Knightian Uncertainty
Too good to be true? Adventures in ChatGPT-land

For decades I have been unimpressed by AI hype. (I had a boss at McKinsey in the early 1980s who insisted it was the wave of the future and I should learn to program in LISP).

Speech recognition continues to perform terribly (for me) and driverless cars scare me.

But I was an early adopter of (dumb) web-search, relying heavily on AltaVista when I started a book on entrepreneurship in 1997, switching over to Google as I went along.

So, I was both hesitant and keen to try out chatGPT for a chapter on behavioral economics in the book I’m now writing (on Knightian uncertainty, as it happens).

I was blown away by the “chatbot.” Here’s why.

I had read quite a bit about psychology and its relationship to economics.  My reading included academic journals, the Encyclopedia Brittanica and International Encyclopedia of the Social Sciences, Cambridge University Press histories – one on the history of behavioral economics and the other on the conceptual history of Psychology, Nobel Lectures and Biographies, and (parts of) an insider tome, Advances in Behavioral Economics.

It’s a jungle out there I found, and easy to get lost in the underbrush.  It was telling that the book on the conceptual history of psychology did not even attempt to define ‘psychology.’  The Association for Psychological Science website likewise doesn’t define “psychological science” or specify its major branches (although it will trumpet its commitment to justice, equity, diversity etc.)

Additionally, the links between psychology and philosophy and physiology seem far deeper and more significant that the links between psychology and economics.  Indeed I was coming around to the point of view that the success of Kahneman and Tversky in synthesizing cognitive psychology and economics turned on how reductive and shallow the ‘cognitive psychology’ they drew on actually was.  They more or less ignored the role of feeling (‘affect’) for example.

This seemed like a good place to start my journey through chatGPT: “What as the difference between cognitive and affective psychology?” I asked

A: “The main difference between cognitive and affective psychology is that cognitive psychology focuses on the study of mental processes such as perception, attention, language, memory, and problem-solving, while affective psychology focuses on the study of emotions and their effects on behavior and cognition. Cognitive psychology is concerned with how people process and store information, and how they use this information to make decisions and solve problems. Affective psychology, on the other hand, is concerned with the psychological processes involved in the experience, expression, and regulation of emotions, including emotion recognition, emotional memory, emotional intelligence, and emotional regulation.”

This was cogent, but I wasn’t sure this was entirely correct. My previous reading had left me with the impression that cognitive psychology does not exclude affect, but it was fuzzy. And if the Association of Psychological Sciences wouldn’t define its major sub-divisions how could I fault chatGPT?  Perhaps there was no clear answer to my question.

I pressed on:

Q: “How did psychology affect economics before the development of behavioral economics?

A: “Before the development of behavioral economics, psychology has influenced economics in a number of ways. For example, many economic models and theories, such as utility theory and the rational choice model, are based on assumptions about human behavior and decision-making that are derived from psychological research. For example, the psychologist and philosopher Herbert Spencer argued that economic behavior was motivated by self-interest, and that individuals would choose the actions that maximized their own pleasure and minimized their own pain. This idea was influential in the development of classical economic theories, which often assumed that individuals were rational and self-interested.

Q: How did the work of psychologists Fechner, Weber, and Wundt influence economists such as Edgeworth, Jennings, and Jevons?

A: Economists such as Edgeworth, Jennings, and Jevons were influenced by the work of German psychologists such as Fechner, Weber, and Wundt. These psychologists were some of the founders of experimental psychology, and they developed a number of influential theories and methods for studying the psychological factors that influence human behavior. For example, Fechner developed the concept of the “just noticeable difference,” which is a measure of the minimum change in a stimulus that an individual can detect. Weber and Wundt, meanwhile, developed theories about the role of sensory experiences in perception and cognition.

These psychological theories and methods were influential in the development of economic models and theories that sought to incorporate insights from psychology. For example, Edgeworth used Fechner’s concept of the just noticeable difference to develop his “diminishing marginal utility” theory, which argued that the utility or satisfaction that an individual derives from a good or service decreases as they consume more of it. Jennings and Jevons, meanwhile, used Weber’s and Wundt’s theories about sensory experience to develop their own theories about how individuals make economic decisions.

  1. How did John Stuart Mill’s psychology affect his economic ideas?

A: John Stuart Mill, the 19th century philosopher and economist, was influenced by psychological ideas in his economic theories. In particular, Mill was interested in understanding the psychological factors that influence human behavior, and he used psychological insights to develop his theories of utility and value. For example, Mill argued that the utility or satisfaction that an individual derives from a good or service is subjective, and that it depends on the individual’s own psychological state. This idea helped to challenge the classical economic theory of value, which held that the value of a good or service was determined by its objective characteristics.

Additionally, Mill’s psychological approach to economics influenced the development of the marginalist revolution in the late 19th century. The marginalist revolution focused on the concept of marginal utility, which is the additional satisfaction that an individual derives from consuming an additional unit of a good or service. This concept was based on the idea that the utility of a good or service is subjective and depends on the individual’s psychological state, and it was a key part of the new economic theories that emerged during this time.

  1. How did William James’s ideas affect economics?

A: William James wrote about the role of emotions in economic decision-making, and argued that emotional factors could influence an individual’s willingness to take risks in economic situations. This idea helped to challenge the assumption, common in classical economic theories, that individuals are purely rational and unemotional in their decision-making.

Skeptical of this response, I asked:

“In which of his writings did William James argue that emotions can influence willingness to take economic risks?

And got the honest answer:

“I’m sorry, but I am not able to browse the internet, so I cannot provide information on specific writings by William James. As a large language model trained by OpenAI, my knowledge is based solely on the text that I have been trained on, and I do not have access to external information.”

I then consulted a James scholar who confirmed my skepticism. In his opinion James had very little influence on mainstream economics.

Nonetheless I am awed.  Apart from the cogency of ChatGPT’s responses, and its ability to parse my questions, where did it find the “training text” that enabled it to answer my obscure questions?

But ever the skeptic, I ask myself whether the system is a form of self-limiting free riding, like stock indexation: It works only to the extent that some human intelligence creates the individual items (good prices for the individual stocks) and shines mainly in synthesizing them.

But what if we all just rely on the mechanized synthesis? As seems to increasingly happen in the stock market.

Or maybe these worries are overblown. Individuals will always have incentives to create the individual originating items.

But what about incentives to game chatGPT, in the same way that search engine optimization has made a mockery of efficient googling, and fake postings flood Amazon reviews?

Also on the issue of incentives: How is this open source enterprise going to fund itself — and come to think of it, how is it currently funding itself?

As my obscure queries show, ChatGPT has somehow been “trained” on a vast trove of obscure, not widely accessible material.

This isn’t cheap. Likewise serving up the answers virtually instantaneously requires expensive hardware and software.  The sweat of unpaid volunteers can’t sustain this.  For now, Microsoft is donating its supercomputing infrastructure to ChatGPT’s maker OpenAI. Sooner or later it will require a commercial return.  Google, which also has the capacity to offer more intelligent responses than its current search engine, will be drawn into the game – but not without monetization.  And if history is any guide the monetization won’t be through a straightforward fee for search.  That would be so old economy.

Get ready folks to give up even more of your privacy.

In the meantime, prepare to get hooked. I sure have…

Tags: