iiinova — where your digital self keeps evolving

Daniel Drucker @danieldrucker.bsky.social · 30/09/2026

How do you think philosophers have (quasi-?) converged on this more nuanced concept? Any clue where it’s coming from? Is that really an intellectual history question?

170

Daniel Drucker @danieldrucker.bsky.social · 29/09/2026

Sorry, link got messed up. Here's a gift link, though! www.theatlantic.com/technology/2...

theatlantic.com

Why Do AI Agents Sound So Frustrated?

"ARGH... WHY IS THIS SO HARD."

060

Daniel Drucker @danieldrucker.bsky.social · 29/09/2026

Sorry, got cut off in the transfer, I think. Here's a gift link, either way! www.theatlantic.com/technology/2...

theatlantic.com

Why Do AI Agents Sound So Frustrated?

"ARGH... WHY IS THIS SO HARD."

023

Daniel Drucker @danieldrucker.bsky.social · 29/09/2026

@kmahowald.bsky.social and I wrote this piece on expressives in LLM agents' chains of thought: www.theatlantic.com/technology/2... We look at some surprising, emotionally expressive language that appears in AIs' "thinking" when they're doing math, or hacking. We argue it's likely functional!

theatlantic.com

2214

Daniel Drucker @danieldrucker.bsky.social · 27/09/2026

It’s very hard to negate specific conjuncts of large conjunctions without repeating the conjunct, so for efficiency, language allows us to focus the likeliest “conjunct” to be rejected. That seems plausible to me anyhow.

120

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Okay, that seems to me to a major concession, though. They are projectible, so they can be real kinds. (I don't buy the conventionality thought at all but maybe you wanna lean on that harder.)

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Why can't functional kinds be projectible? "Where we see an animal locomote, we will also see limbs with hinges", etc.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Why do you think that's the real nature of minds?

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Why don’t they get to be kinds?

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

What matters is that there are real causal relations even involving climbing, and whether mindedness is like climbing, as it seems to be conceptually.

000

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

“Because these bits so arranged realize this function, which is what is essential to being a member of the kind.”

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

That just transfers the mystery to strength ;)

110

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Your reasons remain mysterious for me, because I think it should be common ground that functional similarity is some evidence.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

“No reason” seems wildly overstated to me personally. Surely you think functional similarity is some evidence?

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

You seem to see dualism in a lot of places it’s not!

000

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

We count them as diverse realizers of a common something only because, conceptually, we grasp that realizing the function is what makes them a kind of thing worth one name. Again the dualism stuff is just rhetoric, nothing I’m saying is metaphysically dualistic.

000

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

The existence of them in total I mean

000

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Wild! I think causal relations don’t conceptually or metaphysically depend on physical law being exactly how it is.

110

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Physical commonalities only insofar as they realize the given function, which function realization is not socially constituted. The function is conceptually prior.

110

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

I don’t understand the relevance? Sure, ghosts aren’t nomically possible, suppose; if they were, they could clearly have beliefs.

110

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

I’m not…? Again, folk-psychological/normative reference fixing, but physical powers and realizing.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

The “contingently, in us” is the whole game!

110

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Come on, you know it has nothing to do with dualism. Functionalism is completely compatible with token physicalism.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

I think it's likely that all concrete things are physical, I'd just deny that low-level physical details matter to all causally important phenomena.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

THEORIES are answerable to empirical psychology. "What realizes this function, and thus is minded?" is a question for empirical psychology. But I see no reason to think empirical psychology should have final say in what mindedness IS.

000

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Sure, I've been a little loose with burden of proof here. At a minimum, it shows you need an actual argument that mind is causally inefficacious if defined in broadly folk-psychological terms. I would need an actual analysis on the table to argue from in order to make the more confident claim.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

That's just rhetoric. I disagree. If you have specific concrete arguments here, that'd be helpful. I think it's the best option we have for the central folk concepts we have, like 'belief', 'mind', 'person', etc.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

The only point I'm making is that 'creature that is running' can take part in causal relations, despite being very very multiply realizable. Same with 'is a currency', same with 'is a belief', despite all being very multiply realizable and utterly independent of underlying chemistry or biology.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

There's a whole dialectic about the metasemantics after Kripke. I don't think he showed anything like that Lewis's project can't be made to work. I think it's probably still the most popular and plausible option.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

What allows them to be causally efficacious is that performing the function they do robustly explains many happenings, both in the currency case and the mind case (and vision and running).

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Sorry, where did Kripke show that? Specifically I'm talking about mental concepts, not phenomenal ones. So, belief and desire, not consciousness.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

The point is that functional kinds can absolutely be causally efficacious. (Another functional kind would be, I dunno, vision or running: can perfectly well figure into many causal explanations, but very different physiological implementations.)

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Definitely not! Could not disagree more. I'm taking roughly Lewis's position: we latch onto the kind via folk platitudes, but the kind itself is not social but rather functional (when we investigate what it could be we have latched onto). Social construction is nowhere involved.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Oh, 100%, do not think they're grounded in social practices. The point is that money, like minds, are causally efficacious despite being multiply realizable, and despite function's being the determinant of whether a physical arrangement constitutes a member of the kind.

220

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

(We have strong intuitions that minds and currency are multiply realizable, less strong for tigers, etc.)

000

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Sure, but the nature of the thing we latch onto in our biological ignorance is something that is important but not because of its biological details, as money is important outside of its metallurgical details, etc.

220

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Right, so I think mind can exhibit that kind of homeostasis while also not being dependent on physiological details, i.e., accessible to the folk. "Anything that exhibits these high-level functional features will, for explicable reasons, be relevantly homeostatic."

000

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

I even think folk categories can be important objects of study in the special sciences. Indeed I think probably psychology does centrally and essentially study a phenomenon the folk have latched on to, MIND, even when totally ignorant of the biological details.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

What does "no ontological import" mean? If they have members, that's good enough for me. Maybe you mean something stronger, like "be governed directly by laws" or something of that kind? But I'm not committed to anything like that; I'd be fine if 'mind' turned out to be a (very important) folk kind.

210

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Okay, but then why can't the genus be characterized in functional (folk-psychological, normative, whatever) terms? And then our actual biological realizations are not important to the category as such except epistemically.

110

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

All right, well on that interpretation, it's not relevant to our discussion, it seems?

130

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Could you characterize the argument of this paper to me? I've looked at it before but I wasn't entirely sure what it really was when I thought about it.

110

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

Well, as far as I can tell, you take them to fail _for the reason that_ they don't appeal to biological specifics. That's what I don't understand, what the argument for biology's importance as such is.

120

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

I accept the personal level isn't completely independent of its subpersonal instantiation (not sure what 'sui generis' means here, but I of course accept supervenience), but deny your conclusion still.

100

Daniel Drucker @danieldrucker.bsky.social · 23/09/2026

We have a good (implicit, but perhaps even explicit) grasp of it functionally, I think, such that the physiological implementational details are unimportant, and a fortiori evolution is totally unimportant.

200

Daniel Drucker @danieldrucker.bsky.social · 22/09/2026

Sure, but I wager that's not what we'd see (or are seeing), but sure.

020

Daniel Drucker @danieldrucker.bsky.social · 22/09/2026

Don't worry about it! I tend to think what matters about personality is whether it plays the "personality role" in a system's life. If we can predict and explain a system's behavior over time in the way we normally do with personalities, and it resists change, why does origin mean it's not?

120

Daniel Drucker @danieldrucker.bsky.social · 22/09/2026

arxiv.org/html/2601.10... That's just one example, and they're quite variable, but agents of many model-types resist forced changes to personality, quite strongly. (E.g., Anthropic's Fable.) This isn't true of every language model I wouldn't think.

arxiv.org

The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models

140