Griffin Pion

philosophy of ai competence should require mandatory “date added” field

In another sense, it is actually extremely fitting to describe it as "feeling like" talking to a Ph.D., because if ChatGPT is good at anything, it's generating text that *sounds* very convincing to someone with no knowledge. The danger is *seeming* knowledgable without any understanding!

it’s not a nous paper, but it’s honest work

seconded. this one is especially great if you have any interest in P → Q

this rationale seems to miss the fact that reviewing can actually be a valuable experience for the reviewer, not just some thankless service work. i’ve enjoyed my handful of reviewing experiences!

Thanks also to the editors of the volume, Neil Van Leeuwen and Tania Lombrozo (@tanialombrozo.bsky.social), as well as everyone who has given us comments.

We are excited to post this work and see what people think (and, therefore, believe…).

philpapers.org/rec/PIOBWW

(10/10)

And Eric is an endless well of arguments, knowledge and patience. (I can’t imagine what it’s like to have two grad students nitpick details from my more than 10-year-old dissertation!)
(9/10)

Many thanks to my co-authors! Elliot is largely responsible for clarifying the possible underlying mechanisms for Spinozanism (and now has awesome single-author work on the interaction between negation and belief acquisition).
(8/10)

4. We extend the model from its traditional domain, belief acquisition, to belief updating and action.
(7/10)

3. We draw out some sociopolitical implications of having a Spinozan mind, particularly for fake news and misinformation.
(6/10)

2. In addition to presenting classic evidence, we bolster the model with recent developmental work from Vikram Jaswal and theoretical work from Henry Schiller and Shaun Nichols.
(5/10)

Our paper goes beyond previous work on the Spinozan view – e.g., by Dan Gilbert and Eric Mandelbaum – in four ways:
1. We survey multiple possible mechanisms that could underlie the Spinozan model, characterizing the interactions of acceptance, rejection, and endorsement.
(4/10)

In a new paper forthcoming in The Oxford Handbook of the Cognitive Science of Belief, Elliot Schwartz (@elliotschwartz.bsky.social), Eric Mandelbaum (@ericman.bsky.social), and I present an updated defense and extension of the Spinozan model.
(3/10)

Figures showing each view of belief acquisition: the Cartesian Model and the Spinozan Model.

However, according to the Spinozan model, merely tokening a thought entails believing it. In other words, you believe everything you think. Only through a further, effortful process can one reject these automatically-accepted beliefs.
(2/10)

Griffin Pion, Elliot Schwartz & Eric Mandelbaum, Believe What We Think!: The Spinozan Theory of Mind - PhilPapers How do we acquire beliefs? According to the Spinozan model, merely having a thought entails believing it. Only through a further, effortful process can one reject automatically-accepted beliefs. This ...

📄 NEW PRE-PRINT! 📄

How do we acquire beliefs? According to one intuitive view, we can entertain thoughts, and then choose to accept them as beliefs or reject them.

philpapers.org/rec/PIOBWW

(🧵1/10)

Happy to discuss this more! DM or email me if you'd like.

In this load condition, we also do not see your prediction borne out.

In this condition, we still see the "Uncommon Sense Effect" (compare the 2 figures below). The advantage of looking at this condition is that since we are holding premise/conclusion reading time constant, the RTs are a better indicator of how long responding alone takes.

We also ran load conditions (not reported in CogSci) that might give us cleaner results. One of these load conditions involved time-pressure: participants were shown each premise/conclusion for 1500ms.

One complication here is that this RT data is for a self-paced condition, where the RT value is the total time (a) reading each premise and conclusion and (b) responding.

A simple t-test shows that the difference between the polysemy and valid filler RTs on "valid" response trials is not significant (p-value = 0.3847).

So, restated, the prediction is that P RT > valid filler RT for trials on which participants respond “valid”. Here’s the figure. The left ("0") side are those on which participants respond "invalid"; the right ("1") side are "valid". y-axis is RT.

Instead, the relevant comparison is between P RTs and valid RTs for trials on which participants answered positively (i.e., "valid").

If we consider this in conjunction with the fact that there's a difference in how often people judge each sort of target word as valid (our "Uncommon Sense Effect"), we can’t simply compare the RTs for Ps vs. valid fillers across all trials.

Thanks for reading and for suggesting this prediction! The short answer is that this prediction doesn't look to be borne out. Here's a longer answer:

I should say this project was inspired by @quiltydunn.bsky.social's amazing paper on concepts and polysemy (doi.org/10.1111/mila...)! If you haven't already, go read it!

We're thrilled to present these findings this summer at @cogscisociety.bsky.social 2025 and @socphilpsych.bsky.social!

Many thanks to my wonderful co-authors (especially my co-first-author, Elliot Schwartz). Putting this project together has been, and continues to be, a lot of fun. (9/9)

A bar graph showing that valid filler arguments are deemed valid more often than equivocal polysemy arguments, polysemy arguments are deemed valid more often than equivocal homonym arguments, and homonym arguments and invalid filler arguments are deemed valid at comparable rates.

📊 THE RESULTS! 📊
We found that prediction (2) was clearly borne out, supporting the view that polysemes are not represented like homonyms, and polysemes’ underspecified representations can be used in reasoning. (8/9)

2. However, if polysemes, unlike homonyms, involve a representation shared across senses, and this representation can be used in reasoning, then participants should judge equivocal polysemy arguments as valid significantly more often than they do for equivocal homonymy arguments. (7/9)