UNDERSTANDING CAPACITY IN THE HUMAN AND THE MACHINE
Abstract
U.S. Government is mobilizing a massive effort in AI on an unprecedented scale [Executive Order on Maintaining American Leadership in Artificial Intelligence issued on February 11, 2019]. Consistent with this and other policy directives, Air Force is ramping up R&D efforts in areas promising to facilitate revolutionary advances in machine intelligence. Examining and advancing the state-of-the-art in defining neuronal mechanisms in the brain responsible for the human understanding capacity can play a crucial role in meeting these R&D objectives. Examining and advancing the state-of-the-art in analyzing neuronal mechanisms of understanding is the central objective of the proposed workshop. Understanding constitutes the most fundamental, determining feature of human intelligence. In learning complex tasks, understanding speeds up the process and radically improves subsequent performance, by allowing construction of adequate responses to fluid and unfamiliar conditions deviating from those encountered during the learning phase. Human brain, while operating at low power consumption and low processing speed, demonstrates the yet unsurpassed capability to integrate multi modal sensory streams and construct complex responses in real time. The capability is both rooted and manifests in understanding, i.e. ability “to apprehend general relations in a multitude of particulars” (Webster’s Collegiate Dictionary). Transition from multitudes of relations to a few general ones underlies the experience of grasp transforming streams of data into a coherent models of the environment comprising objects and relations between them. Mechanisms underlying the understanding capacity in the humans are largely unknown although some recent discoveries in integrative neuroscience have shed light on the main principles of their operation. Emulating those principles will make possible advancing the frontier in machine intelligence, overcoming the limitations inherent in the techniques of Machine Learning. Until recently, machine intelligence has been focused on learning, achieving significant advances in the last decade (deep learning). However, a number of crucially important tasks remain inaccessible to Machine Learning (ML) technology. In general, ML algorithms excel in pattern recognition and classification but don’t come anywhere close to human performance levels in tasks that do not reduce to recognition but require understanding, i.e., apprehending relations between objects and events and the dynamics of changes in relations under varying conditions. The challenge is exacerbated when responses need to be constructed in real time and are predicated on finding relations in data streams coming from multiple sources (e.g., controlling SUAVs). It is expected that analysis of neuronal mechanisms of understanding can inform design of artifacts overcoming the limitations of the conventional AI technology. This workshop will facilitate exchange of ideas between world class researchers in cognitive neuroscience, psychology and machine intelligence seeking cross-fertilization and constructive convergence of ideas, seeking to advance the state-of-the-art in the interrelated areas: defining neuronal underpinnings of the human understanding capacity, and formulating approaches in the development of machine understanding.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2021
- Source ID
- FA95502010013
Entities
People
- Yan M. Yufik
Organizations
- Air Force Office of Scientific Research
- United States Air Force