Ethics belongs in the prototype, not the post mortem

I was on a panel for the first GAFAI Insights session, on a topic I care about: prototyping ethics. How do you build AI and fintech products that are new, but also fair, safe and built around people?

Here is what stayed with me.

What we talked about

• Designing across borders. A product that is fine under one regulator can be a problem under another. Cross border AI has to be designed for the strictest rules it will meet, not the loosest.
• Trust in fintech starts with AML. Anti money laundering and counter terrorist financing checks are much easier to build into a prototype than to bolt on after launch.
• Human centred tools. If a tool reads emotion or behaviour, it has a duty to protect the person on the other end, not just to be accurate.
• Accountability. "Harm" means different things in different cultures. Black box models make it harder to show who is responsible when something goes wrong.
• Bias. All data is a record of imperfect human history. Bias is not a rare bug, it is the default, and it has to be looked for on purpose.
• Speed and care. Moving fast is fine. Moving fast without anyone asking "who could this hurt?" is not.

What I took away

1. Ethics at the prototype stage is cheaper, faster and works better than fixing it after release. It is the same reason auditors prefer preventive controls over detective ones.
2. Wide feedback finds blind spots early. The people least like the builders find the problems the builders cannot see.
3. Regulation is not the enemy. Taken on early, it speeds up trust and adoption rather than slowing the product down.
4. Getting this right is good for everyone. Safer, fairer technology is also technology more people are willing to use.

My thanks to the Global Alliance for Artificial Intelligence for the session, and to my fellow panelists and the moderator for a very good discussion.