Back to Blog

Explaining decisions made with AI

Image of Jason Wells
Jason Wells

At the end of May 2020, the UK Information Commissioner's Office (ICO) and The Alan Turing Institute released a guidance paper titled "Explaining decisions made with AI". 


The guidance has three main parts:

  1. The basics of explaining AI – for Data Protection Officers & Compliance teams
  2. Explaining AI in practice – for Tech teams
  3. What explaining AI means for your organisation – for Senior Management

I wanted to share with you a summary of what it contained.  I come from a position of knowing deep tech from a business use case perspective and not a technology expert (my CTO has that covered!).

The topic of AI is deep and rapidly changing as more and more students, researchers and employees are working or developing in the space.  With most technology, there is specific language and concepts used which may not be familiar to those impacted by its use. We should be able to explain AI easily to those who could be materially affected by an AI-assisted decision.

What this means for your organisation

Part 3 of the document covers the various roles, policies, procedures and documentation that can be put in place to help in explaining AI.  I have summarised some of the key points below:

  1. People – Everybody involved in the decision-making processes should be identified and has a role in explaining the AI system.
  2. Support Systems – Put in place support that enables the people involved in the decision to produce and deliver explanations.
  3. 3rd Party AI – If you buy AI from a 3rd party, it remains your responsibility to ensure the system is capable of providing explanations for decisions.

People

The following are likely to have a role to play:

  • Product Manager
  • AI Development Team
  • Implementer
  • Compliance teams & Data Protection Officer (DPO)
  • Senior Management

This shortlist does cover all those likely involved in an AI programme but gives you a starting point.

Tip – Make sure you road test your explanations with your intended audience (this sounds obvious but sometimes is often missed).  It is essential that any explanation addresses critical questions and is understandable by those impacted.

I would suggest a non-Technical team member coordinates the production of explanations.  If the coordinator cannot understand the explanation, it is doubtful your audience will.

Support Systems

Policies & Procedures

When including AI within Digital Transformation, it is essential that all policies and procedures layout:

  • What needs to be considered and actions taken to explain AI.
  • Differentiate between what are hard rules and guidance and be clear on the why, what and who.
  • How to implement the rules and who they should consult for advice.

Design & Deployment Documentation

When design and then deploying AI, make sure you have covered the following:

  • Document what you are required to do under local Data Protection laws, e.g. GDPR or under your policies depending on which sets the higher requirement.
  • Document each use stage of AI adds to the decision and explain it.
  • Keep an audit trail of who you give explanations to and how they were provided.
  • Organise your records well so that can be accessed and used easily by those who have to explain AI to those impacted.

3rd Party AI

If you source your AI from a 3rd party, always remember that you are accountable for explaining it those who are impacted by it.  Make sure the 3rd party provides a clear understanding of why and how an AI decision is made.

The ICO and The Alan Turing Institute document is a great starting point when you begin to adopt AI solutions in your

 organisation.  As AI becomes more embedded into our day to day life I believe it critical that any decision supported by AI must be explainable.

We cannot let AI become a black box that limited people understand when it may have great impacts on people's lives.

What are your thoughts on the use of AI and making certain decisions can be explained?

Let us know below.


Related Posts

Financial Crime Compliance With Luma Zitani

Image of Arwa Al-alshaikh
Arwa Al-alshaikh

Financial crimes are considered a global threat to economies but they can expand to threaten a...

Read more

The Energy Transition and Smart Infrastructure with Tim Rockell

Image of Arwa Al-alshaikh
Arwa Al-alshaikh

The rapid increase in the world's population has resulted in a significant decrease in the ability...

Read more