How to speak AI?
Just over a year ago, I wrote here on the topic of digital ethics: Who is watching the robots? One of the challenges I noted then was that collectively we need to understand what the deployed technology is; how it works, how it arrives at the outcomes it does, what inputs it needs, and to be able to demonstrate that the outcomes are genuinely in customers’ best interests. But I have been increasingly troubled by something which makes getting to the point of collective understanding more difficult; the fact that, whilst there are many excellent commentators on the subject, there is significant difference in the language and terminology used. This makes it a complicated subject matter for the informed, never mind for the non-experts out there.
So, I was pleased to see the publication in May 2020, of guidance developed by the Information Commissioner’s Office and the Alan Turing Institute. Explaining Decisions made with AI combines three guides that helpfully explain the benefits and risks of using AI and makes for a sophisticated beginners guide and a useful engagement tool for those who consider themselves to be ‘fluent’ in AI.
First off - , before they get into the explaining, these Guides do a very important thing: they offer definitions. And these are clearly definitions that have worked through the various considerations and take into account the diversity of the audience. There is no understating how helpful this is, and will be, in widening engagement. The definitions work through key areas, such as the steps of training, testing and deployment. With every organisation potentially using different terminology for the same, or similar things, it is a welcome moment when respected bodies like the Information Commissioner’s Office (ICO) and Alan Turing Institute actively promote the value of a shared lexicon.
Therefore, I would urge you to spend time reading the ‘Definitions’ section of ‘Guide 1 - Explaining the Basics of AI’. It is written in language we can all get to grips with; speaking AI shouldn’t and mustn’t just be something for only the specialists. That necessity I mentioned before, of the need for collective understanding is perhaps a step closer once you have that common language. So, if you are reading this as the bank’s expert or a manager engaged with AI on a day-to-day basis, perhaps you could start introducing and aligning your own terminology to this. For those of you trying to get to grips with the basics – then I would also advise you to start here.
And why does this interest us as a professional body? We come back to my first point: which was around responsibility and accountability. Those impacted by an AI supported decision, and those being supported by or using such technologies, should be able to hold someone accountable for it and our efforts as the UK’s professional body for retail bankers are focused on helping our members achieve and maintain the combination of skills required to be professional, responsible and accountable.
For now, I recommend these definitions to you and then you may wish to return for the next blog in this series when I will delve deeper into the Guides.