Hoping to network with materials scientists and engineering focused peers with respect to work related to my algorithmic complexity and AI reasoning benchmarking system expanding into Material based intelligence (MBI). MBI is a paradigm motivated by my colleague Vladimir whom has done some work at IMEC. [2511.08838] Material-Based Intelligence: Self-organizing, Autonomous and Adaptive Cognition Embodied in Physical Substrates
Motivation
A lot of motivation ties into problems with AI literacy. Concerns regarding existing GenAI reasoning limitations. How can we develop practical expectations on what we can even expect agentic systems to do in terms of their capacity to perform series of tasks overtime that increase in complexity?
There are problems with reliability of these systems in which a lot of cases boils down to generalizing discretized outcomes of SGD. This ties well with Kenneth Stanley’s work on fractured representations. That is, largely language models don’t really understand the language they process. They don’t have deep subject and world understanding that humans embody.
Prospective implications for Material based intelligence (MBI)—presenting a spectrum of artificial to biological systems substrates and their tradeoffs
Existing and prospective utility of AI systems spans beyond the generative capacity of language model architectures and transformer as well as non-transformer variants. The work my colleague Vladimir produced in a recent arxiv preprint depicts a substrate axis of functional tradeoffs between artificial and biological systems.
On the biological substrate side the big tradeoffs leverage agency and operant conditioning dynamics over direct programmability. While on the artificial substrate side, the big tradeoffs are leveraging programmability as opposed to agency.