[rendx]

To me, it's precision in language. "Doing" involves an action. You could in that sense compare the computer part of a robot to the mind of an animal. The mind may be involved in planning or the main source and possibly in control of the action, but it doesn't do. Your example robotic lawnmower is not merely a computer. It may contain one. The computer inside it does not mow the lawn.

To put my comment in context again: I replied to a comment that said "By that logic, nothing computers do is scary." as a response to "Why would it be scary? Claude is just parroting other human knowledge. It has no goal or agency.". I was following the train of "logic" of this chain of comments.

As long as this hypothetical Claude doesn't have control over objects, it cannot do anything. It completely depends on how its output is processed and received. An LLM spits out words, a computer ultimately spits out bits; what makes them "scary" is not what they produce, but fully depends how that product is translated into action by its environment. It cannot be determined in isolation by only looking at the computer/LLM ("mind") part.

The mind part does not do. A computer may be attributed to have "agency", but the "objects" around it can have agency too. A computer cannot force anything on its own; a toddler (or a president) may have agency, but they require cooperation by their environment to exercise that agency. If you break a leg, you can want to move it all you want, it won't do.

It's not merely a "linguistic semantics thing". Think Nuremberg trials. Who is responsible, in a network of 'autonomous agents'?