Assessing and developing AI projects
Beyond the hype, I assess with you whether an idea for artificial intelligence holds up, what it requires in terms of data and law, and what would work just as well without AI. At the end you have a decision you can defend.
Audience
The offer is aimed at public bodies, foundations, associations and research institutions. It fits when a concrete application is already on the table, and just as well when the question is still open. Leadership and the relevant department should take part together, and IT joins as soon as data is involved.
Way of working
It starts with understanding. This is not a foundation course: you learn as much about the technology as you need to share in the decision. Half a day in the opening session is usually enough.
Then comes the assessment. We look at the data available, classify the project legally according to the risk categories of the AI Act, and consider the expected benefit as well as the alternatives without AI. A simpler method often solves the same problem.
After that comes the workshop. Over a full day we develop the application further together and harden it where it would otherwise fail in daily operation. I work with the AI Toolkit, a card set for developing application ideas in mixed groups.
Finally comes the handover. In two hours we go through the assessment together and name the next step, the effort involved and who is responsible, each with a name and a date.
In all, that makes three sessions, spread over six to eight weeks.
Result
You receive a written assessment of around ten pages, with a recommendation, open questions, named risks and an order for the next steps. It is written so that it can serve as the basis for a formal decision.
Not included
Legal advice is not part of it. I tell you what a method does technically and where its limits lie. Whether that is sufficient in your case is for your data protection officers and your legal team to decide.
Nor does it include training your own models or running production systems.
I do not take part in a procurement where the product has already been chosen and only needs justifying.
I advise against it when the data does not support the idea, when a method without AI achieves the same, or when no one is named who will answer for the results once it is live.
Frequent questions
What if the outcome is to advise against AI?
Then that is the result. A considered no costs less than a project started on shaky ground.
Do we need developers of our own?
Not for the assessment. For an implementation I will tell you which roles you have to fill.
Does this also cover tools we already use?
Yes. Existing tools can be assessed for data, law and benefit in the same way.