Ethical AI in a Rapidly Evolving Landscape

By Humera Noor Minhas (Eyeo GmbH)

Transcript

Last Tuesday, I had a biking accident.

It was not that bad, I'm here, but one of the questions that I was asked after this accident was "Humera, were you wearing a helmet?" And my answer was, nope.

The other question that I was asked was, "don't you know it's important to wear a helmet?" Yes, it's not a legal restriction, legal requirement, but you should be wearing a helmet.

And I was like, yeah, I know that.

But to my defense, I was already out of the door when I remembered putting the helmet on, and I didn't want to go back and delay my ride further, and secondly, I didn't think that I would meet an accident that day.

Does this sound familiar when we talk about ethics and AI?

When we build ethical systems, when we build AI solutions, we know we should be following ethical principles, the guidelines are there, but somehow, we just miss out on implementing those principles and incorporating them in our solutions.

The state of the art is that we have those principles ready.

The W3C guidelines are there.

Tess spoke about it, Dom spoke about it as well.

And we also know that they are so much ingrained with our values and principles, so we all know that we want to be fair when we want to develop these systems.

We all know that we want to have transparency, we want to give users the control.

But how do we take that step?

How do we make sure that we do implement them?

It's time to put the theory into practice, actually.

Again, W3C has come up with so many good guidelines.

There are principles, there are workshop templates, there are documentations available, so anybody who wants to do it, they can.

And now the only little thing is taking that step, going back into the house, get that helmet before we go into the journey, or we make sure that the helmet is right there embedded, placed on the cycle, so that you don't think of ethical principles as an afterthought, rather, you have them by design.

And this brings with so many questions.

Yes, there are principles there.

Yes, we should conduct these workshops.

We should do this risk management, we should have this and that.

But what workshop?

Who owns this workshop?

If I do an ethical risk management workshop, who am I?

Am I somebody from the engineering, or is it a product ownership?

Should I think about ethics right from product discovery, or is this something that comes in when I am a little bit into through my POC?

Or, or, or.

And as you can see, these questions are apparently very random.

We are talking about sustainability.

We are talking about, should we go open source or remain closed source?

How do we keep in touch with the upcoming regulations?

How do we make sure that the upcoming regulations are incorporated?

So, there are definitely more questions that we have answers.

And this list, this slide, is actually not meant to be read, because you cannot read it from there.

There are so many more questions.

But at Eyeo, based on our experience with the developing machine learning solutions for ad filtering, here are just three recommendations that I can give to begin with.

Enablement.

We need to enable our teams.

We need to train our teams.

We need to talk about ethics every now and then.

We need to make sure that everyone is taking ownership right from the top leadership, because until then, and unless the top leadership is in there, it cannot drizzle downward.

And the third is specifying the processes.

If it's part of your checklist to think about ethics right when you start the discovery process, then it becomes easier to put it into practice.

But that's not it.

Of course, this is just the beginning of the discussion, and we will continue with this tomorrow in our breakout session.

Please do join, and we would love to discuss how the companies in real world are putting ethical systems in practice.

Thank you very much. (audience clapping)