It has been estimated that artificial intelligence (AI) could add $15tn a year to the world economy by 20301 – which is more than the current output of China
and India combined. Indeed the rise of “thinking” machines has been described as the “Fourth Industrial Revolution”2 (the first three being the loom, the steam engine and the assembly line) with the implication that it will change our economy and our society just as much. Given this enormous potential, investment markets are unlikely to be unaffected, so what do investors need to be aware of, and what opportunities and threats should they be looking out for?
Let me start by defining three different levels of AI. Today what is usually referred to as AI, is a “narrow AI” that teaches itself by applying neural networks to big data. Although this can deliver impressive results for say diagnosing cancer or for self-driving cars, it is arguably a complicated algorithm rather than a true intelligence. The next level of AI is a “strong AI” that can attempt the full range of human cognitive abilities, is genuinely intelligent and probably has some level of self-awareness and independent choice. The final level is known as the “Singularity” – the hypothesis that once an AI is intelligent enough to design another AI that is yet more intelligent than itself, humanity will lose control of this evolution and AI will rapidly undergo an exponential acceleration to levels of super-intelligence that we cannot comprehend, any more than an ant can understand a human being.
I don’t want to discuss or even contemplate the tail risk that a Singularity takes over the world and destroys or enslaves humanity (and all other AIs). This is not because I think this scenario is impossible, but it has been extensively covered in science fiction from the Terminator to the Matrix, and it’s not very interesting from an investment perspective (adopt the standard end-of civilisation portfolio of farmland, firearms and
gold – until the robots get you). So instead let’s think about a scenario where multiple strong AIs are operating within the current financial world, and that humans are still in charge (or at least are allowed to think that we are) of laws, regulations and central banks, and can create new strong AIs to help them
as they wish. What might happen to market
I would argue that intelligence has always been the subject of an arms race in the world of finance. Markets can offer financial advantages to those with hidden information, deeper understanding or more complete data, and hence significant resources are deployed in the hope of advantage, only to be eroded as the competition catches up. It is telling that the phrase “computer” originated in the early 17th century, long before electronic computers existed, as the name for people who performed and checked calculations
in science or finance. There was even a hydraulic computer named MONIC built in 1949 to model the UK economy. Modern finance mostly operates as teams of intelligent independently-minded people supported by cutting edge computers (which are still mostly below even narrow AI) and data.
So for many purposes the machines have already arrived in the markets, but they are operating in teams with people and do not have sufficient competitive advantage to let them dominate the markets. However the rise of AI still has some implications that investors
Firstly, as with any revolution many businesses will be disrupted in the same way that the advent of the motor car wiped out many thriving businesses relating to horses, or that Amazon is weakening many high street retailers. But predicting specific winners is hard and there is a tendency for markets to buy into an interesting story leading to a market bubble (such as the railway bubble or the tech bubble). So make sure you are well diversified and avoid overpaying for the latest hot tech name.
Next, innovative investment channels will develop such as initial coin offerings as a form of crowdfunded debt investment or equity markets managed via blockchain rather than an exchange. Recognise that regulatory protections in such markets will be limited and treat them like other frontier markets.
Investors need to recognise that their personal data has a value, especially when combined with data of other investors. Such data can provide asset managers with more timely and a deeper understanding of the economy and the behaviour of other investors. Therefore expect that this will give rise to new investment strategies, but that many of these will be undermined as the value of the data is recognised and privacy and ownership rights are asserted.
Investors should be sceptical of any claims that a particular investment strategy is better because it is run by AI. Make sure you understand enough about what is going on inside the black box or you are in danger of buying into back-testing and survivorship bias.
We can also expect ethical challenges to arise as AI discovers old forms of market manipulation in new guises. But remember that the market regulators will set their own AIs to catch misbehaving machines, and regulation will develop. Never invest in anything that sounds too good to be true.
So in summary we expect that the pace of change will continue to accelerate and there will be many disruptions arising. However, unlike in the world of competitive chess playing, the psychological underpinning of the markets means AI will not be able to dominate sufficiently to crowd out human traders, at least not those that are backed up by sophisticated sub-AI computers.
How might artificial electronic intelligence have any competitive advantages over humans supported by conventionally programmed computers?
Speed: Trading ahead of the crowd will often deliver an advantage. But investors have responded to this with so called “high-frequency trading”, where proprietary trading strategies are executed by sophisticated computer programs situated close to an exchange. If speed is the sole requirement for success, then a true AI will tend to lose out, as the need to process its intelligence will mean it is slower than a more focussed program.
Does not get tired or make mistakes: Mistakes generally cost money, but computers (properly programmed) do not make mistakes either and human teams with high quality controls can effectively banish mistakes. Additionally in real world markets the optimal rules are not known and there is no advantage in accurately following the wrong rules. Imagine that a theoretically perfect trading rule did exist, it would rapidly get copied and its trades would be crowded out until the opportunity for profits disappeared. Avoiding mistakes is therefore not a long-lasting source of advantage.
Perfect memory and able to use more data: Conventional systems also provide perfect recall. Additionally given the ability to analyse more data it becomes progressively harder to separate the signal from the noise, and this increases the chance of relying
on spurious or incorrect data (garbage in – garbage out).
Free from heuristics: Heuristics are “rules of thumb” that people use to simplify their thinking. These can sometimes go wrong when clear thinking is needed, so a well programmed computer can help avoid them. But it is now clear that big data techniques to develop narrow AI can lead to uncritical adoption of such heuristics from the test data. It is likely that strong AI will discover new and untested heuristics, but it may be very difficult for people to be aware of these or understand the risks arising.
Different ways of thinking: This could take the benefits from diversity into entirely new realms if ever adopted – but so far the use of narrow AI in robo-advisors and trading algorithms seem to mainly be variations on thinking and techniques that already existed. Strong AI might be more likely to discover entirely new ways of thinking, but how will we then know whether we can trust it to trade or invest our money?
Cheap To the extent that intelligence becomes an “always-on” commodity that can be accessed online whenever required at low costs, this should be easier and cheaper than hiring people to think. However the set up costs to develop the necessary computer infrastructure, experienced artificial intelligence and a relevant data set are not trivial.
1: Professor Al-Khalili, Surrey university according to the Financial Times on 6th September 2018
2: Frank Roehrig and Pring, What To Do When Machines Do Everything (2017)
For more information contact firstname.lastname@example.org