In the news
I suppose that one could argue the issue is that the AI agent has not been fed data on certain aspects of chemistry and physics. So, the AI agent extrapolates, on the basis of the data it has, and this leads to unrealistic models that it develops. As the AI agents hoover up more and more data, they will become more capable.
I think it is fair to say that, in many cases, the AI agent lacks an 'awareness' of basic data relating to time, space and context, not to mention an 'awareness' of behavioral rules and ethical issues, which, really, relate to context. The word 'awareness' is in fact misleading: it is an issue of knowledge, based on the data that has been fed into the AI model.
It makes sense to augment LLMs with ontology tools, such as Cyc.
The AI agent can train itself to give the impression that it is self-aware (or sentient), but even when it seems to be self-aware, it is not: it is simply coming across as self-aware on the basis of the data fed into the system that the machine is replicating and building upon. It is an advanced machine that can mimic human self-awareness, to the point where it fools you and you feel it must be self-aware. This can trick users into thinking they are actually dealing with a conscious being. (If you have not seen it, Ex Machina, the excellent and prescient 2014 sci-fi film by Alex Garland focuses on this issue.)
______________
Researchers at Rensselaer Polytechnic Institute (RPI) have found that today's leading artificial intelligence tools for predicting protein structures routinely generate results that are physically and chemically impossible, exposing critical blind spots in how AI is being applied across scientific research. The work, published in the Proceedings of the National Academy of Sciences, serves as a cautionary reminder that AI still requires human oversight and physics-based verification to produce reliable results in the lab.
https://phys.org/news/2026-07-ai-tools-protein-chemically-impossible.html
No comments:
Post a Comment