ActInf GuestStream 083.1 ~ "Embodied intelligence", Joshua Bongard

"Embodied intelligence" Joshua Bongard

It is still laughably easy to foil AI with adversarial attacks. Why? Because such systems lack embodiment. But dropping deep learners into robots and calling the result Embodied AI misses the mark: Embodiment is about more than just having a body; it is about change. Consider how much the world changes from the perspective of a human as she grows from one cell into 10^13 of them. Grappling with such massive internal change makes grappling with external change, like learning to read or drive, easy by comparison. Thus, to realize safe AI, we must similarly create autonomous technologies whose internal physical changes pretrain them to handle external change, like new tasks or adversarial attacks. I’m excited to discuss a future filled with soft and biological robots that are capable of this “morphological pre-training”, and how it will make them useful, smart, and safe.

https://www.uvm.edu/cems/cs/profiles/josh_bongard https://scholar.google.com/citations?user=Dj-kPasAAAAJ&hl=en

Xenobots, Biobots, and more!

Active Inference Institute information: Website: https://activeinference.org/ Twitter: https://twitter.com/InferenceActive Discord: https://discord.gg/8VNKNp4jtx YouTube: https://www.youtube.com/c/ActiveInference/ Active Inference Livestreams: https://coda.io/@active-inference-institute/livestreams ... https://www.youtube.com/watch?v=yknM6SFlPPc