Yes, you are. In attributing innovation to ‘the model’, you collapse agency.
Interestingly, at an earlier point you (correctly) rejected ‘great man theories’ of innovation, but now you threaten to make the same move as them: the ‘lone inventor’?
If you tell a machine it has two marbles and if you give it another two marbles what does it have, and it doesn’t have the rule of addition, what does it tell you?
You’re assuming interpolating is extrapolating.
It might be if we were also were asking about AI rights?
Then the dis-analogy to women arises. Granting rights to women served equality & justice, whereas granting rights to AI would only hamper accountability.
Agreed. Parents are associated with personhood. In all legal jurisdictions as far as I know.
Hence why I made the point that “AI generated patent” is a legal misnomer, before we even get to it hiding the humans in the process.
What does it mean to imagine “models running on their own”?
The model is not a miraculous birth, popping into existence sans designs, intents, protocols etc made by humans.
But here you slip straight into agency for the model, rather than disaggregating the process to find the agency.
I just think that slippage is deeply problematic.
Yes but you then IMO opinion implied novelty could be attached to AI if patents resulted from their activity.
That’s is akin to calling toes speedy on the fast runner.
You suggested patents are novel and if AI-generated then AI is engaged in novelty. It’s there in B&W tweet.
You then suggested the issue is an empirical one. Fine. Only natural persons can hold patents (legal fact). AI is a tool in the human process (techno-social fact).
What’s the problem?
Fair enough if you don’t understand what was written.
I tend to advise caution in interpretation at that point.
FYI those of us that study knowledge generation professionally are in fact in disagreement about AI consciousness.
The moving of goal posts affects your defense of AI innovation more than my critique of it.
Also, you assert AI can patent & we should let the query be empirically determined. I pointed out AI is not a natural person thus no patents. All of a sudden facts are irrelevant to empirical enquiry?!
Accepting AI consciousness involves a dubious form of ‘isomorphism hunting’: low-level dissimilarities brain-to-machine are trivialized & focus placed on similarities at high levels of abstraction.
A contrary is @davidgunkel.bsky.social’s relational account for NOT dismissing AI consciousness.
AI-generated drugs are using machines to augment a human testing & creating process.
E.g.: Underneath Exscientia’s ‘the machine did it’ hype is a straight scale & speed claim.
I’m rejecting the approach that claims AI-innovation not AI as a tool.
www.technologyreview.com/2023/02/15/1...
In all jurisdictions on earth, inventors are not natural persons. Patents go to natural persons.
This is just current legal fact.
AI augments the process whereby people create something that is patented. Why the need to cut the human out just to ‘recognize’ that augmentation role?
You of course know that humans are in the loop even in your AI outputs.
Training data is the human in the loop.
Animals can be inventive but they’re biological beings like humans.
I am arguing explicitly that you lack a good theory of why or how AI innovates.
Above you backtrack from a behavioural account, which I predicted you would.
Below you also backtrack from the claim innovation requires no thought.
“It just has to” is where you’re at!
No, that is not the problem.
It’s the non-discriminatory part that is the problem. I’m not sure you see the point of Chomsky’s ‘anything goes’ argument?
To avoid the behavioural theory of innovation
But now you need to explain how and why AI filters the outputs.
I think you’ve landed on a form of behaviorist theory of innovation and I suspect you’ll now need to backtrack.
Backtracking is good, though!
If you think it’s special pleading to refer to human specialness, the real issue becomes whether you face squarely the fact of AI non-specialness.
If we know all AI’s principles and modes of operation, we can readily discern incapacity.
Chomsky said the definition ‘anything goes’ captures all the rules of nature (including the fact of innovation).
But it’s so non-discriminatory it rules nothing out and is thus useless.
Applied to these random recombination = innovation arguments for AI?
They’re useless. Technically.
This novelty definition has the disadvantage of not escaping the Monkey on the Typewriter trap. As it’s said, with infinite time the monkey will type McBeth plus trillions of nonsensical plays.
Introducing goal-directed but devaluing intentionality = monkey on the typewriter.
Anti-pedestal allergies, targeting unwarranted abstractions = ✅
I suggest, though, you’ve went too far the other way. You’ve made innovation coterminous with whatever recombination manifests from a recombination operation.
Like fleeing principles only to embrace behaviorism.
Devils Advocate etc
Just because innovation is more random than suggested by Great Man Theory (GMT) only shows GMT is flawed, not that innovations are random. Affirming the consequent is a fallacy.
Hostile reading, maybe, but I’m prodding at looseness that lowers a bar just to save AI output.
Your argument for AI novelty has moved from conflating language & thought, to denying thought any role in innovation, settling for blind chance alone.
‘Recombination sans thinking’ is a high price to pay just to call AI slop ‘innovative’. You’re digging a hole
Angus Taylor talking cost of living pressure in Canberra’s wealthiest suburb - “hard times hit everyone” he said - is code for “you folks will be OK?
Where is the ‘magic touch’ assumption in my argument?
It seems you’ve imported it in, for reasons unknown.
Is it wise to conflate language and thought? That’s the methodological query I raised.
“A complete joke”
After the starting gun fires, Australian researchers have to wait 2–3 years before even starting the race.
Really clear article explaining the impossibly long new time-frames for Australian Research Council grants.
By @liammannix.bsky.social
FYI your argument to the effect novelty by LLM’s cannot be foreclosed, because human & machine both recombine prior ideas/training data, is flawed. You have to assume you can conflate language & thought to make the recombination argument work, and that conflation assumption is flat wrong.
We cannot assess degree of substantive or innovative claim because Kustov’s Claude-generated paper is like a Seinfeld episode in AI hype. Claims that claims are made but no claim, just self-referential reference.
You know what’s missing from that entire Kustov article?
A single word about what substantive argument any AI-generated paper made.
Think about that. The claim to fame is words were made to follow each other. Groundbreaking.
