LONDON, May 19 (Reuters Breakingviews) – “My colleagues, they are working on artificial intelligence,” Israeli psychologist Amos Tversky once quipped. “I study innate stupidity.” The co-founder of behavioral economics, who died in 1996, won’t live until 2023. At that time, many of his academic peers jumped on the AI bandwagon, along with venture capitalists, corporate leaders and equity investors. But investors should pay more attention to Mr. Tversky’s expertise. Behavioral economics, which studies how psychological, emotional, and social factors influence human decision-making, has some important pointers for those who want to profit from AI. .
The first lesson is the most obvious. Watch out for bubbles. Since OpenAI released his ChatGPT chatbot last November, the steady flow of capital into all things AI has turned into a torrent. Shares of Nvidia (NVDA.O), the world’s leading maker of chips used to create AI, have surged more than 100% in the past six months. Software giant Microsoft (MSFT.O) has increased its market capitalization by nearly $500 billion since it announced in February that it would incorporate AI into its Bing search engine. Alphabet (GOOGL.O) investors became Google owners just one day last week after CEO Sundar Pichai unveiled a new AI product at the company’s annual I/O conference. Increased value by $60 billion.
In fact, AI enthusiasm is a ray of light breaking through the stock market dark clouds caused by the record rise in U.S. interest rates. SocGen analyst Manish Kabra estimated last week that the S&P 500 index (.SPX) would be down 2% this year, excluding AI-related gains. Instead, it rose by 8%. Booms also have macroeconomic implications. Ireland’s Chancellor of Finance Michael McGrath last week announced a new €90 billion sovereign wealth fund largely funded by corporate tax windfalls from tech giants such as Ireland-based Apple (AAPL.O) and Microsoft. announced plans for the fund.
For other companies, perceived vulnerabilities to AI could spell disaster. Shares of Chegg (CHGG.N) plunged earlier this month after the maker of learning materials admitted that so-called large-scale language models such as ChatGPT were encroaching on the company’s market.
Orthodox asset pricing models suggest that these volatility reflect reasonable, albeit variable, estimates of future profitability. However, behavioral economics has long offered an alternative explanation by rogue enumeration of systemic deficiencies in human decision-making. These range from crowds and overconfidence to confirmation bias and fear of missing opportunities. For investors, now is a good time to pay particular attention to the natural tendency of stupidity that pushes stock market valuations to unrealistic and ultimately unprofitable extremes.
But the most important lesson of behavioral economics relates to a more fundamental question: Will the new generation of AI do what it promises? This technology has already achieved some very impressive achievements. In November 2020, Google DeepMind’s AlphaFold stunned the scientific community by achieving major changes in one of his grand challenges in molecular biology. It predicted the structure into which a protein would “fold” based solely on the sequence of its constituent amino acids. Nobel Prize winner and then-President of the Royal Society Venki Ramakrishnan called it a development that would “fundamentally change biological research”.
AlphaFold has demonstrated what is widely understood to be AI’s greatest strength: its ability to recognize patterns that evade both human intuition and traditional statistical analysis, and leverage these patterns for predictive purposes. The same ability characterized the AI’s astonishing achievements in defeating human opponents in strategic games like chess and Go, and enabled ChatGPT to generate eerily coherent prose.
The big unknown is whether AI can replicate this extraordinary predictive power in the commercial, financial and political spheres where the rules are more ambiguous. Behavioral economics offers some caveats against such attempts to apply AI in practice.
One potential gremlin is the so-called sampling bias problem when building predictive models based on statistical learning. The problem is that the dataset used to train the model may omit rare but significant events. For example, stock market earnings can be affected by rare but extreme stock price fluctuations. As a result, quantitative trading firms often eschew pure data mining strategies in favor of an approach that assumes rather than learns so-called tail risk probabilities. Less technically minded investors employ their own versions of the same tactics when implementing simple heuristics such as legendary investor Benjamin Graham’s “safety margin”.
Behavioral economists have described the problem of sampling bias when studying how humans learn. However, neural networks can have similar drawbacks. Intelligent machines, like humans who are naturally stupid, will have to face the infuriating fact that the absence of evidence is seldom proof of absence.
Then, perhaps the most frustrating of all problems when it comes to modeling and manipulating human behavior is Goodheart’s Law. This is a paradox first clarified in 1975 by Charles Goodhart, an official at the Bank of England, that when a measure becomes a policy target, it becomes unreliable. For example, total money used to be a good predictor of inflation. But once central banks adopted targets based on those numbers, the stable correlation disappeared.
The root of this problem is that human systems, unlike physical systems, are inherently adaptive. When people feel that anticipating their behavior is against their interests, they understand it and try to defeat the effort. Amino acids involved in protein folding are not.
Again, these real-world challenges are well documented in the investment space. Stock trading is a zero-sum game. One investor’s capital gain becomes another investor’s capital loss. As a result, there is a strong and automatic incentive in the rest of the market to adapt and deactivate historically successful trading rules as soon as they are identified. Goodhart’s law explains why the excess returns earned by systematic investment strategies usually decline over time. Whether AI can successfully escape gravity is still an open question.
Investors would be hesitant to ignore the amazing work AI has produced so far. But when it comes to broader applications, it needs to be handled with caution. Artificial intelligence may have more in common with natural stupidity than humans and machines currently think.
follow @felixmwmartin on Twitter
Editing: Peter Thal Larsen and Pranav Kiran
Our standards: Thomson Reuters Trust Principles.
Opinions expressed are those of the author. They do not reflect the views of Reuters News as committed to honesty, independence and freedom from bias based on trust principles.
