RiskTech Forum

Chartis: AI: A crashing wave, or a gradual flood…?

Posted: 28 November 2017  |  Author: Phil Mackenzie  |  Source: Chartis

It seems like a day can’t go by without a new Artificial Intelligence (AI) story in the headlines. AI is learning to beat Go champions and drive cars. It’s making inroads into law, medicine and finance.

The impression is of an unstoppable tide, one that will crash over every area of life and technology with equal effect. This vision is reinforced by pundits who rarely deviate from the extremes of breathless utopian futurism or apocalyptic doom. But it seems misleading. AI is not magic. It is, by and large, a collection of statistical processes used to build systems that possess a combination of rules-based and iterative or adaptive capabilities. These have strengths, and they have weaknesses.

The very idea of AI often lends an air of mysticism to things which are often less exciting or new than they can first seem. For example, the idea of artificially intelligent chatbots or virtual assistants is a popular one, particularly for compliance or advisory roles. However, there have been past attempts at this sort of thing, like Microsoft’s infamous Clippy. ‘Would you like some help with your risk management?’ Perhaps not.

At the other end of the spectrum, some successes have become routine and everyday so quickly that they fade into the background. Robotic Process Automation (RPA) sounds futuristic and exciting. But every time your wi-fi goes down and you click on ‘troubleshoot connection’, an automated workflow process kicks off which runs through a series of steps to attempt to resolve the problem. This is RPA, and it is undeniably useful, but it also feels disappointingly prosaic.

Mysticism about AI can impede perception of its progress. If failures are forgotten, and successes are internalized, then AI can constantly be seen as a static wave about to burst and break over everything any minute now. The truth is likely to be both blander and more complex – namely that AI has already flowed into a number of areas.

The machine-learning processes which underpin ‘modern AI’ are nothing particularly new, but they have been powered by a massive explosion in the quantity of usable data, and by improvements in the capability to process that data, including Graphics Processing Units (GPUs) and advanced databases. So the areas in which AI is making its impact are mostly those that involve processing large amounts of data, and where iterative optimization is particularly useful. These include rules extraction, document analysis, client and counterparty segregation, and the tuning of RPA processes.

Many of these can seem a bit dull when compared with robot cars and man-vs-machine Go. But in many ways boring and repetitive is the name of the game.

So rather than a huge obliterating wave, perhaps AI is better thought of as a slow, relentless flood into a vast and complex problem-solving landscape, creating pools of activity in areas where it does especially well.