AI : Validate before we accept

It has been a while that I wrote a blogpost. I lost the energy due to various reasons, but I got triggered by a blogpost from a good friend of mine, Alessandra Moreira.
In here blogpost, One Reason Why AI Won’t Replace Testers: The Power of Human Expertise, where she mentions that “AI cannot replace human experience or the core function of testing: human-driven insight.”

I wanted to respond on here blog in the comment but thought by myself, why don’t you respond via a blogpost yourself. So here it is.

In the beginning of 2024, I participated in a Ministry of Testing event, 30 Days in AI Testing. There the participants worked on a task that a Ministry of Testing team member posted each day for 30 days long where the participants worked on. In one of the tasks, the participants were asked to come up our ‘own’ AI in testing manifesto. One of my principles was the following:

Validation over Acceptance
The output generated by AI must be validated before it can be accepted. It is not known (yet), because of the lack of transparency, how the AI came to the conclusion of the output it generated. As the A.I. can quickly come up with proposals, checks must be done that be sure that these proposals are accurate. Always be aware that the given output is a ‘prediction based on a set of probabilities’

AI or Generative AI, is simply said a Large Language Model with a neural network that generates an answer on the question (a.k.a prompt) which is asked to it. That generated answer is based on probabilities. The AI collects a lot of information related to the prompt and comes up with the best probable answer. Can we accept this answer? No, we can’t. We need human-driven insight, i.e. we need that validate the answer before we accept it (and use it)