Post Event Report - Building the Future of Finance with AI and Machine Learning
Thu, 06 Apr 2017 05:53:37 GMT
Post event report by Stef Weegels on the Verne Global and Realization Group “Building the Future of Finance” roundtable, held at the SAS Innovation Lab, London 16 March 2017
Artificial intelligence is still in its infancy and depends on effective data and computing power, but the case for harnessing the technology in financial services is growing rapidly, according to speakers at a Verne Global roundtable.
What is the precise difference between artificial intelligence (AI) and rules-based systems? How must data and compute power be optimised to exploit the opportunities offered by AI? And what are the most compelling use cases for this technology within the financial services sector?
These were some of the questions tackled by a panel of speakers at a roundtable hosted by Verne Global in London on March 16. In a wide-ranging discussion, contributors sought to define what constitutes true AI and to deal with the complexities of bringing machine learning solutions to market.
With the discussion led by Verne Global and several companies that provide analytics and data services, including SAS, Sybenetix and BT Radianz, there was a natural focus on the importance of data. In today’s increasingly complex market structure, any effective analytical model must be able to mine vast quantities of data very quickly, and this is a hurdle that must be overcome before the benefits of AI can be realised.
The challenge for financial institutions is that data typically resides in disparate formats in multiple locations and while many firms have significant budget assigned to data management, they often struggle to invest it effectively. Without getting to grips with data, it becomes more difficult to implement the latest machine learning techniques that rely on both data and compute power.
While the natural objective may be to concentrate data in as few locations as possible, an alternative is to bring the computing power to the data, according to one participant. This would avoid the legal difficulties that may arise when data is moved across borders and banks have to comply with multiple sets of regulatory requirements.
As market participants and service providers consider the best ways to optimise data and compute power, a parallel stream of work must consider how the paradigms of machine learning can be best used to improve the industry. Contributors agreed that while AI may have the ability to transform the way things are done, there is still a need to establish the problems that could best be solved by the technology.
A key question is where the boundary lies between rules-based algorithms that apply a predictive logic to particular workflows and true AI that replaces human decision-making with machine learning and is equally if not more effective than the human brain. Advanced as the technology may be today, it was agreed that real AI must apply a non-deterministic logic and this may still be decades away.
One participant likened the writing of this kind of non-deterministic logic to trying to explain the appearance of waves at the top of the ocean. One might have a highly sophisticated understanding of the physical parameters causing the waves, but to explain exactly why a wave breaks at a particular time is a challenge of an entirely different order. In the same way, understanding why a human decision is made at a particular time and writing technology to replicate that decision-making may take a long time to perfect.
In the mean time, speculation continues over where the business case for AI is most pronounced. Given the importance of trading and the success of algorithmic execution over the past decade, there is a natural tendency to focus on the trading desk as the first step for this kind of innovation, but other functions within financial institutions may well have a more compelling need for machine learning.
Surveillance, for example, is an area of increasing importance where large volumes of structured and unstructured data are used to power early warning systems that will identify possible sources of malpractice both within and outside institutions. Investment strategies might also benefit from machine learning techniques that would mine data from unconventional external sources to drive asset allocation and deliver enhanced performance.
Compliance and legal may be another area where AI can reap significant rewards in the future. Recent news stories reported that JP Morgan has rolled out an automated learning programme to interpret commercial loan agreements in seconds, while the process previously consumed 360,000 hours of lawyers’ time eachyear.
These kinds of efficiencies cannot be ignored, and if financial institutions can use machine learning to automate cumbersome legal processes and reduce the costs and errors that arise from manual processing, then the take-up of AI may be faster than some have predicted.
But one participant cautioned against putting excessive reliance on the development of AI, as the world is rarely black and white, and the majority of institutions are likely to adopt machine learning where it makes the most sense rather than looking to implement it on a wholesale basis across businesses.
While AI and machine learning are clearly rising fast up the industry’s agenda, the contributors agreed in their conclusions that the technology is still in its infancy. In addition to the need to establish use cases and manage the associated data requirements, financial institutions also need to keep up with the increased computational demands that will be generated by AI and consider how best to source the required power.
The real breakthrough in the coming years may come not from computation, data or technical innovation, but rather from neuroscience, as a fuller understanding of how the brain works and how decisions are made will be a pre-requisite to realising the true potential of artificial intelligence.