Question of the Month: Would you trust an AI to select components for your next system design?

AI is NO substitute for good engineering calculations.

I have been following AI for over 40 years and it even more over hyped now than any time in the past.

Until you can verify what happens within the various levels, you have an untestable application.

Buyer beware!

Trust is something that has to be earn. Don't trust AI.

I said I'd accept limited input.  That comes with the caveat that I'd vet everything the AI selected.  In this capacity the AI would act as an enhanced search engine of sorts.  For what it's worth  I was very close to answering no but eventually thought there might be a balance where AI could be used in a limited way to help optimize the initial component search.

Nope!

AI is only as good as the company that created it. Companies have constraints. I'm leary of what constraints are reflected or not reflected in their product.

DAB, I think that for this particular domain, it is actually a possible fit. The main drivers for engineering are domain knowledge, math and experience based on the past. And no feelings. I think that AI algorithms may be able to capture that.

Once it's created by a number of skilled people, they may outsmart single you's skills. 

I don't see why not. Hardware design is composed of patterns all over the place.

It's just a matter of time until people find ways to get online netlists and schematic PDFs to become good training data. It will even be able to improve them once there are better digital representations of devices and what's inside them, and telemetry from all the subsystems, so AI has rich sources to learn even more.

Already, completely generic AI systems can do some limited stuff, even building up simple VHDL for instance. It's already specifying what logic to use and how to assemble it. And that's with zero attempt by firms at developing AI for electronics circuit design.

Already I trust it to write code for me, and then I'll vet it. It needs a helping hand so far.

It depends. A lot of what is being called AI today is applied statistics. A lot of what has been called AI in the past is clever programming. I'd want to be sure it's actually modeling animal-level cognitive processes before I call it AI, at which point I'd probably trust it as much as a human that had been similarly trained.

I would accept suggestions to evaluate, if the process is efficient and if the suggestion come with appropriate justification. Not automatic trust, but assistance is welcome.