Will Embodied AI Become Sentient?
- Embodied AI
- RC Trust
Abstract - Today's large language models have relatively limited interactionwith the physical world. They interact wth humans through theInternet, but even this interaction is limited for safety reasons.According to psychological theories of embodied cognition, theytherefore lack essential capabilities that lead to a cognitive mind.But this is changing. The nascent field of embodied robotics looks atproperties that emerge when deep neural networks can sense and act intheir physical environment. In this talk, I will examine fundamentalchanges that occur with the introduction of feedback through thephysical world, when robots can not only sense to act, but also act tosense. Processes that require subjective involvement, not justobjective observation, become possible. Using theories developed byTuring Award winner Judea Pearl, I will show that subjectiveinvolvement enables reasoning about causation, and therefore elevatesrobots to the point that it may become reasonable to hold themaccountable for their actions. Using theories developed by TuringAward winners Shafi Goldwasser and Silvio Micali, I will show thatknowledge can be purely subjective, not externally observable. Usingtheories developed by Turing Award winner Robin Milner, I will showthat first-person interaction can gain knowledge that no objectiveobservation can gain. Putting all these together, I conclude thatembodied robots may in fact become sentient, but also that we cannever know for sure whether this has happened.
Bio: Edward A. Lee has been working on embedded software systems for more than 40 years. After studying and working at Yale, MIT, and Bell Labs, he landed at Berkeley, where he is now Professor of the Graduate School in EECS. He is co-founder of Xronos Inc. and BDTI, Inc. He is a founder of the open-source software projects Lingua Franca and Ptolemy and is a coauthor of textbooks on embedded systems, signals and systems, digital communications, and philosophical and social implications of technology. His current research is focused on the Lingua Franca polyglot coordination language for distributed cyber-physical systems. More details can be found at https://ptolemy.berkeley.edu/~eal/biog.html.
