Michelle Ramey

@michelleramey.bsky.social

Assistant professor studying episodic memory and how it interacts with visual attention, schemas, and aging (using eyetracking and computational modeling) | https://michellemramey.com/

I'm going back to in-class exams this semester, which are harder to schedule, less intellectually rigorous, slower to grade, harder to secure, more prone to error, harder to provide disability acommodations for, and use up a lot more paper than online exams. AI makes everything worse.

I assigned random gender/ethnicity labels to scientific abstracts from the literature and then asked Claude to do a thematic analysis. Claude identified a clinical versus computational split for female/male authors and a DEI focus for Black/URM authors. All in completely random data.

BildBild

The evidence is stacking up that AI use rapidly degrades cognitive ability/effort. From RCTs in >1200 people by Oxford, MIT, CMU & UCLA: AI use impairs people's future AI-unassisted performance and reduces their persistence. This happens after only 10 minutes of AI use. arxiv.org/abs/2604.04721

AI Assistance Reduces Persistence and Hurts Independent Performance

People often optimize for long-term goals in collaboration: A mentor or companion doesn't just answer questions, but also scaffolds learning, tracks progress, and prioritizes the other person's growth...

arxiv.org

What does it mean for cognition to be Bayesian? The view that cognition is Bayesian inference is often intended at Marr's computational level. In a new paper in Nat Rev Psych, @kaixue98.bsky.social and I formulate the "explicit-Bayes" hypothesis that lives at the algorithmic level. rdcu.be/ff1XA

The explicit-Bayes hypothesis for cognition

Nature Reviews Psychology - It is often asserted that human cognition is Bayesian, but that broad claim is difficult to test in a falsifiable way. We suggest that researchers specifically assess a...

rdcu.be

This -- a model bill to effectively decimate non-STEM research by requiring 3-3 teaching loads outside of "STEM or Americanism and western civilization" -- seems...extremely bad but also quite plausible.

Proposed Model Bill Would Change College Tenure, Teaching, & Research

Three conservative groups have proposed model legislation that would dramatically change faculty tenure paths, teaching loads, research activities and hiring authority

forbes.com

New paper in Psych Review on a model of false recognition in Deese-Roediger-McDermott DRM task. Not just recognition responses, but also associated RTs! And not just the semantic task, but also the structural task - where words overlap in orthography/phonology! A thread!

APA PsycNet

psycnet.apa.org

I spend so much time verifying everything now from art to cited sources to photos to historical references to citations to quotes to legal and medical information that it's really hard to fathom how much work AI has collectively added to the world, not reduced

Dr. Casey Fiesler@cfiesler.bsky.social · 8mo ago

Here's the reality this example illustrates: It's not even just about people blindly trusting what ChatGPT tells them. LLMs are poisoning the entire information ecosystem. You can't even necessarily trust that the citations in a published paper are real (or a search engine's descriptions of them).

Memory problems will change how you see the world...literally 👀 Across two new papers, we examined the eye movement patterns of younger adults, older adults, individuals with mild cognitive impairment, and amnesic cases. 1/5

New paper out! Imagery can directionally modify memory encoding, to manipulate later recognition for changed faces. Essentially, imagery can be used to simulate effects of higher (or lower) study-test similarity for an item itself. @psychonomicsociety.bsky.social link.springer.com/article/10.1...

Using visual imagery to manipulate recognition memory for faces whose appearance has changed - Cognitive Research: Principles and Implications

Real-world recognition requires our memory system to accommodate perceptual changes that occur after encoding; for example, eyewitnesses must recognize perpetrators across changes in appearance. However, it is not clear how this flexible recognition ability can be improved: Standard encoding strategies not only tend to be ineffective, but can in fact be detrimental for recognizing people across appearance changes. Given the effectiveness of visual imagery in creating and modifying memory representations, we examined whether counterfactual visual imagery could be used to manipulate flexible recognition by simulating an increase in encoding–retrieval similarity. Across two experiments, participants (n = 317) encoded faces with neutral expressions and were cued to imagine the faces with either happy or angry expressions. During later retrieval, participants saw lineups of old and new faces with either happy or angry expressions, and selected the old face and provided recognition confidence. Old/new recognition discriminability and confidence were higher when a face’s expression at retrieval matched the expression that it was imagined in during encoding (i.e., congruent imagery); interestingly, however, there was Bayesian evidence for no benefit of imagery congruence for face-choice accuracy. Moreover, congruent imagery improved recognition for old arrays irrespective of whether participants correctly selected the old face, suggesting that the imagery manipulation influenced a diffuse sense of recognition without influencing the ability to attribute that sense of recognition to a specific stimulus. Together, these findings indicate that visual imagery can directionally manipulate recognition for changed faces and produces a novel dissociation between old/new recognition and forced-choice accuracy.

link.springer.com

We make predictions based on general knowledge and/or specific memories. Different brain areas are active when these distinct predictions are violated – and hippocampus selectively responds to prediction errors based on episodic memory. Cool work by @chrismbird.bsky.social @ayab.bsky.social et al!

Hippocampal mismatch signals are based on episodic memories and not schematic knowledge | PNAS

Prediction errors drive learning by signaling mismatches between expectations and reality, but the neural systems supporting these computations rem...

pnas.org

Our new paper on how episodic memory and semantic knowledge interact to influence eye movements during search is out now in Psychonomic Bulletin & Review, with @jmhenderson.bsky.social and Andy Yonelinas! (summary below) link.springer.com/article/10.3... #psynomPBR @psychonomicsociety.bsky.social

Episodic memory and semantic knowledge interact to guide eye movements during visual search in scenes: Distinct effects of conscious and unconscious memory - Psychonomic Bulletin & Review

Episodic memory and semantic knowledge can each exert strong influences on visual attention when we search through real-world scenes. However, there is debate surrounding how they interact when both are present; specifically, results conflict as to whether memory consistently improves visual search when semantic knowledge is available to guide search. These conflicting results could be driven by distinct effects of different types of episodic memory, but this possibility has not been examined. To test this, we tracked participants’ eyes while they searched for objects in semantically congruent and incongruent locations within scenes during a study and test phase. In the test phase containing studied and new scenes, participants gave confidence-based recognition memory judgments that indexed different types of episodic memory (i.e., recollection, familiarity, unconscious memory) for the background scenes, then they searched for the target. We found that semantic knowledge consistently influenced both early and late eye movements, but the influence of memory depended on the type of memory involved. Recollection improved first saccade accuracy in terms of heading towards the target in both congruent and incongruent scenes. In contrast, unconscious memory gradually improved scanpath efficiency over the course of search, but only when semantic knowledge was relatively ineffective (i.e., incongruent scenes). Together, these findings indicate that episodic memory and semantic knowledge are rationally integrated to optimize attentional guidance, such that the most precise or effective forms of information available – which depends on the type of episodic memory available – are prioritized.

link.springer.com