In this brief dialogue, Speaker A challenges the notion that current generative AI systems are truly cognitive, describing them as statistical next‑token models that predict words based on prior context. A past controversy is referenced where a former Google employee claimed Google had a "cognitive" AI, which Google denied. Speaker B counters by outlining emerging approaches that augment generative models with non‑human digital inputs—advanced cameras, environmental sensors, and direct integration with CAD software—to automatically generate workflows for part design. The exchange ends with lingering skepticism about whether such systems can be considered genuinely cognitive.
Chapters
So when you say cognitive AI, so my understanding of generative AI is it's really just an extension of traditional model-based statistics kind of thing. You have half of a sentence and the system figures out what the next word is, and then from there it figures out what the next word is, etc. And so it's really not cognitive. And you remember, there was this Google guy that got let go a couple months— well, not probably 6 months ago— when he was claiming that Google had a cognitive— Cynthia AI, right? Exactly. And Google kept on saying, no, we don't, no, we don't, it's not cognitive, right? Because that's the first time I've heard from generative AI claiming that it's— that there are some cognitive— I mean, it seems very cognitive when you, when you have a prompt engineering conversation with it, right? But in reality, it's still just based on statistics, right? So how are these companies kind of going in the direction of cognitive, or what's the difference there?
So 3 that I've seen, and there's actually a 4th I'll be going in to see that in talking to the CEO, I think they're probably the closest to a true cognitive. They're tying in rather than human inputs into the generative AI system, they're tying in digital inputs. So it's, you know, advanced camera systems that are examining the work area of scope, its sensors in the area, it's tying into, you know, directly into CAD programs and that. So basically, as you're developing on CAD, the AI is automatically starting to generate the actual workflow to produce the part.
Hmm, okay. So are they truly cognitive? Probably not.
Probably not. But they're getting there, right? Yeah. They're, they're trying to equip the generative AI basically with eyes and ears so it can learn from its environment and create its own work instructions.
