How investors should place bets on AI

{"text":[[{"start":null,"text":"

"}],[{"start":6.42,"text":"The writer is a former global head of research at Morgan Stanley and former group head of research, data and analytics at UBS"}],[{"start":15.86,"text":"Annoying, amusing or indispensable. Every investor will have a sense of how far artificial intelligence has progressed in this sequence. But a hunch won’t get us far. Instead, understanding the kind of future we expect should define our bets. It is not the same to wager on a tidal wave of productivity benefiting most companies as it is to bet on a grinding development favouring a few incumbents. Or a rupture to a new paradigm, bringing the next superstar firms."}],[{"start":52.36,"text":"Perhaps a good place to start is the most widely accepted assumption — that AI will boost productivity. We may not know who the winners and the losers will be, but we could surmise that most industries will profit from AI as the technology is infused within companies as happened with steam engines and electricity in the past."}],[{"start":75.76,"text":"If we accept this proposition that AI is a normal technology — advanced by Princeton University Professors Arvind Narayanan and Sayash Kapoor along with others — the right approach could be to buy broad market indices as the rising tide will lift most companies."}],[{"start":95.74,"text":"But investors should consider some nuances. Estimates of productivity gains are all over the place. For the US, estimates of productivity increases range from at least 1.4 per cent growth in a year (a lot) to 1 per cent over the course of five years (not much)."}],[{"start":115.56,"text":"This is understandable. We do not really know how AI changes the productivity equation once we account for the costs of adoption, adequate oversight and ensuring the right specification of the models. Often at times, humans are even slowed down when they work with AI given the current state of the technology."}],[{"start":137.16,"text":"Some investors might bet that whoever makes the most of Jevons’ paradox will capture the lion share of the economics. Stanley Jevons, a Victorian economist, thought that when technology advances make a resource efficient to use, instead of reducing the expense, the demand will grow much more, leading to higher overall consumption."}],[{"start":160.84,"text":"If the paradox is true to form, the best profits will be captured by firms that help corporations or individuals integrate a myriad of agents and applications in their enterprises. The game, then, is to forecast who out of the current AI hyperscalers can capture the “attention” of clients in more ways and forms. That is the equivalent of spread betting in sports, comparing predictions of relative gains in share against what the market prices."}],[{"start":193,"text":"Another investor though may bet on finding the new multi billion-dollar company produced by the next big tech leap. That is a riskier wager because tech paradigms are not stable. Lots of new technological solutions will emerge and gradually, but then suddenly, the market will converge on one or a few. This is the story of Google’s search engine or Nvidia’s chips."}],[{"start":219.84,"text":"As shown by past cycles of innovation, very few firms stay at the frontier of change over five years. No one knows which companies will emerge winners in the next convergence. Or whether large language models with open-source software coding or closed versions will lead the way."}],[{"start":240.26,"text":"The guessing game would be simpler if testing the performance of LLM’s performance would discover the next best thing. But it is not. Professors Narayanan and Kapoor make a convincing case that rather than tracking how many trillion parameters are in a model, or its ability to pass the long list of difficult questions in the AI performance benchmark known as Humanity’s Last Exam, we must assess the practical uses — the applications that make our job more productive. The size, compute time, and accuracy of the answers by a model seem no different than the details of a candidate’s education in the résumé. It tells you something, but it does not guarantee they can do the job. Instead, we must monitor the development of AI applications and agents."}],[{"start":292.46,"text":"Brave investors could look at early-stage AI companies as an option on a speculative future with very idiosyncratic risk and reward characteristics. As a study by professors Lubos Pastor and Pietro Veronesi of the University of Chicago shows, peaks in the valuations of shares of companies with emerging technology are associated with high uncertainty about the productivity of the innovation. Valuations multiples then come down as the technology becomes adopted at scale."}],[{"start":328.22,"text":"So caveat emptor; unless the player has some clear edge, it may be better to spread the bets, accepting the radical uncertainty that they entail."}],[{"start":342.02,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1759138191_4595.mp3"}