I don’t know if you noticed a small news item in some of the newspapers sometime last week. It described the experience of an investor with ChatGPT – the AI that is making all the waves in the media currently.

The headline of the news item said “ChatGPT beats top investment funds in stock-picking experiment”.

ChatGPT was asked to select stocks and create a portfolio (of 38 stocks) using criteria, such as history of growth of the company and low debt levels etc. Over a period of 7 weeks the value of this bouquet of stocks on the London Stock Exchange was tracked and compared against 10 of the top mutual funds available in the UK. It was found that over these 7 weeks the ChatGPT bouquet outperformed all of those 10 funds by a wide margin!

Based on these results, it was declared that “This demonstrates the potential of AI-powered chatbots like ChatGPT to help investors make informed investment decisions”!

I did a quick calculation which showed that instead of the period used in the experiment 6th March to 28th April, if I used a different period – say from 17th March to 24th of April, a simple, passive index fund based on FTSE-100 index (the most popular stock market index in the UK) would have delivered almost twice as much return as the ChatGPT bouquet!

In other words, if you enter and exit at precisely chosen moments, there is no limit to how much money you can make in the short term, regardless of what stocks you choose! The point of this calculation was – whether an experiment that ran for just about 7 weeks was enough to conclude that “chatbots like ChatGPT can help investors make informed investment decisions”.

Would the ChatGPT bouquet continue to outperform the market by the same margin observed in the 7-week experiment for the next one year, 3 years, 5 years?

If a human fund manager was asked to pick stocks based on the same criteria which were given to ChatGPT, would that fund manager’s bouquet of stocks perform worse than the ChatGPT bouquet?

The comparison in the experiment does not seem to have considered whether the stock selection criteria of the 10 mutual funds which ChatGPT bouquet beat were the same or at least comparable! Without this, the comparison is an apples to oranges comparison. At least the news report did not say so.

I think these are the kinds of questions (and there are many more I can think of) which need to be legitimately asked before blithely concluding that “This demonstrates the potential of AI-powered chatbots like ChatGPT to help investors make informed investment decisions”.

However, even this is not the point of this writeup.
It is entirely possible that ChatGPT’s stock picks are actually better than any, in the short term. However, the “Act of Investment” implies (or it should imply) something more than just short-term stock picks alone.

For example, we have all heard exactly the same or very similar stories before about Robo-advisors. 7-8 years ago, Robo-Advisors were making waves in the market. Similar claims were made about their superior capabilities in picking stocks or other assets for investments, to deliver better returns (compared to human professionals). At that time, it was claimed that this superior ability derives from science (Like Modern Portfolio Theory) which they rigorously employ to compute their decisions (which a human professional is incapable of).

Today, if you check, you will find that in the last 7-8 years only a small fraction of all the money that is invested in the market has been invested purely (i.e., without any human intervention) through Robo-advisors. Again, this is not because Robo-advisors were demonstrably inferior to human professionals in picking the best mix of assets to invest in. It was more because Robo-advisors could not take into account in their calculations the investor’s wants and needs or greed and fears. There is no demonstration, yet that ChatGPT will do any better in this regard i.e., understanding the investor’s wants and needs or greed and fears and incorporating them in its investment advice! In the experiment that I described above, for example, the selection criteria did NOT include any factors related to the investor’s wants and needs.

What if the investor’s needs are better served by investing in assets that offer a greater degree of capital protection and predictable returns rather than the highest possible returns? What if the investor’s aspirations (given his personal circumstances) are better served by taking longer term but riskier bets? What if the investor is prone to knee-jerk reactions to even small turbulence in the financial markets and therefore in his own interest, needs to be counseled against taking impetuous, irrational decisions?

Perhaps it is possible to “train” ChatGPT or similar AIs to consider all this and tailor its recommendations accordingly.

But (a) As yet there is no evidence that has been presented to even hint at that possibility. The 7-week experiment described above is totally inadequate to provide any such evidence.

But even more importantly (b) an honest financial professional is “motivated” to “care” and “be concerned” about safeguarding the investor’s interests purely because he/she prospers if the investor prospers!

Would ChatGPT be ever driven by a “concern” to safeguard your interests?