What Embodied AI Means for Discovery and Humanity

Robots are quickly becoming more common and increasingly capable. Soon, robots that are deeply integrated with AI systems — an area that researchers call embodied AI — will be ubiquitous in the operating room, on the factory floor, and even in the home. However, one critical piece is still missing: a true robot brain.

In a wide-ranging keynote address on March 4, 2025, at the Columbia AI Summit, Sami Haddadin shared his vision for an advanced robotic brain capable of sensing and responding to the dynamic physical world, interacting safely and intelligently, and learning from experience. 

“We are entering an era where AI is no longer confined to screens and datasets but will actively contribute to discoveries in the physical world,” he said. “Just as telescopes and microscopes expanded our ability to observe the universe, embodied AI could be the next great tool to extend human potential.”

After the talk, Dean Shih-Fu Chang joined Haddadin for a deeper dive into the impact that embodied AI may soon have on research and the pursuit of new knowledge.

Shih-Fu Chang: You argued that embodied AI systems will soon help humans in the process of discovering knowledge, both at the individual level and scientifically. When these assistants become widely available, what role are humans best suited to play?

Sami Haddadin: The human will be asking questions, and the AI will be doing what’s necessary at much higher speed and accuracy to provide answers, in particular the ones that humans cannot see directly. Humans will bring intention to discovery by setting boundaries and using domain knowledge to tune the AI to its task. This should enable scientists to spend less time in the lab and more time doing highly intellectual work. 

Chang: What skills and knowledge sets should students pursue to prepare for this paradigm of Embodied AI?

Haddadin: There’s no one answer, but I think it’s important to be intentional about becoming a developer of embodied AI or a user of embodied AI. For example, medical doctors stand to benefit so much from these technologies. In addition to training them to use these systems, we should also train them to give active input regarding what competencies we should be developing. It typically takes a long time to bring these two fields together. In my experience, it often takes a year to two [years] of collaboration before we are able to speak roughly the same language. In training young talent, we have the opportunity to prepare them to be early adopters and communicators who can accelerate progress significantly. 

If you think of becoming a developer, it’s important to acknowledge that overcoming the limitations of current AI is a very hard problem. This will require a lot of mathematical and engineering knowledge as well as bridges to data science. I see a lot of dogmatism in AI research. The future depends on integrating and expanding these areas of knowledge far beyond what we see today. 

Chang: Columbia AI’s concept is AI + X, which advocates for cross-disciplinary approaches across a wide range of areas. What is the best way to conduct cross-disciplinary AI research?

Haddadin: The best areas of collaboration emerge around solving problems. I think of environmental sustainability or health as domains with real problems that AI can solve. You have a wonderful colleague, Sunil K. Agrawal, who is an expert in robots that assist with rehabilitation. He is a faculty member in mechanical engineering and his lab is at the Columbia University Irving Medical Center. This is one way these collaborations can succeed. There aren’t many places in the world where you can have this kind of intense integration. It’s why so many of the things he’s been developing translate to clinical reality. This proximity to a broad range of prestigious programs and researchers is the unique opportunity I see at Columbia. There are so many unique pockets of expertise that we can connect with to use AI in understanding domains, finding solutions, and building systems. 

Lead Photo Caption: Sami Haddadin (left) and Dean Shih-Fu Chang (right).

Lead Photo Credit: Eileen Barroso/Columbia University