As of Feb. 2, the AI Regulation requires every organization dealing with AI to be AI literate. To help organizations with that, the AP published a handbook "Getting Started with AI Literacy" last week. The VNG already released a report on this last December.
These publications and the book "Co-intelligence" by Ethan Mollick I used as inspiration for the 5 tips.
But first things first: what do we mean by AI literacy? AI literacy means that employees have a basic understanding of what AI is, how it works and which applications are relevant to their work. AI literacy also requires employees to realize the risks of use. Risks to privacy, the occurrence of biases in output and the need to be transparent about use.
A misunderstanding I encounter quite often in practice: we do not develop or offer AI. We only use it on a very limited scale. Therefore, we have nothing to do with the AI Regulation and the requirement of AI literacy. Not is less true. The AI Act and therefore the requirement of AI literacy applies even if your organization only uses AI. Article 4 from the AI Regulation talks about "user responsibility" and thus makes the use of AI subject to the regulation.
So if you are using Copilot from Microsoft or Chatgpt in your organization, you are already user-controlled and need to work on AI literacy.
This may sound like an open door, but all too often I still encounter an organization enthusiastically getting started with AI. Pilots are started and use is encouraged. In itself a good thing (see Tip 5), but more is needed for responsible use of AI. Learning by doing is a great way to get started with AI. But a future-proof approach requires a multi-year action plan. This builds board support, establishes responsibilities, sets budgets and facilitates employee development, AI literacy.
Unfortunately, I cannot give you a general approach to for AI literacy. What is an appropriate approach for your organization depends on a number of factors.
First, the risks of the AI systems being deployed come into play. The AI Regulation distinguishes 4 risk classifications from "prohibited" to "minimal or no risk." The higher the risk level of the AI systems, the higher the requirements for AI literacy.
A second factor is the employee's role within the organization. A higher level of knowledge is expected of policy makers, directors and supervisors, for example. After all, they must be able to make the right policy choices.
Finally, the context in which the AI system is deployed also affects the level of AI literacy required. Whether a system is deployed to develop course materials or for assessments of new employees, for example, makes quite a difference.
AI literacy is not just about the technical operation of AI systems, but perhaps much more about the social, ethical and practical aspects of their use. Employees must be able to interpret the output of the AI system. They must be able to look critically at the output of the system. Realize how an AI system decision affects the people the system decides on.
In short beware the effect of the "computer says no" from Little Britain.
Developments around AI are happening at lightning speed and are irreversible. As Mollick puts it in his book: this is probably the worst AI we will ever work with. At the same time, he is achieving remarkable results by using AI in his lectures and having his students use it.
I know this goes against the advice of Surf, the ICT collaboration organization for education. They advise against using Microsoft's Copilot given the privacy risks. Good that the organization is in talks with Microsoft about this, but waiting for the outcome risks falling behind.
And yes, there are still many questions and uncertainties regarding the AI Regulation that still require further elaboration. Be aware of these risks and uncertainties, but at least start doing some finger exercises with AI systems already and thus also work on the AI literacy of your organization.
