Skip to main content

From Machine Learning to Machine Psychology

Save to calendar

Sep 11

Date and time: 11 September 2023, 16:00-18:00 CEST (UTC +2)
Speaker: Robert Johansson, Stockholm University
Title: From Machine Learning to Machine Psychology

Where: Digital Futures hub, Osquars Backe 5, floor 2 at KTH main campus OR Zoom

Picture of Robert JohanssonAbstract: In the quest for Artificial General Intelligence (AGI), this work advocates for the incorporation of operant conditioning — a fundamental principle from behavioral psychology extensively studied within the context of learning and adaptive behaviors with animals and humans — into AI research.

We introduce Machine Psychology, a novel framework that merges behavioral psychology principles with the Non-Axiomatic Reasoning System (NARS), an AI model known for its real-time learning and adaptation capabilities under constraints of limited knowledge and computational resources.

This framework presents a promising path for AGI advancement by leveraging the profound understanding of learning, adaptation, and complex decision-making processes provided by behavioral psychology. Specifically, the design of NARS aligns well with the application of operant conditioning paradigms, offering a method to explore the dynamics of intelligence evolution and adaptability with computer systems.

We propose a methodology that involves the simultaneous manipulation of NARS’ experiential inputs and operational dynamics to investigate the conditions necessary and sufficient for achieving behavioral changes aligned with operant conditioning principles. This methodology enables a systematic exploration of adaptive behaviors with NARS, following a behavioral psychology-informed approach to progress towards AGI.

Bio: The main area for my research is in the field of Artificial General Intelligence (AGI). Our research is based on the premise that general-purpose intelligence can be seen as an instance of a kind of abstract response patterns called arbitrarily applicable relational responding (AARR). For a longer definition of AARR and how it relates to AGI, please see my 2019 paper linked below:

Johansson, R. (2019). Arbitrarily applicable relational responding. In International Conference on Artificial General Intelligence (pp. 101-110). Springer, Cham. Link to paper.

I also do research in clinical psychology, where I have a broad range of interests, with a particular interest in emotion-focused psychotherapy models. Please see my Google Scholar page for representative publications.

Link to speaker profile