AI Can Play Poker, but It Can’t Play the Markets Yet

Facebook’s AI-powered Pluribus system astounded the world when it recently defeated six professional poker players – including World Poker Tour title record holder Darren Elias in games of no-limit Texas Hold Em. The AI revolution’ is being lauded in pretty much every sector you can imagine from healthcare and agriculture to fast-moving consumer goods and even reading the news.

Worldwide spending on AI systems grew to nearly $35.8 bn this year a 44% increase in 2018. It’s expected to more than double to $79.2 billion by 2022. In my own sector of international capital markets, the chorus of voices proclaiming the dominance of machine-learning is growing by the day.

After retail, the banking and finance sectors are the second biggest beneficiaries of AI investment. Such levels of spending are undoubtedly going to change the make-up of international capital markets and the traders who operate in them. But we must ask ourselves if we are headed in the right direction.

That AI is so popular with venture capitalists now is clear, but what’s not clear is whether it’s working. For every dollar invested in new technology, a dollar is quickly withdrawn from a project that fails to make money. In many cases, investment pours into programmes without a clear sense of purpose. In some cases, it is technology for technology’s sake, rather than addressing a genuine need.

Chasing autonomous trading unicorns misses the bigger opportunity of AI and machine learning. The goal should not be to overtake humans but instead augment their abilities including in the making of trading decisions.

Theoretical neuroscientist and entrepreneur Vivienne Ming is part of a growing cadre of like-minded experts calling for a better understanding of augmented intelligence’ – an alternative conceptualisation of AI focused on how it should assist and advance human capabilities instead of replacing them. In capital markets, that means helping the traders who operate in them become faster, smarter and more efficient at what they already do.

Intelligent machines have big potential for trading in terms of sifting through large quantities of data and other less analytical tasks. While robots are more powerful and efficient than they were decades ago, we have yet to see any conclusive evidence to suggest that robots can consistently interpret trading patterns to make the complex decisions that traders do for their clients.

Algorithms are not yet good at determining causality and analysing the meaning behind patterns in the market – this remains a task best reserved for human insight. Machines that decide when to buy and sell are struggling to keep up with a world where many of the rules have been checked out the window. Uncertainty over President Trump’s trade policies and Brexit are just a couple of examples of macro drivers of the international trading environment which AI is poorly equipped to handle. That these are also a challenge for human traders cannot be ignored, however, I believe humans still tend to do better in response to unexpected outcomes as they are not driven by a straight reliance on past patterns of behaviour.

Data-crunching computers making investments predicated on historical price trends did worse than human managers in 2018. These computers were handling some $220 billion of capital, but the losses got so bad investors were forced to pull billions out. The irony is that these types of investments had been regarded as an effective way of protecting portfolios against downside risks in previous years when they were managed by humans.

One also has to ask when things go wrong in AI investing, is the machine legally responsible? I have argued they should not be, but rather it should be the individuals and firms responsible for the development of the AI who should be held to account.

Unlike computer algorithms, human traders with expertise and insights can respond to inefficiencies in the market, shock news or unusual activity. The marriage of human judgment with cutting-edge technology enables traders to provide valuable liquidity in more thinly traded stocks which are generally outside the focus of AI-driven strategies. This is essential in providing investors with more options and opportunities, while also encouraging capital formation.

For decades now, a large swathe of companies that operate in the capital markets have attempted to replace human judgment with automated systems. They have also largely ignored stocks in areas of more limited liquidity, smaller capitalisations and emerging growth companies. If we want more competitive and stable global markets, more firms should embrace augmented intelligence to support these more thinly traded types of stocks.

Workers around the world fear that AI and automation will make their jobs redundant and leave them out of work. But the potential of augmented intelligence can be applied to a plethora of sectors beyond capital markets and finance, and I am optimistic that the most likely outcome is a reimagination and reframing of work.

AI and automation can with the right public policy and private sector commitments give humans across all sectors more time and capacity to focus on being more creative and insightful in the tasks they do well, and ones which robots are often ill-equipped to handle. Augmented intelligence is the best way forward for the capital markets industry and the global economy.