Artificial intelligence is beginning to change the world, will it make it impossible for private investors to beat the markets?
The human touch! Can mere mortals compete in the investment arena with AI?
Actually, AI is nothing without data. AI needs data, and lots and lots of it. These days, when people talk about AI, they are usually referring to machine learning — a method by which machines are able to learn how to perform certain tasks, by initially studying data.
This is the means by which AI is developing image recognition, for example. Take the way we identify a raccoon. Young children can manage this feat, but up until recently machines found this very difficult. This largely relates to differences in the way we learn. Try to design an algorithm that can identify a raccoon, a four legged animal with a striped tale and markings around the face, and it is a hopeless task. We don’t learn that way: the human brain learns and comprehends in complex and subtle ways. On the other hand, teach a machine to identify a raccoon by getting it to study thousands of images and it can achieve this task. We don’t know how specifically, merely that it does.
AI also pulls upon other technologies such as neural networks — computers which are meant to broadly simulate the brain, with its neurons forming synapses. But a key development making this possible has been the cloud. To develop a neural network in-house is enormously expensive, accessing one via the cloud becomes a pay as you go model.
To put all this in context, when IBM’s Deep Blue defeated Gary Kasparov at a game of chess in 1996 and over a series of games in 1997, the IBM computer won by brute force: it could quite simply look more moves ahead, and never made a mistake. The AlphaGo machine, developed by Alphabet/Google subsidiary, British company DeepMind, was able to defeat the world champion at the Chinese game of Go, using a quite different approach. Initial versions of the machine studied data on thousands of Go competitions involving both amateurs and professionals. In October 2017, however, DeepMind revealed AlphaGo Zero, a machine/AI program that learnt the game solely by playing itself. And it was able to do this incredibly quickly. Three days after the machine had been playing against itself, it could defeat Alpha Go, 37 days later, claims DeepMind, it was arguably the greatest player of Go, ever.
But AI has limitations. It requires context. Recently, I was at a conference when we showed how AI that is normally very good at image recognition — almost spookily so — fails when images are out of context. For example, it described an image showing a number of goats up a tree as birds on a tree.
But can machine learning be applied to investing? The short answer is yes and indeed it is.
At first, investment strategies were applied using algorithms. High frequency trades are carried out by algorithms that can buy and sell stocks following certain rules in nanoseconds. Such algos can backfire, and have been accused of causing flash crashes when as asset can either plummet or surge in value in seconds, as different algos Independently choose to either buy or sell at the same time.
I am not sure you can describe such algos as AI — rather they consist of a set of rules — if certain conditions are met: sell, or buy.
AI using machine learning, by contrast, would work out when to buy or sell based on its own examination of data.
And the more data there is, the more sophisticated AI becomes. Machine learning can be highly effective if it can have access to huge volumes of data: including not only historical stock performance, but company results, announcements, economic data, customer surveys, or even reports on the weather.
Indeed machines may discern pattens and relationships which have eluded human analysts.
Thinking the same
I wonder whether an AI dominated market place will end up seeing each fund that employs AI pretty much making the same returns as all others in any given sector. If all AI powered trading strategies have access to the same data, then they might all buy and sell simultaneously.
The premium will go to any fund manager that has access to unique data.
Bloomberg recently explored how AI could be used to create new ETFs. A company called Kensho, for example is trailblazing in this arena. “Why would you ever limit yourself to aged financial data when there’s a sea of information out there,” said John van Moyland, managing director of S&P Kensho Indices, in an interview with Bloomberg.
The human touch
Does that mean that for us poor old humans, we will soon find it impossible to compete?
I don’t think it does, and I have more than one reason for saying this.
Firstly, as Gary Kasparov says, humans working with machines can outcompete a machine working on its own. Kasparov says that the winning strategy is machine plus human.
Secondly, I do think that AI determines the bulk of market movements you will see a massive number of buying and selling occurring simultaneously. I wonder whether the AI will cancel each other out: hardly any variance away from average. In such an environment a human trader, by being less predictable, may have an opportunity to greatly outperform the market, although equally could underperform to the same extent.
But for me, there is a third reason. I am a great believer in buying what you know. I have often said I can predict which way the M&S share price is going by watching my wife’s reaction every time she goes into the store. If she walks away empty handed, it is time to sell shares, if she walks away saying she was spoilt for choice, buy.
On a similar vein, I became an Apple bull the day I fell in love with my iPhone. That moment I realised that the Apple share price was on its way up.
Maybe, in time, via customer surveys, machine learning will master what I might call the human touch, too.
But I doubt that this date has yet arrived.
These views are those of the author alone and do not necessarily reflect the view of The Share Centre, its officers and employees