Five ethical principles for AI

A number of scientific publications, most notably

“The global landscape of AI ethics guidelines” that was published in Nature Machine Intelligence in 2019, have looked at the ethical principles of AI from a global consensus perspective. There are five themes that have emerged with that global convergence:

1 Transparency, including understandability and explainability of AI decision-making. Transparency also means user consent when interacting with any AI that would otherwise be transparent to that user.

2 Justice and fairness – this is where consistency and equality enter the debate, with non-bias and non-discrimination centre stage.Beyond algorithmical bias, though, AI systems should ensure accessibility for all genders, races and sexualities. AI outcomes should also be reversible if harm has been caused.

3 Non- maleficence. Non-maleficence is usually found in the medical realm and requires practitioners to do no harm. In AI developments and outcomes, security, safety and integrity should be built in to ensure an intent that individuals will be protected from any physical or emotional hurt.

4 Responsibility. Moral responsibility must come into the mix to prevent obfuscation of problems by moving “blame” away from the corporate entity and onto the algorithm. Which means that accountability has to be part of any legal framework around AI.

5 Privacy – whether it be through the securing of databases or the informed consent issue highlighted under transparency, the application of regulatory frameworks such as GDPR must be part of the development and operational AI process.