Buried in the sprawling, nightmarishly complex and totally pointless landslide of petty interference that is Mifid2 – there is a very interesting provision. I know, right – I was as surprised as you! Sitting in yet another interminable briefing on the topic, drawing spirals on my notepad, I suddenly heard this intriguing provision. From next year, Algorithms involved in electronic market making must carry a unique identifier – and the trades they do will be reported to the regulator with that identifier attached. This to me is absolutely momentous. For years, regulators and banks have effectively conspired to create a system of individual monitoring – individual bank employees now assume a huge amount of personal risk as they are responsible for upholding countless daft regulations, and what might be market practice one day is considered illegal the next. But this new provision effectively extends that kind of supervision of individuals to include individual algorithms. The next flash crash, at least in Europe where Mifid2 applied, will be analysed on the level of which individual bundles of code were trading, and presumably why. To me – the implication is clear. Algorithms are now being treated as participants in markets – not just tools that banks use.
When people talk about the “Singularity” in technology, they typically mean that some kind of AI will become intelligent enough to start creating and propagating other, even better AI’s, which is presumed to lead to some kind of either glorious utopia or terrifying robot world domination. But there’s a very crucial flaw in this scenario – namely the idea of “intelligence”. It’s very easy as human beings to go around using this concept, because we have a shared experience rooted in our biology of wanting certain outcomes, and so intelligence is easily understood in terms of how we overcome obstacles to get those outcomes. For machines, there is no such goal/obstacle relationship. Of course, one can program an algorithm to contain a utility function, something to maximise or minimise depending on a set of constraints. It’s possible to imagine that at a sufficient level of complexity, these utility functions and algorithms maximising behaviour might lead to something that looks a bit like human intelligence. But that’s hardly the point. The algorithms already have a perfectly serviceable provider of obstacles and goals – us!
Seen like this, humans have largely been working in the service of algorithms for a lot longer than computers have been around. After all, a publicly listed company is a kind of algorithm – with a utility function (profits) to maximise, and a set of constraints imposed by law, physics etc. The humans who work for this algorithm provide it with a kind of mapping – interpreting the world into inputs that the algorithm understands. The humans here are giving the algorithm its goals, but the algorithm is providing the muscle to achieve them – sometimes literally in the case of a big company, sometimes computationally in the case of a modern trading algorithm – but the relationship between us and our technology is more complex now than that of tools. It is a two way street. Once a company or a trading algorithm starts, it becomes a part of creating the world that we live in. No human being ever decided that we should spend the majority of our time in offices, but the logic of the machines we created interact with our understanding of the world to make that decision for us. Computers like to be in air conditioned buildings. They like lots of humans in the same place to maximise network effects, and specialisation. In financial markets, they don’t like heterogeneity, and have no regard to the actual needs of investors. That’s why we have triple leveraged ETF’s on broad stock market indices, but no nominal GDP futures – the machines are configured to maximise profit, not fulfil the proper function of markets of allowing firms and households to reduce risk plan for the future.
In this context, the Mifid 2 insistence that Algorithms be identified, rather than firms, is quite radical stuff. But the substitution of firms for algorithms gives us a clue for how the AI alarmists could be very wrong. Ultimately, firms have nothing driving them but the goals set for them by people. Granted its a two way street where people’s thinking can be driven by the conditions set by the firm they work for, but the willpower comes only from the people themselves. No matter how smart AI gets, and how much computational might it can muster, the involvement of human beings, with our single minded focus on serving our own best interests, means that there’s no reason to expect machines to take over from us in the realm of setting goals in some sort of sudden revolution. Like public companies have affected the material conditions of the world, ever more advanced computers are doing the same – but its a gradual crawl as humans get used to being the thinking appendage of the systems we’ve built. Rather than a singularity, and an AI taking over the world, the machinery we’ve built is gradually assimilating us – as we become ever more inseparable.