Filip Dvořák - Agentic AI -where symbolic and subsymbolic modeling converges

While subsymbolic modeling (machine learning, large language models) excels at capturing statistical patterns of the world from historical data, symbolic modeling (constraint satisfaction, SAT solving, planning, knowledge graphs) provides efficient reasoning on top of explicitly captured mechanics of the world with various levels of abstraction. The efficient interleaving of these two paradigms remains an open research challenge — one that underpins the next evolution of composite AI, neurosymbolic AI, and, in contemporary terms, agentic AI. In this talk, we provide an overview of the state of the agentic AI field, and we dive into the principles and techniques developed and deployed to production by Filuta AI across industries such as energy, logistics, automotive, dual-use, and digital entertainment.

Also check out other Contests in Prague.