Lightnews — Scholar-powered news

Jonathan F. Kominsky

@jfkominsky.bsky.social

1.4K followers 550 following 710 posts

(he/him) Assistant Professor of Cognitive Science at Central European University in Vienna, PI of the CEU Causal Cognition Lab (https://ccl.ceu.edu) #CogSci #PsychSciSky #SciSky Personal site: https://www.jfkominsky.com

ccl.ceu.edu)

Posts Media Videos Starter Packs

Pinned

Jonathan F. Kominsky @jfkominsky.bsky.social · Sep 29

#HiSciSky I study the cognitive science of causality: How we identify cause-and-effect in the world and how we think and reason about it. I'm the PI of the CEU Causal Cognition Lab in Vienna, Austria. (ccl.ceu.edu)

#CogSci #PsySci #DevSci

Causal Cognition Lab

ccl.ceu.edu

5 12 31

Reposted by Jonathan F. Kominsky

Sickos Committee @sickoscommittee.org · 3h

When the Yankees get eliminated

26 450 2.7K

Reposted by Jonathan F. Kominsky

Melanie D’Arrigo @darrigomelanie.bsky.social · 1d

I feel like this photo of masked, armed men pepper spraying a pastor protecting his community is going to be a defining picture of this moment in America for a long, long time.

1.7K 21K 55K

Reposted by Jonathan F. Kominsky

Gary Lupyan @glupyan.bsky.social · 8d

A wild exchange with Gemini Pro 2.5. In preparing some examples for my seminar (cogsci-llms.netlify.app) I remembered @randomwalker.bsky.social old tweet about playing Rock paper scissors with LLMs. Decided to see what it's like now ....

1 8 22

Reposted by Jonathan F. Kominsky

Cameron Patrick @cameronpat.bsky.social · 19d

kids these days just don’t understand their computer’s filesystem. gotta organise your files into folders, not just rely on cloud and search!

erika @yeeeerika.bsky.social · Jan 13

i don't want to hear your most boomer complaint. what's your most millennial complaint?

5 15 110

Jonathan F. Kominsky @jfkominsky.bsky.social · 20d

I think that's beyond the scope of what we looked at. There are other people who have looked at predictability and such that might be more relevant to that question

Jonathan F. Kominsky @jfkominsky.bsky.social · 21d

Wasn't the first journal we submitted to, however! It's been a long journey for this paper

Reposted by Jonathan F. Kominsky

samuel mehr @mehr.nz · 22d

Nothing. I use it for nothing at all because AI is good at zero of the tasks I do regularly

Honestly I don't even know what its web address is, is it like a 2000s style ChatGPT.com or something funkier like chat.g.pt

9 11 64

Jonathan F. Kominsky @jfkominsky.bsky.social · 21d

Reposting for the morning crowd (and with proper tags this time): My Ph.D. student's first paper, on how causal coherence affects our episodic memory.

#CogSci #PsychSciSky

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Very excited to announce my student Andreas Arslan's first paper, "Causal coherence improves episodic memory of dynamic events" in Cognition!

Out now open access: www.sciencedirect.com/science/arti...

Andreas isn't on bsky, but he very kindly wrote a summary thread for me to share.

🧵 (1/24)

Causal coherence improves episodic memory of dynamic events

“Episodes” in memory are formed by the experience of dynamic events that unfold over time. However, just because a series of events unfold sequentiall…

www.sciencedirect.com

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Epilogue: Just to say as Andreas's advisor, I'm tremendously proud of his work on this. He learned 3D modeling an animation for this project to make the stimuli, and designed brilliant experiments to confront foundational questions about causality and memory. So thrilled to share it with all of you1

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

‘For now, these findings highlight that the episodic memory system is prone to forgetting when made to serve as a vessel of disjointed sensory impressions but latches on to whatever causal patterns can be detected in the noise.’ (24/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Right now, we don’t have enough data to decide which mechanism explains our findings best – it may even be a mixture of all of them. We’ll conclude the thread by quoting the last sentence of the paper, which encapsulates its main take-away well: (23/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

In paper, we spend a lot of time discussing potential mechanisms underlying these behavioral results. We propose three kinds of processes that could account for them: a) reconstruction, b) gating, c) encoding/postdictive updating. (22/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

The results showed that people really were better at detecting lures related to coherent videos. When it came to reject fragmented lures, they actually performed at chance level! (Exp. 3 also replicated our order-related findings from Exp. 1 and 2). (21/24)

A box-plot showing accuracy on the Y axis with two causal status conditions, "coherent" and "fragmented", and two groups of items, "original" and "lure". Read the paper linked in the first post of the thread for more information.

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Basically, only half of the images participants saw were actual stills from the video (originals); the rest (lures) had been doctored in a manner that misrepresented causally relevant details. Here, the red burger and the barrel switched position. (20/24)

An image with four panels showing two "original" images (left column" and corresponding "lure" images (right column). The only difference between the original and lure images is that the two objects in the image (in both cases a barrel and a sphere) have their positions swapped.

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Finally, in Exp. 3 we investigated whether causal coherence would result in better recall of causally relevant details. We adapted our cued memory task, so that participants now not only had to report event order, but also detect lures. (19/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Exp. 2 showed that coherence also improves order recall of longer videos: In Exp. 1, we had used videos with only three segments; in Exp. 2, we also presented 5- and 7-segments-long videos. Again, causal structure improved order call. (18/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

This was assessed via a cued memory task: Participants first saw a still image from a video (the cue). Other still images followed – for each, they had to indicate whether it was taken from a part of the video that came before or after the cue. (17/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

In Experiment 1, we tested the hypothesis regarding recall of event order. It was a within-subjects study, meaning each participant saw both coherent and fragmented videos. As predicted, people were better at remembering the order of coherent events. (16/24)

A box-plot showing accuracy on the Y axis with two causal status conditions, "coherent" and "fragmented", and two answer types, "in-episode" and "non-episode". The accuracy for "coherent and in-episode" items is higher than for "fragmented and in-episode" items, but there is no difference between coherent and fragmented for non-episode items. Read the paper linked in the first post for more detailed information.

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

You’ll notice the examples so far rely on event schemas (car accidents) and common knowledge (familiarity with movies/apps). So is the influence of causal structure on episodic memory something really fundamental or not? Let’s look at the data. (15/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Similarly, to maintain a causal interpretation of an event, you have to remember some essential causally relevant details. So while you might misremember a car’s color or even exact shape, you probably won’t confuse a microcar with a monster truck. (14/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Event order depends on causal structure in a straightforward way: Causes have to precede effects. If you witness a car crash, for example, you are unlikely to ‘falsely remember’ that a windshield shattered before the vehicles made contact. (13/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

We expected that causal structure would improve two specific aspects of the episodic memory representing an event:

1) The order in which things happened

2) The causal relevant details of an event

Let’s consider them one by one.
(12/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

We should probably note that all the previous GIFs were just snippets – here’s a complete fragmented video from Experiment 1 (these videos were the shortest), in case you are curious: (11/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Coherent videos also had cuts, but they only changed the point of view (i.e., position of the camera) and did not interfere with the causal structure of the event. We hypothesized uninterrupted causal links would lead to more accurate event recall. (10/24)

Jonathan F. Kominsky @jfkominsky.bsky.social · 22d

Importantly, fragmented videos were not completely devoid of causal structure. Instead, they contained causal discontinuities: After each cut in a fragmented video, objects would abruptly be in different positions and move at different speeds. (9/24)