A Skinner Box for the Age of Blockchain+AI
I feel old sharing with you that 25 years ago, my third patent was originally titled “A Skinner Box for the Information Age”. At that time as a young Organizational Psychologist, I was working in a semiconductor factory, ramping up the new processes to select, develop and manage the new technical workforce. Through the job and process analyses, I learned that the engineers were struggling to predict when the robotic arms needed preventative maintenance. It turns out that when robotic arms break, the entire production process breaks down, and cost the factory millions of dollars. I realized that the behavior of a robotic arm was similar to human performance, just more predictable, more reliable, and less uncertain.
A chronic challenge for all of psychology is to measure human behavior in experiments in ways that are as naturalistic as possible, while still being able to make good causal inferences. The problem with us pesky humans is that when we know we’re being measured, we tend to behave differently, as famously demonstrated in the well known “Hawthorn Effect”. I had learned about this as a special case of demand characteristics that psychologists try to avoid in studying human behavior, because we want to understand real, not contrived, behavior. My insight was that the lasers the engineers were using to monitor some aspects of the arms could be used unobtrusively to monitor human behavior, avoiding these demand characteristics, when using spectra of light that is not visible to the human eye. My system also enabled a new type of password, where you’d have to “dance” to get access to a computer system. I called it a “Skinner Box”, saluting the famous psychologist B.F. Skinner who used boxes to train pigeons (and even his daughters), using classical conditioning methods in the 1940’s and 1950’s.
While I was happy that I found a way to address demand characteristics, that invention was expensive, and required immobile hardware. That was long before Marc Andreeson famously said “software is eating the world” noting that many of the tasks that hardware used to do can now be done better, faster and less expensively with code. It was also before modern machine learning methods turned into the current exciting set of capabilities, such as the large language models ability to write text, and perform as a synthetic rater in an assessment. As I’ve written before in this blog, there are many exciting use cases for novel forms of AI assessment that are quality-assured by psychometric AI.
What I saw today that made me even more excited was another solution to a traditionally wicked problem in psychology. Historically, especially in organizational psychology, we struggle to measure job performance well, and in ways that give trustworthy feedback. Job performance has multiple dimensions - task execution, Organizational Citizenship Behavior, and Counter-productive Work Behavior, but many studies just lump them all into an unsophisticated rating scale that often has too many biases and noise that drown out the signal. Too often, managers are forced to live with bad measures of job performance that affect pay, promotion, and ultimately business performance.
I knew that AI’s ability to automatically tag meta-data would change this eventually, especially because firms like HireView have been doing this for many years with pre-employment testing of cognition, traits, knowledge and skills. But until today I was unaware that it was cost-effective to use to do the same with job performance measurement. Because this AI is expensive and tedious to create, I thought we were years away from using those types of data in a more widespread way for job-performance. What changed? Today I learned about the tremendous prototype called “Deep Strike” of the startup jabbr.ai. The image below shows real-time classification of specific types of punches, to specific locations in the opponent, using particular stances, all while synthetically scoring each blow!
This portends a new era of AI performance measurement that compliments the sort that HireView does on the attributes of employees. In the case of boxers, their physical abilities, personality traits, age, and skills all likely predict the degree to which any given punch may be more or less effective, and this level of detail not only helps fairly score competitions, but also helps fine tune the training of the athlete.
In the era of immutable, composable, decentralized smart contracts, where some user and engineer behavior is automatically written to a transparent blockchain, it’s easy to see that this sort of approach to workplace AI is a new version of my beloved digital skinner box idea. There are so many possibilities, on-chain, to leverage this type of AI without violating people’s privacy, but allowing them to get feedback and their organization (e.g. DAO) to monitor performance and automatically pay them. Given that world-class athletes have contracts that can be bought and sold (e.g. traded to another team), it also portends a new class of real-world sports contracts that have on-chain bonuses paid for developmental progress, and continuous improvement in performance that ultimately predicts winning competitions.
Just as one Decentralized Autonomous Organizations is trying to buy the Denver Broncos, perhaps DAOs can also become boxing managers, with on-chain triggers for effective job-performance, and ongoing development so that each fighter is more likely to win more fights (and prize money). In working with CryptoTrigs recently on web3, I came to learn that there is actually now a Real World Blockchain Association for physical assets, and I suspect they would also consider sports contracts like this in-scope for their work, along with commodities like real-estate.
If the potential for web3 and AI for HR intrigues you, feel free to join CryptoTrigs and me, sponsored by Wonderverse and Kleomedes DAO on twitter spaces on February 16, 2023 here https://twitter.com/i/spaces/1PlKQpkMBznxE?s=20
