Canadian AI standards, too!

I have lamented the lack of Canadian AI regulation efforts. And for the past couple of posts I have referenced how the creation of standards in other jurisidictions made be an entrypoint in to regulation. And lo and behold, we have a set of Canadian AI standards with regards to ethical design and use of AI. This is a national standard from the CIO Strategy Council but is a standard developed in accordance with the Standards Council of Canada, a Crown Corporation. As such, it has some significant element of a government imprimatur.

The standard sets out minimun requirements for protecting human values and incorporating ethics in the design and and use of automated decision systems. Those are systems that use artificial intelligence, and in case your wondering what artificial intelligence is, this standard fully adopts the definition we first saw here. Which is a bit of a thing, because if a defintion keeps being adopted and replicated, eventually it’s the one that everyone goes along with.

The standards themselves are really helpful. If you know nothing about AI, like me, but you are in governance of an organization, they tell you exactly what you probably should be doing. This includes oversight functions and risk assessment functions. Notably, an organisation is directed to consider national and international guidance or standards. That one seems a little difficult, given that standards are fairly inconsistent internationally at this point (could an organization adopt a lower standard by cherry-picking internationally?).

The standards go through management, design, deployment, and monitoring and maintenance of AI systems. At the macro level, this creates a best practices checklist from which deviation would have to be explained, should something go wrong. It is not a short set of standards–39 items all in. One does wonder if in a small, lean organization it will be difficult to support the bureaucracy to track all of these. However these are ethical standards and it may be that in many applications of AI, the impact of the application will be narrow enough that compliance with these standards will not be onerous.