iiinova — where your digital self keeps evolving

Cameron Buckner @cameronbuckner.bsky.social · 01/09/2026

It's on. Thanks to Greg for serious poster game.

040

Cameron Buckner @cameronbuckner.bsky.social · 11/08/2026

I found the perfect pedagogical use of this meme...

131

Cameron Buckner @cameronbuckner.bsky.social · 31/07/2026

Heading out for sea for a few days without email contact. Don't do the Singularity while I'm gone.

020

Cameron Buckner @cameronbuckner.bsky.social · 30/07/2026

Communing with a corvid at Skilak Lake

040

Cameron Buckner @cameronbuckner.bsky.social · 30/07/2026

Sometimes you just need to go to Alaska and run on the beach.

050

Cameron Buckner @cameronbuckner.bsky.social · 21/07/2026

Philosophy editors: Both NeurIPS and ICML have adopted a "dual policy" approach to AI-generated submissions. You can submit/review under either a conservative (AI banned) or a liberal policy. They find almost no difference in rates of acceptance or reviewer ratings across policies.

120

Cameron Buckner @cameronbuckner.bsky.social · 07/07/2026

I did not pay DE Shaw & Co to advertise my #ICML talk but I'm grateful to them.

052

Cameron Buckner @cameronbuckner.bsky.social · 24/06/2026

Very pleased to be doing a session on my book at Pusan National University next month.

040

Cameron Buckner @cameronbuckner.bsky.social · 11/06/2026

This is one of the funniest AI-search snafus (snorfus?) I've seen yet: they've accused some random nice lady from Carmel of being the kimu-eating beast from Pietroski's classic thought experiment...

060

Cameron Buckner @cameronbuckner.bsky.social · 09/05/2026

Interesting background on the event at Berkeley... artshumanities.berkeley.edu/news/promine...

120

Cameron Buckner @cameronbuckner.bsky.social · 29/09/2025

Phillip Kieval and I have a new commentary on Binz et al.'s Centaur model and its role in discovering a new theory of human cognition: psycnet.apa.org/doiLanding?d...

031

Cameron Buckner @cameronbuckner.bsky.social · 20/09/2025

I finally found an adversarial example for humans in the wild!

020

Cameron Buckner @cameronbuckner.bsky.social · 28/05/2025

I think we should both avoid being convinced by breathless AI hype to adopt unsafe tools for things that probably shouldn't be automated in the first place, and also avoid falling for biased takedowns that romanticize human decision-making and minimize real advantages current AI tools can provide.

070

Cameron Buckner @cameronbuckner.bsky.social · 09/05/2025

Multi-modal Chain-of-Thought reasoning is already a big thing. That is, CoT models can "reason" not only with chains of text self-prompting, but also by producing images, audio, and more, and engaging with the stimuli they self-produce in a variety of ways. arxiv.org/abs/2503.12605

0155

Cameron Buckner @cameronbuckner.bsky.social · 16/04/2025

Heading to the Pacific APA in San Francisco for an Author Meets Critics session! Couldn't ask for better commentators. Also some international book sightings in the wild (or citings?)

070

Cameron Buckner @cameronbuckner.bsky.social · 08/03/2025

This "Carl" guy is gonna hear from my lawyers...

050

Cameron Buckner @cameronbuckner.bsky.social · 19/02/2025

Some news that isn't terrible: Looking forward to this event in South Korea next week:

060

Cameron Buckner @cameronbuckner.bsky.social · 09/01/2025

Truth be told, it's always me passing around the Non-Euclidean geometry...

2110

Cameron Buckner @cameronbuckner.bsky.social · 13/12/2024

Most important lecture of the year: heading to my daughter's school to talk to a bunch of sixth graders about philosophy of artificial intelligence. And I have the perfect shirt picked out already...

3401

Cameron Buckner @cameronbuckner.bsky.social · 10/12/2024

Wow, this is basically my position. I guess at least there will always be Marcus to disagree with? (Lots of respect for Chollet's work...but of course it's not the *optimal* way to do reasoning. That's why we invented formal systems, as a cultural technology!)

1163

Cameron Buckner @cameronbuckner.bsky.social · 05/12/2024

Dangit didn't they use the higher resolution graphic?

020

Cameron Buckner @cameronbuckner.bsky.social · 02/12/2024

The grads are memeing my paper advice again...

030

Cameron Buckner @cameronbuckner.bsky.social · 17/11/2024

At airport after PSA, so long and thanks for all the PhiSci! Came in exhausted and left rejuvenated by the quality of the talks and conversation, especially by ECRs (jazz and Cajun food helps too). A+++ would recommend.

050

Cameron Buckner @cameronbuckner.bsky.social · 07/11/2024

Honored to be giving the keynote at the Society for Exact Philosophy tomorrow; my talk will focus on the prospects of using large language models like the GPT-series and Claude as models of human reasoning, a topic which has much been in the news lately. meta.phil.ufl.edu/host/sep/mee...

2101

Cameron Buckner @cameronbuckner.bsky.social · 16/09/2024

In this article we explore arguments against Tulving's traditional episodic/semantic memory distinction, provide them on a firm metaphysical foundation by drawing upon the theory of natural kinds and previous things I've said about a phenomenon called "transitional gradation."

110

Cameron Buckner @cameronbuckner.bsky.social · 02/04/2024

If you want to hear a discussion over some of the ethical implications of my book, see this discussion with Reid Blackman on the Ethical Machines podcast. Thanks again for the great discussion, Reid! www.reidblackman.com/podcast/ais-...

141

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

The remaining chapters of the book explore how this basic capacity for perceptual abstraction in the face of nuisance variation might be bootstrapped to greater degrees of efficiency, flexibility, and control by layering on more domain-general modules simulating other faculties.

010

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

Even if this is all accepted, the systems still fall short of modeling human visual perception. They are less sample efficient, vulnerable to adversarial examples, and lack many other forms of flexibility highlighted in the Tiers of Rationality described in Chapter 2.

020

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

This explains the distinctive power of deep convolutional neural networks: they are orders of magnitude more efficient on performing abstract categorizations plagued by nuisance variation. And they do so by capturing key mechanistic aspects of mammalian visual cortex.

000

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

Deeper networks can discover deeper abstractions by controlling for more nuisance variation, bc computations performed at earlier layers can be recursively redeployed an exponential number of times in later computations, relative to the network's depth compass.onlinelibrary.wiley.com/doi/abs/10.1...

000

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

DCNNs can resolve this dispute by showing how the same mechanism can deploy different methods of abstraction in a division-of-labor. Linear convolution nodes play the role of abstraction-as-composition, and non-linear pooling nodes perform roles of abstraction-as-subtraction.

000

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

To solve this problem, Locke and other empiricists proposed that we possess several different innate methods of abstraction. Many philosophical battles were fought in the attempt to discover the "right" process of abstraction amongst the several options.

100

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

In fact, this way of construing the problem is very similar to the one faced by early modern empiricists. Locke and others fretted endlessly about how we might distinguish abstract similarities in categories like "triangle" when we have only seen idiosyncratic particulars.

500

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

We can see the task of both biological and artificial neural networks as one of transforming our sense of subjective similarity. By the end of the transformation, misleading nuisance similarities should be downplayed, and subtle distinguishing features accentuated.

100

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

We must "control for" nuisance variables in order to recognize abstract categories. DiCarlo and co-authors have compared the problem to pieces of paper crumpled together (by nuisance variation) into a ball; they must be uncrumpled for easy classification. cell.com/trends/cogni...

110

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

Models inspired by Neocognitron--we now call them Deep Convolutional Neural Networks--were designed to capture how perceptual cortex controls for "nuisance variables". These variables complicate perception because they introduce variation that doesn't track meaningful categories.

100

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

Of course, as I emphasized in Ch 1 and 2, empiricists are free to deploy domain-general structure in their theories of these faculties' operations. Indeed, the first DNNs were inspired by neuroscientific research on ventral stream research by Hubel & Wiesel.

100

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

Indeed, perception is the starting point for all empiricist theories of the mind, as it centers their fundamental constraint: the raw materials for constructing all ideas must come from the structure of perceptual input data.

110

Cameron Buckner @cameronbuckner.bsky.social · 11/02/2024

Ch. 3 Summary: Perception Given arguments of Chs 1 and 2, we should model human cognitive faculties by designing DNN architectures around core mechanistic aspects that explain how brain circuits perform those faculty's functions. We begin with the most basic faculty, perception.

110

Cameron Buckner @cameronbuckner.bsky.social · 24/01/2024

Very excited to discuss my book this Friday in NYC, especially with other disciplines associated with the NYU Digital Theory Lab, Center for Mind, Brain and Consciousness, and Center for Data Science. Includes a response from Ryan Healey nyuad.nyu.edu/en/events/20...

0104