iiinova — where your digital self keeps evolving

Together, these results suggest that human randomization ability is much stronger than previously acknowledged, as long as it’s tested in the right contexts! This is just one piece of a larger set of ideas we’ve been exploring recently on random behavior; more to come!

230

In three additional studies, we probed this process even more, showing that participants flexibly (and optimally) incorporated information about the reward structure and/or their opponent's knowledge state into their behavior.

results from Studies 4–6, demonstrating optimal random behavior across task and opponent information.

In other words, the presence of a strategic opponent pushed people to produce very strong random behavior — in many cases, indistinguishable from uniform randomness. This is surprising, and contrary to what we might expect given the persistent failures in human randomization!

Across three studies, we found evidence for a 'strategic gradient' in randomization: In the game with a live opponent, behavior was near-random; in a closely matched game against a computer opponent, random behavior worsened; and in the non-strategic task (‘be random’), behavior worsened further.

results of Studies 1–3. random behavior improves along a strategic gradient.

perceptionstudies.github.io

The idea here is that, in general, the generator must behave randomly to win, otherwise the guesser can exploit their tendencies. This game actually works for any two people playing online, so you can even play it yourself against a friend! See here: perceptionstudies.github.io/strategic_ra...

Near-random behavior in strategic games

To examine this possibility, we built a custom, online game of Battleship on a 3x3 grid. Then, we compared behavior in the ‘standard’ context (e.g. ‘be random’) to behavior in a live, dyadic game with another opponent where randomness was instrumental.

schematic depiction of our tasks. randomness holds instrumental value in these tasks, allowing us to compare random behavior there to behavior in format-identical non-strategic tasks.

Here, we asked: Are people bad randomizers in general, or only in some contexts? Most experiments simply order participants to produce (say) random sequences of numbers. This looks nothing like how we use randomness in our daily lives, so maybe we’ve been inadvertently setting people up to fail.

Randomness underlies many key aspects of the world and of our minds, from anti-coordination to decision-making. Yet when we try to generate random sequences, we fall prey to several biases (e.g., over-alternation). This is quite puzzling; why are we so bad at this very important thing?

130

Chaz Firestone @chazfirestone.bsky.social · 15/09/2026

We often say that humans struggle to produce random sequences… but what if that's not the whole story? In a new paper, @dorsaamir.bsky.social, @samiyousif.bsky.social & I show that context matters (a lot!). When playing strategic games, humans can produce near-random behavior osf.io/preprints/ps...

17828

Reposted by tal boger

Out today! www.nature.com/articles/s41...

06714

Reposted by tal boger

Dorsa Amir @dorsaamir.bsky.social · 14/09/2026

Hey folks, I'll be considering PhD applications for the 2026-27 cycle over here at Duke Psych. To learn more about the lab, check us out at www.mindandculturelab.com and/or read through this short FAQ for prospective students: www.mindandculturelab.com/faq

05027

Reposted by tal boger

mahwishkittur.bsky.social @mahwishkittur.bsky.social · 10/09/2026

After years of analyzing thousands of drawings from hundreds of adults and children, I am excited to share my first ever paper (w/ @samiyousif.bsky.social )! Using a simple drawing task, we show how topological structure is represented in the mind and its importance in spatial memory.

elifesciences.org

Topological structure in human spatial representation revealed through drawing

16024

Reposted by tal boger

Chaz Firestone @chazfirestone.bsky.social · 04/09/2026

here at @jhu.edu we believe so strongly in collaboration that @talboger.bsky.social gave his latest internal department talk about work he did outside of my lab, our department, and our university! (lucky you @samiyousif.bsky.social @dorsaamir.bsky.social)

tal’s title slide: “randomness revisited”

a slide with names and faces of tal’s collaborators (sami yousif and dorsa amir), with the text “(sorry chaz)” beneath it

1202

Reposted by tal boger

Chaz Firestone @chazfirestone.bsky.social · 26/08/2026

Out today! elifesciences.org/articles/111... @elife.bsky.social

0274

Reposted by tal boger

Sami Yousif @samiyousif.bsky.social · 26/08/2026

Excited about a new paper (w @russellepstein.bsky.social): "Cognitive maps as a medium for thought", in @cp-trendscognsci.bsky.social! We discuss how maps are an important 'middle ground' rep. medium, and how they may interact with a 'language of thought'. www.sciencedirect.com/science/arti...

sciencedirect.com

Cognitive maps as a medium for thought

Cognitive maps are essential for navigation, but to what extent are they involved in thinking more broadly? In this opinion article, we examine whethe…

29528

In the commentary (philpapers.org/archive/BOGC...), I draw out some implications of this issue (including a problem of reverse inference) and discuss a potential solution (focusing on representational format). Looking forward to reading the other commentaries (+ D&S’s responses)!

philpapers.org

030

In lossless compression, a minimal code facilitates perfect recovery by using rich, structured symbols. This mode of representation — prevalent in both minds and machines — crosscuts the target article’s capacity regimes, revealing a challenge for the framework.

140

The target article proposes a few capacity regimes that vary in (a) their ability to memorize their past experiences and (b) the amount of representational resources they use to do so. These regimes are supposed to be “distinct”, but…

Figure 1 from Dubova & Sloman's target article, providing a schematic depiction of different capacity regimes.

Dorsa Amir @dorsaamir.bsky.social · 06/07/2026

BBS recently published a great piece by @mdubova.bsky.social & @sabinasloman.bsky.social on “Excess capacity learning”. I argue D&S’s framework overlooks a special kind of representation that encodes information both minimally and perfectly: lossless compression. philpapers.org/archive/BOGC...

Figure for my BBS commentary ("Compression without compromise") on Dubova & Sloman's target article ("Excess capacity learning")

13011

Reposted by tal boger

When we consider children in cultural evolution, we tend to cast them as recipients of adult culture. In our BBS target article, @sheinalew.bsky.social & I argued there's more: through 'peer cultures', children actively drive cultural adaptation. Our final paper + a rich set of responses is out now!

cambridge.org

Children as agents of cultural adaptation | Behavioral and Brain Sciences | Cambridge Core

Children as agents of cultural adaptation - Volume 49

412242

Reposted by tal boger

Chaz Firestone @chazfirestone.bsky.social · 18/06/2026

@talboger.bsky.social’s poster-coaster experience lands in Baltimore this afternoon! Come see him present his work at the @socphilpsych.bsky.social at this evening’s poster session, then grab a coaster as a souvenir. Available while supplies last!

031

Come find me at SPP (at my poster on June 18) to hear more, or even just to grab a coaster! Demo page here: www.perceptionresearch.org/anagrams_ani...

Coasters of our anagram stimuli, including for real-world size (right).

062

*(An open question is whether some high-level property other than animacy — e.g., natural vs manmade — might explain our results. Our experiments don’t rule out that possibility; but even then it would be a high-level property driving performance, which is the main motivation for our project.)

Together, this work provides evidence for visual processing of animacy itself*, beyond lower-level correlates like shape or curvature. This adds to the growing use and appeal of “visual anagrams” as a tool for high-level perception (see: bsky.app/profile/talb...).

100

But when we rotate the anagrams to vary their animacy, we also vary their orientation. Might that explain our results? We conducted the present/absent task again, but with silhouetted versions of the anagrams. The effects disappeared here — so orientation alone seems not to explain this effect.

The search effects disappeared with silhouetted version of the anagrams, suggesting that such effects do not merely arise due to differences in orientation (perhaps the only lower-level feature that varies with animacy in our stimuli).

100

We also replicated this result with a more typical present/absent search task. As before, participants were faster to determine that a target was present when it differed from distractors in its animacy. Thus, this suggests genuine sensitivity to animacy in visual search!

Another set of visual search tasks, this time present/absent (instead of location search). We find that animacy guides search here, too, replicating our prior results.

100

Next up: visual search. We tested this in several ways because of the many paradigms and results in the literature. We found that, in a location-search task, people were faster at locating targets when they differed from distractors in their animacy (e.g. animate target among inanimate distractors).

Results from our location search task. Participants were faster to locate a target when it differed in animacy from its distractors.

100

First, we asked whether changes in animacy are encoded in visual working memory. People did a simple retention + change detection task. Changes that altered animacy (e.g., rabbit→boot) were more detectable than changes that did not (e.g., rabbit→dog), even though animacy was task-irrelevant.

Results from our visual working memory task. Participants were better at detecting changes when those changes altered animacy than when they did not.

100

Then, we used these stimuli to test effects of perceived animacy on visual memory and visual attention. If the anagrams produce stable effects, then we can be confident that animacy itself (and not its lower-level correlates) guides these visual processes.

100

Using this technique, we generated images that differ in animacy subject to orientation. So, they hold nearly all lower-level features constant across the two interpretations — because they’re all the same pixels, just rotated!

These stimuli differ in animacy while holding nearly all lower-level features constant. This is because the animate objects and the inanimate ones are the same exact images, just rotated!

This challenge may seem impossible to overcome. But, as in our previous work on real-world size (bsky.app/profile/talb...) we used a new technique from Geng et al. called “visual anagrams”, which allows you to generate images whose interpretations vary as a function of orientation.

A schematic depiction of the "visual anagrams" model (Geng et al., 2024).

In fact, many findings about perceived animacy have been contested in precisely this way. For example, some studies find that animacy drives visual search, but others find that these effects reduce to differences in curvature between animates and inanimates.

100

Animacy is among the most studied high-level visual properties, with influential effects on attention, memory, and more. But objects that differ in animacy also differ in shape, curvature, and other low-level stimulus features. How can we be sure these effects reflect animacy itself?

The mind seems to encode differences in animacy; but objects that differ in animacy also differ in shape, curvature, texture, and more.

130

Gabe Waterhouse @gabrielwaterhouse.bsky.social · 03/06/2026

Do we perceive animacy itself, beyond its lower-level visual correlates? In @elife.bsky.social, @chazfirestone.bsky.social and I leverage “visual anagrams” — images whose interpretations change with orientation — to suggest the answer is: yes! elifesciences.org/reviewed-pre...

47622

Reposted by tal boger

When you look at a complex scene, you perceive various ensembles — sets of items of various sizes, numbers, etc. Are some more salient than others? In a new paper (w/ samiyousif.bsky.social), we provide an answer to this question! osf.io/preprints/ps...

osf.io

OSF

22012

Reposted by tal boger

Roman Feiman @romanfeiman.bsky.social · 26/05/2026

Excited to have been wrong about this! A few years ago, Tal told me he was working on intuitions about complexity. I knee-jerkily said, “but surely that’s not a natural kind! Why think that people have intuitions about complexity across domains?” Good thing this is the last time I’ll ever be wrong!

1202

You can find the paper here — osf.io/preprints/ps... — and you even try your hand at some of our tasks @ www.perceptionresearch.org/complexity_t....

osf.io

OSF

Together, these results implicate a type-independent representation of complexity, in ways that make contact with several bigger-picture issues in cognitive science, philosophy (e.g., the language of thought hypothesis) and computer science (e.g., universal compression algorithms).

151

Finally, we asked whether higher-level judgments (e.g., aesthetic beauty) across different stimuli may arise from unified representations of complexity, in a way that grounds individual differences. If you like highly jagged shapes, do you also like melodies with many notes? Yes, you do!

Results from our aesthetic preference task. Participants who like complex shapes also like complex dot-arrays, melodies, and so on.

150

Next a modified Stroop task found that representations of complexity in one domain automatically interfered with representations of complexity in another. When told to ignore the shapes and only judge the complexity of the words, participants did better when shape and word complexity were congruent.

Results from our modified Stroop task. Shape complexity automatically interfered with judgments of word complexity.

130

This pattern even extended to tactile forms. Participants who learned to associate reward with the feel of a jagged (vs. smooth) shape spontaneously transferred that association to a diversely colored (vs. uniformly colored) visual array.

A participant completing the tactile-visual transfer. Having learned that more jagged-feeling shapes are rewarded, participants inferred that more diversely colored dot-arrays were also rewarded.

This transfer went beyond a simple matching of ‘more’ to ‘more’ along dimensions (e.g., more turns in a shape to more notes in a melody), because we did not observe transfer to more brightness, more saturation, or greater size — suggesting complexity itself is what’s being transferred.

First, reward transfer: After learning to associate a greater reward with, for example, complex (or simple) shapes, participants spontaneously inferred that complex (or simple) melodies, dot-arrays, mathematical expressions, and so on would be rewarded too.

Results from our reward transfer tasks. After learning to associate rewards with complex (or simple) stimuli of one domain, participants spontaneously inferred that this reward rule applied to a totally different domain.

160

We put this hypothesis to the test by studying how representations of complexity transfer across different modalities or domains. We explored such ‘complexity transfer’ in three different kinds of paradigms, spanning 11 experiments.

When we study complexity in psychology, we seem to study several disjointed concepts — booleans for concept learning, skeletons for shapes, etc. However, foundational ideas from philosophy and computer science give clues that complexity may be cognitively ‘unified’ across domains.