Artificial Intelligence #41: A practical way to implement #AI ethics

Like most practitioners in AI, I follow the discussion on AI ethics with interest. With the forthcoming EU regulation on AI ? this space is likely to get even more complex (because it is a regulation and not a guideline).

However, I am concerned that there is now emerging an army of consultants and advisors who see AI ethics as a business model i.e. a way to sell their services

?There are various ways to obfuscate the issue

a)?Mix narrow AI threats?with general AI threats

?b)??Anthropomorphize?AI (Anthropomorphism is the attribution of human traits, emotions, or intentions to non-human entities) ex see AI as a Godfather like puppet master in all scenarios?

c) Always frame the discussion as man v.s. machine. Paint the future like a 'planet of the apes' scenario where if we are not careful, bands of marauding?AI will hunt the last remnants of humanity?

You get the picture!

Entertaining as this may be sometimes, it has a serious problem?

If we are not careful, we could end up stifling innovation

If we are not progressive, our entrepreneurs would not be able to compete on a global scale

AI ethics is a good idea but we need to have a discussion on how to implement AI ethics at scale. These problems are often lost in the conversation if we articulate the problem of AI ethics but not the solution.

Hence, I was pleased to see the work of the Alan Turing institute on Artificial intelligence, ethics, and safety . You can download their report here .

For example, they define transparency of AI ethics as

a)????The workings of the object are clear(transparent) but also that

b)????the quality of a situation or process that can be clearly justified and explained because it is open to inspection and free from secrets.

The first part is explainability / ?interpretability of a given AI system, i.e. the ability to know how and why a model performed the way it did in a specific context and therefore to understand the rationale behind its decision or behaviour. (overcoming the black box model)

The second part involves justifying the processes that go into its design, implementation and outcome i.e. justification of use. Thus, the design, implementation and outcomes should be ethically permissible, non-discriminatory/fair, and worthy of public trust/safety-securing.

Hence, they propose three critical tasks for designing and implementing transparent AI

Justify Process: In offering an explanation to affected stakeholders, you should be able to demonstrate that considerations of ethical permissibility, non-discrimination/fairness, and safety/public trustworthiness were operative end-to-end in the design and implementation processes that lead to an automated decision or behaviour.

Clarify Content and Explain Outcome: ?In offering an explanation to affected stakeholders, you should be able to show in plain language that is understandable to non-specialists how and why a model performed the way it did in a specific decision-making or behavioural context. You should therefore be able to clarify and communicate the rationale behind its decision or behaviour.

Justify Outcome: In offering an explanation to affected stakeholders, you should be able to demonstrate that a specific decision or behaviour of your system is ethically permissible, non-discriminatory/fair, and worthy of public trust/safety-securing.

A bigger picture is as below

The point here is – transparency, as defined in this way could be adapted by every company to create their own set of values.

By being transparent both in mechanism(explainable) and process(outcome), you could create a system that could be practically implemented at scale