When AI reads every 10-K, the only edge left is the CEO's flop sweat

What happens to active (human) investors as AI gets better? Will there even be a role for humans in the process or will we be out of a job?

It’s a smaller version (and more selfish, given I am an investor!) of a bigger fear on AI displacing / replacing knowledge workers. Those fear probably peaked this April when people started resurfacing a video of Anthropic’s CEO saying AI would destroy 50% of knowledge work and lead to 10-20%+ unemployment. I’ve generally thought those fears were overblown, but I will admit that AI is so revolutionary that my belief that we weren’t about to face some mass unemployment event wasn’t quite as steadfast as I’d like it to be.

For investors, the implications of the rise of AI are basically infinite. You can spend endless hours thinking about how AI will impact you individually as an investor (or the process of investing in general), building out tools to streamline or improve your investing process, or looking for opportunities that AI creates (whether it’s shorting AI losers or buying AI beneficiaries)…. and I have spent a lot of time doing just that!

But today’s article I wanted to focus on something slightly different: the outlook for humans in investing.

A lot of investors¹ think of a super-powered AI in investing as some all knowing god that can predict the future and thus sets prices based on things before they even happen (a la Minority Report or some other scifi work). While that’s a cool (and very scary) concept to think about, even super-powered AI cannot predict the future. So the AI bear case for investors isn’t some prophet that knows how stocks will move in advance; the “AI fully displaces humans” bear case would look something like this: at its core, investing is about reading and processing information. An investor generates alpha by either having more information or reading the same amount of information but processing it better. If you imagine a world where a super-powered AI can process all information basically instantaneously and with a super-genius IQ, what hope do human investors have in that world?

That’s not to say that stock prices won’t move or move dramatically! Again, the AI is not predicting the future; there will still be massive stock pops if a company is acquired out of nowhere or stock drops if a company’s plant blows up or something. So new information can and will move stocks, often dramatically²…. but in the AI world I’m laying out you basically have a perfectly efficient market where all known information is already in the price. You could still make (or lose!) a fortune in the stock market by making a big bet on something…. but you’ll make a fortune in the same way you could make a fortune by going to Vegas and betting it all on black. Your fortune came through luck or beta, not through alpha generation.

In that world, I think all of the “analysis” edge goes away. One of the classic stories every investor dreams of having is finding the needle in the proverbial 10-K: some line or number buried deep in an SEC filing that shows a company is undervalued or some event is about to happen. I suspect all of that edge gets competed away (if it hasn’t already!) by AI that is instantaneously reading every SEC filing in real time and comparing it to every SEC filing ever made. Humans simply can’t compete with that!

However, “analysis” edge going away does not mean that all edge goes away. If computers can perfectly analyze all information that they have, that means you can still generate edge through unique data. In fact, that means having unique data might have more alpha.

What does this look like? Well, an extreme example will show this nicely: if your best friend is the CEO of a company and tells you they’re announcing a deal at a huge premium in two days, that is a very unique piece of data. In the perfectly efficient market I laid out above, you will make even more money with that information because a perfectly efficient market means lower trading and liquidity costs and that tail events (like unexpected mergers) generally come with bigger payoffs.

Again, this is an extreme example: what I laid out above is clearly insider trading, so while your bank account might go up on the actual day of the announcement you’ll end up giving all of that up in SEC fines and jail time. But I think it illustrates the opportunity nicely.

So where could we see a (legal) data edge?

I could imagine a few places. Let me start with the simplest, because I think my next ideas build off it: historical stock market data. This might seem basic and easy if you’ve never tried to use historical stock data…. but anyone who has tried is almost certainly nodding their head in agreement at this. Companies get delisted or go bankrupt all the time; how are they handled in the historical data that you’re training your AI on? If the answer is “we’re missing them,” then your AI is going to have a very positively biased view of the world. An example might show this simply: consider a binary like a drug readout and imagine a world where if the drug reads out successfully the stock goes up 10x and if it fails the stock immediately files for bankruptcy and delists. If you train an AI on drug readouts and all of the failures aren’t captured because those companies are gone, then your AI is going to think every drug readout leads to a 10x and buying every drug readout on max margin is a ticket to world beating returns. That’s a recipe for going bankrupt fast! Now, that is a very simplified example…. but you could start getting more and more complicated with it. Is there a world where there’s a cascading data advantage for large firms that have decades of detailed (and accurate!) stock market data to train their AI on?

Of course, historical stock market data is generally publicly available. Perhaps the AI gets so good that it can clean everything up and adjust to those errors I mentioned on its own. So maybe that edge is fleeting or non-existent to begin with.

Perhaps future edge comes from nonpublic data. That kind of makes sense: if the AI can perfectly interpret and price all data, the way to outperform is to have data that no one else has.

What could that look like?

Well, MNPI like the insider trading example I mentioned earlier is one clear example. But let’s assume that you’re trying to avoid jail time and try to think of a few others.

The first that comes to mind is private meetings. If you have a 1x1 with a CEO or some other member of a management team, by its nature that meeting generates data that no one else has. Yes, Reg FD limits just how juicy the data a management meeting can generate, but even small amounts of data could be enough to have an edge in a perfect market, and there are lots of things that are material to an investment thesis that can be revealed in an investor meeting and certainly aren’t part of Reg FD. A somewhat extreme example might show this nicely: imagine that a company has their Q2 earnings call at the end of July and the CEO says “we are confident in our full year guidance” and all of the AIs immediately incorporate that into their models. Then pretend that you have a company visit to see the company in mid-September, and you’re the only one on that visit. After the tour, you ask the CEO about guidance, and the CEO pauses for a second, nervously takes a sip of water, breaks into a flop sweat, and says “I said on the earnings call that we are confident in our full year guidance.” That would be a very juicy piece of information! Feed that into your AI and you’d have a heck of a unique data point.

Now, you might be thinking “but that data point is already unique / actionable in today’s world.” That is certainly true! The point is what happens to the value of that data point in a world where everything else gets priced instantly; I’d suspect it’d be increasingly valuable in the hands of an AI who can use it to tune their model of all sorts of different derivative plays.

The stronger pushback against the “unique data is edge” thesis is probably alt data over the past two decades. A huge amount of money was spent on alt data in the 2010s (credit card data, satellite images of parking lots, etc.) and that edge decayed fast (if you’re the only one with credit card data, you can destroy everyone else when it comes to trading quarterly prints…. if everyone has that data, then you’re all out the cost of the credit card data and no one has an edge!). I get that pushback, but that alt data was generally broadly available if you were willing to pay up. There was nothing unique about credit card data; everyone paid basically the same price and got the same data. The stuff I’m talking about isn’t broadly available; analysts who are good at getting management teams to open up or reveal things that they maybe don’t mean to are creating genuinely unique data. Anything with a price list gets arbitraged away. What’s left is the stuff you have to go get yourself.

And I wonder if the value of getting that unique data goes up in an AI world; maybe knowing the company is likely going to miss guidance causes the AI to come up with all sorts of strange trades that a human could never find (i.e. Apple is going to miss their quarter, so tax revenue in California is going to be lower than expected, so California deficits are going up and we should short Cali bonds while buying Texas munis?).

Again, that’s an extreme example. But you can imagine softer examples that create smaller edges that might be really relevant. Expert calls are the ones that really come to mind here; could they become increasingly important as a tool for generating unique data that feeds the AI’s view of the world in real time?

Maybe I’m overthinking this. Buffett famously said that, in investing, if your IQ is >130 you’d be better off selling the extra points. Maybe the rise of 200+ IQ AI that can instantaneously read and interpret every data point and compare it to all of human history changes nothing for active investors. I sort of doubt that, but who knows. The world is weird!

But I’m increasingly of the view that, in the future, alpha is going to be generated more by people who can find, source, and feed unique data to our AI overlords, and that the major question for most investors is going to be how to do exactly that. Is it being part of a big shop with a huge expert call budget? Is it developing deep sector expertise and industry relationships? Is it remaining small and nimble so you can get unique data across lots of different industries?

I don’t have the answer, but it’s something I’m thinking about all the time (and how I can best position myself for it!).

1. And at my most pessimistic I’d throw myself into this bucket

2. In fact, I think you could argue unexpected news will move stocks more dramatically in a perfectly efficient world