A view from Brussels: Lessons learned on the practice of AI literacy

There is a famous syllogism about cheese, but it only works if you picture Emmental. The more cheese there is, the more holes there are, but the more holes there are, the less cheese there is. Therefore, the more cheese there is, the less cheese there is. 

I was thinking about the role of consensus-building in EU lawmaking and the challenges of reconciling dozens of positions. "Consensus is the business of politics." Somewhat unexpectedly, this quote is from the author of the Jurassic Park movies Michael Crichton, who is primarily a renowned physician. 

Another fun quote comes from former U.K. Prime Minister Margaret Thatcher who said, "Consensus is the absence of leadership." 

So, if we apply a mean syllogism trick, we could find that the business of politics is the absence of leadership. But of course, there is always more cheese than there are holes. 

Artificial intelligence literacy is the same: the less there is, the more there is. The European Commission may propose to suppress that obligation from the AI Act two years after its entry into force, yet companies continue to strengthen AI literacy programs across the board.

During the AI Act negotiations a few years ago, the pertinence of AI literacy obligations was one that easily garnered consensus. And indeed today, many companies report that well over 80% of their workforce has received basic AI and AI awareness training, according to the empirical cataloguing of AI literacy practices led by the Commission. Some report having trained up to 100% of employees that have access to the company's internal AI models. These figures are even more interesting because many programs kickstarted before the AI literacy obligations came online. 

The AI Act's Article 4 on AI literacy mandates providers and deployers of AI systems to take best-extent measures to ensure a sufficient level of AI literacy of staff and people dealing with the operation and use of AI systems on their behalf. That should equate to equipping staff with the necessary notions to make informed decisions regarding AI systems, depending on context and other factors. 

When looking at the approach many companies have deployed, AI literacy actions also contribute to a broad AI governance approach, serving obligations across data governance, human oversight, risk management, documentation obligation and cross-functional collaboration.

Of course, this is not without challenges. Training cannot be static or a one-off. The pace of AI changes and adoption means training will have to be continuous and focus on the fundamentals that stand the test of time. 

This can mean significant investment in time and resources. So, it is quite interesting to see that many organizations also comment on gaining tangible, if not measurable, benefits. Some fall neatly into the governance and compliance bucket, for example, when they improve risk assessment and decision-making capabilities which can serve AI and other digital responsibility domains. 

Companies also list AI literacy benefits in two broader buckets. First, many underline the role that AI literacy plays in serving business operations, from improving the feedback that products teams get to the way cross-function collaborations happen. Second, they attest to AI literacy's broader contribution to building a responsible-AI awareness culture in the organization, one that puts trust as a cardinal point of its approach.