@vipejuan.bsky.social
PhD student @Max Planck UCL || RL and decision-making || Trying to understand how we process (dis)information ๐ง ๐๏ธ
Honey, we fixed Signal Detection Theory (SDT)! In this preprint, Constantin Meyer-Grant, David Kellen, Sam Harding, and I critically evaluate the (unequal-variance) Gaussian SDT model in recognition memory and pursue the Gumbel-min model as a principled alternative: doi.org/10.31234/osf...
๐งต
๐จ New study alert! ๐จ
Ever wondered if rats and humans learn in the same way? ๐ญ๐งโ๐ฌ
We tested this โ and the answer is yes, at least when it comes to how we value rewards in context.
(with @shaunaparkes.bsky.social Lachlan Ferguson, Magdalena Soukupova)
๐งตThread ๐
1/
www.biorxiv.org/content/10.1...
Honest people donโt lie. Or do they? Liars arenโt honest. Or are they?
One puzzling conundrum in contemporary politics is that politicians who seem to be estranged from facts and evidence are nonetheless considered honest by their followers.
1/n
Again, a big thank you to @ranimo.bsky.social and Ray Dolan for guiding this work!
In the full paper, we go in depth into these results, and propose several mechanisms of how some of these biases can emerge, escalate and progressively bias our beliefs.
osf.io/preprints/psyaโฆ
13/13
However, you may still under-correct these news, perceive neutral sources as biased in favor of vaccines, and, when receiving factual information, revise your opinion of the source rather than your vaccine beliefs. This will make you more vaccine-skeptical over time!
12/13
So what does this mean in the real world? Imagine you frequently read anti-vax news. You know itโs biased. You think youโre reading critically.
11/13
We found that biases systematically distorts beliefs, even when:
โ๏ธBiases are non-ideological, simple and additive
โ๏ธParticipants are highly motivated to learn
โ๏ธThey have clear chances to detect/correct biases
Bias silently takes holdโeven when we're trying to resist it!
10/13
3๏ธโฃThird finding: People care for learning about the sources over getting money
Participants directed too many cognitive resources to learn how sources are biased, but this hurt their ability to make good bandit choices. Sometimes attempts to correct for biases may backfire!
9/13
2๏ธโฃSecond finding: people misperceive neutral sources as being biased.
After interacting with a biased source (e.g., favorable), a neutral source was perceived as biased in the opposite direction (e.g., unfavorable). And this only emerged after the ground truth was withheld.
8/13
So, what did we find?
1๏ธโฃFirst big finding: People don't fully correct for bias.
Even when theyโve had ample opportunity to learn that a source is biased, they still under-debiased. Participants became biased in the same directions as the sources that informed them!
7/13
In phase 2, these feedback sources can be treated like our "biased weight scale".
By adding/subtracting 3ยฃ to estimates of unfavorable/favorable sources respectively one can fully correct for their reports and learn the true value of paintings!
6/13
The task had two phases:
๐ขPhase 1: true outcomes and source feedback were shown, so that could learn about source biases.
๐ Phase 2: only source feedback was shown (no true outcomes), so they had to infer the values of paintings.
We also asked them to classify the bias of each source.
5/13
Instead, they relied on external sources that estimated the selling price of selected paintings. But these sources could give biased estimates:
โFavorable sources overestimated true selling prices by ~3$.
โซNeutral sources (unbiased) โUnfavorable sources underestimated by ~3$
5/13
We tested this using a multi-armed bandit reinforcement learning game where participants played art dealers selling painting copies (=bandits).๐ผ๏ธ Paintings varied in price.
The goal: to choose more expensive paintings.
The challenge: they didnโt get to see the TRUE prices
4/13
Even more interesting, bias is theoretically correctable!
Imagine a scale that always adds 5kg. If the scale reads 75kg, you can infer your true weight is 70 kg. So, in principle, if we know an info-source is biased, we should be able to adjust for it. Right?
Not quiteโฆ
3/13
First, bias is not noise.
โขNoise is like a coin flipโrandom and directionless.
โขBias is systematicโit consistently skews things in a certain direction.
And here's the kicker: while noise cancels out over time, bias can accumulate. 2/13
๐จ New preprint alert! ๐จ
w/ @ranimo.bsky.social ๐ osf.io/preprints/psyaโฆ
From partisan news to algorithmically curated content, we constantly receive biased misinformation. With biased input, can our beliefs be accurate?
Turns out, biased misinformation distorts our beliefs! ๐๐งต 1/13
โญ๏ธPhD in Cognitive/Computational Psychologyโญ๏ธ Use Reinforcement Learning to study how mis/misinformation affects us. For full funding, one has to be eligible for UK home fees. Please Share!!
@queenmarycbb.bsky.social
Deadline: April 20. For more information:
www.findaphd.com/phds/project...
When populist regimes target scientific institutions - as is happening in the US today - it is not because their core constituency is anti-science but exactly because even they respect the authority of science.
Science is a dangerous counter-power for the populist leaders.
(2/4)
We know that economic anxiety & conspiracy beliefs are related. Often this is used to argue that it is key to fix economic conditions to avoid widespread conspiracy beliefs.
But a new study shows that causality runs the other way. The conspiracy beliefs drive the anxiety: doi.org/10.1111/pops...
Last year, we published a paper showing that AI models can "debunk" conspiracy theories via personalized conversations. That paper raised a major question: WHY are the human<>AI convos so effective? In a new working paper, we have some answers.
TLDR: facts
osf.io/preprints/ps...
