now building something new: a "multiverse."

Auto-research, he said, is one of the most exhilarating frontiers of all. Define a metric that can be measured, and an agent will try, fail, adjust and try again, without tiring.

Given the right permissions, such systems can operate a computer and carry out tasks on their own. Lee described sending his AI agents into a research contest on quantum computing and into Frontier-CS, a computer science challenge. What makes these contests work is that every submission can be checked and scored, so competitors can measure themselves against a public leaderboard and keep reaching for a better answer. The mathematical tools and infrastructure he had built for causal modeling and multiverse simulation, he noted, proved just as powerful when it came to solving these problems.

Yet handing the work to AI does not write humans out of the story. To push his results further, Lee had to keep learning how his agents actually worked, refine the tools and strategies he gave them, and trade techniques with competitors from fields far from his own. The contest community even went on to turn its discoveries into a research paper, so that what the agents had attempted became something people could study and learn from together. 2026 is the year of AI agents, and the gap between those who know how to use them and the many who still do not is precisely where our opportunity lies.

During the Q&A session, students asked about the implications of these developments.. One asked whether total automation would cost us our connection to one another. Another asked whether AI could ever truly understand love. Lee smiled and turned the question around: "Do you understand love?"

Humans may end up doing less of the work. But we are still here, and our task is to recognize what is truly great.

A person may look perfectly calm while their heart races at 150 beats a minute. Affective data can see what a composed face conceals, yet AI will never predict a human being with complete certainty. Lee was blunt that no one writes code by hand anymore. The engineer's job now is to direct AI and make sure its results can be verified. Don't trust what AI says, he told the room. Trust what it did, because that is something you can verify.

And so the story returns to where it began. Lee still regrets that he did not heed those early messages from the future. More than twenty years later, standing at a lectern at NCCU and speaking of causality, of verification, of where people stand before their machines, he may have been delivering another such message. When AI can go to the frontier for us, write our code and win our contests, we still have to know how to ask the right questions, how to verify, and how to recognize what is truly worth a life's pursuit. The message has been sent. What remains is who will write it down, and then act on it.