Defining an AI fund is far too valuable to be surrendered to marketing departments. Investment history is littered with ground-breaking good ideas that have become debased by inadequate and, in some cases, dangerous implementation. Absolute return, cash plus, ethical, passive, smart beta all spring to mind. Definition creep should be avoided at all costs.
So what does an AI driven investment strategy look like? How does this differ from what has been done before, and what are conventional investment managers doing today or planning to do in the future? As a starter, AI investment strategies should be able to process diverse inputs, interpret and learn from these, and generate good directional actions. An AI system should operate autonomously, without external human assistance. Learning should be adaptive, rather than programmed, and able to respond to novel situations – not just those that have happened before.
Is it already too late to rescue AI from public debasement? Back in January, the JP Morgan flows team reported that AI funds had, ‘lost an unprecedented 7.3% in February even surpassing the losses by CTAs’. The source information for this comment was the Eurekahedge AI Index, which showed that February was indeed a bad month, the worst since records began in 2011. Was this a fair reflection of AI fund performance? Well perhaps not, as a review of the index constituents a year ago showed that of the 12 funds being monitored only three were close to real AI. An early, but minor skirmish, but it does emphasise the importance of defining what an AI-based fund is, and in doing so establishes early on the classification for a fund universe that, as of today, may be limited but is anticipated to become meaningful in the future.
Meanwhile, there are other important questions for those attempting to work out how AI will shape our futures. Artificial intelligence versus human intelligence is playing out in every aspect of our lives. How will HI respond to AI? Humans, and consequently, markets, are chaotic in the scientific sense of the word and the better for it. We can be very perverse when we choose to be. Secondly, will AI funds lower costs and risk, which are very valid ambitions, or will they improve returns?
When I used to swim a lot I met Dick Jochums, who was one of the top coaches in the US during the 1970s. At that time he had a squad of world record holders and Olympic champions. The secret of his success was to reduce the amount of time his swimmers spent in the water, from a mind-numbing six hours a day, which is what the Australians and Russians were doing, to a more sane three hours. The pay-back was that the intensity bar was raised to maximum. No more plodding up and down the pool simply to put in the miles. It worked. Unfortunately, others concluded that if they could train for six hours a day at maximum intensity then that was even better and so the benefit of his insight had only a short shelf life.
What has this to do with investment? Well, AI definitely has advantages over HI, most obviously in terms of speed of interpretation and action, particularly within closely defined parameters. HI may, however, still have the edge when it comes to breadth, ranging as it does across all facets of life, to judge what is relevant to an investment decision and what can be safely discarded. If AI manages to bring intensity to both breadth and depth, then human fund managers may as well move aside. I am sure that we will find other games to play. Time will tell, but we can be sure that attempts to control innovation, whether by legislating to control social media, banning the use of genetically engineered crops, or avoiding morally questionable research are doomed to failure. Technological progress will ultimately determine the future of investing, not regulation. For now though we will see more hybrid strategies that combine HI and AI, making the measurement of pure AI fund performance ever more important.
It is sometimes said that one of the main problems with AI is that it is amoral. Those that think about where morality comes from consider that, in evolutionary prehistory, consciousness emerged as a side effect of language and that morality is a convenience to be relied on only in normal times. We cannot attain the amorality of wild animals or the choiceless automatism of machines. Understanding the nuances of human language may be hard to program into a computer, but looked at from the other side, surely it is possible that machines will develop a language that we cannot fully understand. Then what happens if we want to question the AI judgements that we have come to rely on? Will we have the language needed to interrogate the system?
Definition, implementation and measurement is step one. There will be many more complicated questions to address in the future.
Investors should remember that the value of investments, and the income from them, can go down as well as up. Investors may not recover what they invest. Past performance is no guarantee of future results.
Any mention of a specific security should not be interpreted as a solicitation to buy or sell a specific security.