3 Comments

Interestingly, coming of Richard Herrnstein's primary field of expertise, Skinnerian behavior analysis, a new area within behavioral psychology called Relational Frame Theory (RFT) has emerged in the last twenty years. RFT attempts to develop a behavior analytic framework for studying the development of verbal behavior, cognition, and emotional conditioning. Its application has made some promising advances in enhancing human intelligence.

Relational Frame Theory extends the "Stimulus Equivalence" research conducted by Murray Sidman in the 1980s. Herrnstein and Sidman worked together at Walter Reed Army Research Institute in the 1950s, where Sidman published his seminal articles on avoidance training that became known as "Sidman Avoidance" or free-operant avoidance conditioning.

It has always intrigued me that Herrnstein was so interested in human intelligence because it was so much out of his field of expertise. He primarily performed research with pigeons and was well known for developing the "matching law" of reinforcement. When I read "The Bell Curve" in the 1990s, I also was struck by the lack of appreciation of the statistical subtleties and problems such as multicollinearity in the data they discussed. Skinnerian behaviorists typically perform very little statistical analysis in their research as they tend to rely on single-subject research designs.

Although I do agree that genetics accounts for a significant degree of variance in IQ, RFT-based training appears to condition the underlying cognitive skills that create what we have come to know as intelligence. After some promising initial pilots, we have the first somewhat larger scale study with school children replicates the earlier findings with a sustained effect of 8.9 point improvement in non-verbal IQ. This improvement was sustained with only 240 minutes of training over a four-month period. Earlier studies showed more significant gains in the 15 IQ point range with more intensive training. These training effects also have generalized to improved reading comprehension.

Another interesting factoid is that machine learning researchers at DeepMind have significantly improved training outcomes in reinforcement deep-learning protocols when adding relational frameworks in their training protocols. Unfortunately, most AI folks have little knowledge of the new RFT field. The traditional cognitive psychology folks at the big AI schools like CMU and MIT have a visceral disregard for anything coming out of the behaviorist camp. I was mocked and ridiculed by psychology professors at CMU when I attended grad school after working in behavioral psychology at Johns Hopkins School of Medicine. I think the RFT camp is on to something and could do an end-around the traditional mental-model building cognitive psychology folks. RFT could also have an impact on developing truly autonomous deep learning machines.

I have included reference links below. (Charles, I can forward you the pdf's)

Recent Study: Non-verbal IQ Gains from Relational Operant Training Explain Variance in Educational Attainment: An Active-Controlled Feasibility Study - https://link.springer.com/article/10.1007/s41465-020-00187-z

Background paper on RFT: Ian Stewart, The fruits of a functional approach to psychological science. https://doi.org/10.1002/ijop.12184

DeepMind article: Deep reinforcement learning with relational inductive biases | DeepMind -https://deepmind.com/research/publications/2019/deep-reinforcement-learning-relational-inductive-biases

Expand full comment

Fascinating read. Would love to get in contact with you again.

Expand full comment