Silicon Vs Steak: How the AI Apocalypse Skipped Wall Street

In 2021, at the height of the ESG craze, there was a UK based company called Drax Group that decided it wanted to produce “green energy”. Their methods were innovative: stop burning coal, clear-cut forest in Canada, convert the timber to wood pellets, ship them to the UK, and burn them in the formerly coal-fired power plant (yes this is true). It takes far less than the mathematical abilities of your average toddler or even an impressive ape to figure out the practice produced far more CO2 emissions than simply burning coal. Yet Drax Group received billions of pounds of subsidies. It is, in a field of strong contenders, one of the more egregious examples of greenwashing: the environmental equivalent of putting lipstick on a pig. What I think we are seeing today is AI washing.

A brief history of almost nothing

Algorithmic trading has been around since the 1980s. But it wasn’t until the heady days of 2013 that it came to the fore. It was a scary time: the age of the silicon-based analyst was upon us, and the meat-based analyst was advised to cultivate a broader range of interests. This claimed a few scalps, but not as many as feared.

Then we had machine learning, the second assault on the not-so-diminished ranks of the meat-based analyst. Machine learning investors found out the hard way that if you discover a source of “edge” then throw a few hundred billion dollars at it, it tends to stop performing. It revealed the fundamental intellectual dishonesty of these models: their expectation of profitability is based on the continuation of a history that by their very existence they ensure will never happen again.

We’re two assaults in, and at this point, it seems the only things that will survive the nuclear apocalypse and the resulting Mad Max-style dystopia will be giant, irradiated cockroaches and meat-based analysts searching for arbitrage opportunities between Gastown and The Bullet Farm. I’m not even sure Immortan Joe (Google him) would be the scariest portfolio manager I’ve ever known.

Anyway, now comes the age of AI, we find ourselves somewhere in the third wave of silicon v meat. And at this point I’m hoping for a fourth act in which something actually happens. Because so far, a series of increasingly complicated IF(THEN) statements masquerading as sentience portending the doom of human investors has delivered, well, not much doom. While they have made some substantial changes, mostly by helping human investors do their jobs in new and interesting ways, the seismic shifts they promised have not come to pass.

The Wizard of Oz

Let’s be honest about what LLMs actually are: stochastic token generators based on 2.7 billion web pages, every book ever written, and all the videos on YouTube. AI has achieved something remarkable, in that it may be the first entity since Erasmus to have read every book in existence. The intention is that by running the “sum total of human knowledge” through a lot of computer chips, and performing a tremendous number of calculations, we can for a given input, guess with increasing accuracy what the next word or sentence might be… and ultimately create something artificially sentient.

The problem here is that the map is not the territory. By that I mean that you’ll struggle to reverse engineer true thought from pure text (particularly if you source the majority of it from the world’s largest unfenced insane asylum: Reddit). In real terms what that means is this post won’t capture everything that went into it; it’s a map of my thinking, it’s not my actual thinking. How much of what I thought about actually ends up on the page or whatever glowing rectangle you read this on? The point being: could you reverse engineer all the thoughts that went into this post by simply reading it? The same can be said for investment strategies: can you reverse engineer my thinking simply by examining my returns? Could you do it even if I gave you all my write-ups?

AI can’t actually think; it just regurgitates the old using ever more sophisticated IF(THEN) statements. While it does that it will be a tool used by investors, not a replacement for them.

Back to the laundry

Now we come back to AI washing. I’m deeply suspicious of people touting the use of artificial intelligence in investment management. It’s not just what AI’s advocates say, it how they say it. While pandering, snake-oil-salesman pulpit-speak is a mode of discourse I find endlessly entertaining, in this case I think it’s at least dishonest and often flat-out dangerous. “This chat bot selected a portfolio that outperformed by 160%”. I hate to rain on your parade mate, but compared to what? If you selected 100 50-stock portfolios at random from the S&P500, about half of them will outperform, some of them substantially. You might have been fooled by randomness. Get back to me when your chat bot has delivered statistically significant alpha for 5 years.

You may have read these passages and thought to yourself “James hates AI” or “he’s some kind of Luddite”. I can see how you’d get there, but the truth is, I am thoroughly pro-AI. But, I am even more thoroughly pro-reality. AI will probably make substantial changes to the way we invest, in my case it already has. But rise of the machines? I think we’re a way off.

The obvious follow-up question is: even with its limitations, how can AI be used in its present form to help pick stocks? The short answer is: it is an incredibly powerful tool, when used correctly and within its limitations. For the long answer, check back here next week.

Subscribe