New Weapons in the War Against Crypto-Parasites
I was thrilled to see a new study today released by Receptiviti, co-founded by the eminent Social Psychologist James Pennebaker. Pennebaker pioneered the use of rules-based AI for psychological assessment in the 1980’s and 1990s, and in this new paper, his team convincingly shows how they can detect the deceptive speech in FTX Founder and ex-CEO, Sam Bankman-Fried who appears to have been the Bernie Maddoff of the crypto industry in 2022, defrauding billions. Thoughtfully, the paper compares SBF’s pattern with typical CEOs, and another crypto industry pioneer who is known for being conscientious, Brian Armstrong, the CEO of Coinbase. Receptiviti is able to show that they can detect his deception, and it’s noticably higher than a typical CEO, or Brian.
Pennebaker’s estimates show it is possible to detect bad actors in the crypto industry before they cause harm to teams and organizations. As great as the paper is, it suffers from psychometrics from the 1950’s. In the newer, interdisciplinary approach to metrologically-oriented psychometrics, what Receptiviti does, while impressive, isn’t even measurement.
For example, all measurement has uncertainty, but Receiptiviti can’t estimate these, and because they use norms rather than objective measures, the user can’t be sure they can trust whether those norms generalize to new populations, especially in other countries, industries, etc. These “Classical Test Theory” approaches also have unequal amounts of noise in the very areas (low and high) that matter the most, with bad actors like SBF. Those problems don’t exist with Rasch Measurement.
Similarly, Pennebaker’s approach doesn’t use either of the most exciting parts of Computer Science. First, Receptiviti didn’t (and doesn’t) use any of the new Large Language Model (LLM) machine learning that substantially outperforms his rules-based AI approach. Similarly, it executes it’s code in web2 cloud APIs meaning it is not immutable, composable, censorship resistant, or unlikely to crash the way blockchains are.
Without modern psychometrics, it’s hard to know how much noise is in each measure, and without modern computer science, we won’t be able to quickly create a wide variety of other automated job-relevant assessments, on-chain, with immutable, composable properties that blockchains provide.
Even so, it is exciting that even with Pennebaker’s outdated methods, the blockchain industry can detect and mitigate this multi-billion-dollar risk. Imagine a future where multiple dimensions are on-chain, automatically screening Discord, Telegram, and Twitter, and using modern psychometrics AI to not only detect these types of human risks, but also give automated feedback to help DAO members and decision makers improve (and lower the odds of rugs, exploits, scams and bad performance).
In conclusion, while this study is a step in the right direction, it highlights the need for modern approaches to detecting bad actors or less-than-effective behavior in the crypto industry. With the advancements in AI and psychometrics, it is now possible to create a safer and more secure environment for people to work, grow, and thrive in the blockchain industry.
