Automated trading systems are widely used in the asset management industry but at present even the most complex algorithms rely on relatively basic machine learning technology. This may be about to change.
“AI is going to impact the economy across a wide range of sectors,” said Professor Nick Bostrom, founding director of Oxford University’s Future of Humanity Institute, who was speaking last week at the Jersey Finance funds conference in London.
Huge sums of money are being pumped into artificial intelligence (AI) research and investment in the technology is set to hit $72bn by 2021, according to the International Data Corporation.
Bostrom, one of the world’s leading thinkers on the implications of artificial intelligence, said development in the field of deep reinforced learning systems over the last decade has been rapid.
Deep learning systems can now find increasingly intricate patterns in ever-growing amounts of data. AI systems can, for example, now construct photo-realistic profile images of people that have never existed. In the not-too-distant future, they will be able to read and digest one million pages per second about a particular subject.
The extent to which this potentially revolutionary technology impacts the asset management industry, Bostrom said, may depend on what kind of investments people want to make and the kinds of sectors they want to invest in.
Machine learning systems are increasingly adept at finding small signals within enormous data sets, allowing firms to shun the conventional rationale that only humans can make big-picture decisions based on hard data. In time, AI systems could make apparently counterintuitive predictions beyond the capabilities of the human mind.
Tech giants dominate
The AI development space is dominated by cash-rich technology giants such as Facebook, Amazon, Google and Microsoft in the United States, and Baidu, Alibaba and Tencent in China.
“The FANGs – Facebook, Amazon, Netflix and Google – are estimated to invest more than $5bn billion annually in AI,” said Nick Hartley, co-head of active equities at Legal & General Investment Management, which has more than $1.2trn of assets under management.
Amazon and Microsoft are seeking to commercialise their AI capabilities by renting them out via the cloud, potentially opening AI up to companies of every size.
“Everyone wants to claim they are doing something AI-related – it has become the fashionable thing,” Bostrom said.
“A lot of start-ups in the sector are developing off-the-shelf machine learning tools and there is opportunity there. But a lot of these tools are not specific to fund managers.”
Hiring top AI talent
The key question, in Bostrom’s opinion, is: How can asset managers invest in AI and achieve some sort of sustainable and defensible advantage?
“If you can get the most qualified people then you will have a commercial advantage. However, in terms of software, there is a lot of competition for the cutting-edge talent in the field of machine learning research,” he said, adding that tech giants like Google and Facebook offer handsome salaries to attract the best talent.
Google, for example, has invested heavily in a raft of AI-related areas where there is no immediate focus on product application in a bid to stay ahead of the development curve. It is aggressively pushing ahead with its deep learning project Google Brain.
The tech giants offer a place where machine learning experts can work in their chosen field for an attractive salary and publish their findings among their peers.
“If people can work for a great salary at Google or Facebook, it’s going to take a lot to persuade these people to withdraw from their peer community and work in a place where you are not allowed to publish your findings. The best people benefit from being able to publish,” Bostrom continued.
Access to big data
An organisation’s ability to develop an AI capability with a sustainable and defensible advantage may, therefore, depend on the size and quality of the datasets it is able to access, Bostrom said.
“If you are sitting on a huge dataset – especially a valuable dataset – that other people cannot easily replicate, such as a massive statistics unit, then you may be able to do more things with it.
“There may be specific data that would facilitate the work of a fund manager trying to predict markets. Some people are of course already working with quants, but there are different lines of attack.
“You would need really fast computers and have access to the right kind of people who can point the computers in the right direction.”
Once you have access to a large volume of data from which AI technology can learn, and the ability to deploy and implement AI at scale, the biggest challenge is ensuring that it keeps learning and improving, said Hartley. “The final step – ensuring that it keeps learning and improving – may be the hardest to achieve and is potentially a major barrier to fully harnessing AI,” he said.