Moving past the generative AI hype to focus on real governance and liability

Yes, GenerativeAI is what people are talking about. Yes, it presents specific risks (notably re disinformation). But from an aigovernance and liability perspective, the risks associated with predictive/discriminative AI are just as relevant.

In her speech today re the State of the European Union, Ursula von der Leyen said that we should “not underestimate the real threats coming from AI, mostly GenAI”. Surprising, given that GenAI wasn’t much of a topic back when the AIAct was proposed (or even the AI Liability Directive, for that matter).

Don’t focus only on ChatGPT and the like. Many other kinds of “AI systems” (the OECD’s definition is very useful) are commonly used by companies across the globe – from cybersecurity threat detection to algorithmic recommendations of products to consume. Some can influence people’s behaviour, others their health.

Any organisation should look at managing several kinds of risk, notably:

1) the risk that someone using AI within the organisation does something that triggers the organisation’s liability towards others,

2) the risk that the use of AI affects the assets of the organisation (e.g. IP-protected assets or confidential information),

3) the risk that something the organisation produces (products, information, etc.) is used to automatically affect others, in particular in a negative manner (// automated decision-making, with also its GDPR implications).

That’s what proper AI governance is about. Not just about “Can we use Bing/Claude/…?”.
If your processes don’t yet cover all that, well, maybe you could use a hand!