Warren Pearce
@warrenpearce.bsky.social
560 followers 220 following 97 posts
Academic. Digital methods. Climate change. Expertise. Imagery. Occasional music posts. #AcademicSky #STS https://sheffield.ac.uk/spir/people/academic/warren-pearce
Posts Media Videos Starter Packs
Pinned
warrenpearce.bsky.social
How has computer vision changed Google's knowledge infrastructure? 🤔

*Extremely* happy that our pre-print now up at SocArXiv. Our amazing team dig into Google Images, the #AI technology driving it, and the impacts for users.

osf.io/preprints/so...

#STS #digitalmethods @digitalmethods.net
Screen shot of web page reading: 
From authority to similarity: how Google transformed its knowledge infrastructure using computer vision
Authors
Warren Pearce, Maud Borie, Laura Bruschi, Daniele Dell'Orto, Matthew Hanchard, Elena Pilipets, Alessandro Quets, and Zijing Xu
Data visualisation showing the ranking of Google Images results for climate change in Australia, Brazil, China, Mexico, Netherlands and Nigeria. Some images such as 'earht in hand' and 'landscape' appear multiple times across different countries Data visualisation showing Google Images search results for biodiversity loss in Australia, Brazil, China, Mexico, Netherlands and Nigeria. As with climate change, some images appear multiple times across different countries, such as scientific charts and 'lonely animal', but there is slightly more diversity than for climate change Data visualisation showing that most search results from Google Images are different than those from Google Search. This applies for both climate change and biodiversity loss, and across all six countries.
warrenpearce.bsky.social
On way home from superb Cultural Climate Models meeting in Graz with @davidhiggins.bsky.social @caroschwegler.bsky.social et al. Looking forward to seeing in print the new approach to modelling from this truly interdisciplinary international collab

gewi.uni-graz.at/en/unsere-fo...
Entrance of Graz train station with Graz Hauptbahnhof written in capital letters on a light coloured stone building against clear blue sky
warrenpearce.bsky.social
Arguably, Wikipedia occupies a comparable role to IPCC in terms of epistemic authority (albeit via a totally different route). The Foundation‘s policy, including a strong statement of values, is a good starting point for IPCC deliberations wikimediafoundation.org/news/2025/04...
Our new AI strategy puts Wikipedia's humans first – Wikimedia Foundation
The Wikimedia Foundation's new AI strategy doubles down on the volunteers behind Wikipedia.
wikimediafoundation.org
warrenpearce.bsky.social
Excellent analysis of the dilemmas presented to the IPCC by AI! The question of whether there should be an IPCC chatbot is crucial. Given close control over communication products in the past, an “official” chatbot would signal a v different attitude to risk.
hollyjeanbuck.bsky.social
How can the #IPCC navigate generative AI?

What does it mean for scientific assessment more broadly?

New working paper looks at scenarios for AI adoption & resistance, w/ @dralaaclimate.bsky.social @shinasayama.bsky.social and Oliver Geden

Reflections welcome!

www.swp-berlin.org/publications...
www.swp-berlin.org
Reposted by Warren Pearce
dralaaclimate.bsky.social
New working paper out: "Four scenarios for an @ipcc.bsky.social navigating Artificial Intelligence"

Led by @hollyjeanbuck.bsky.social, with @shinasayama.bsky.social and Oliver Geden

Thread 🧵

1/13
.
Reposted by Warren Pearce
hollyjeanbuck.bsky.social
How can the #IPCC navigate generative AI?

What does it mean for scientific assessment more broadly?

New working paper looks at scenarios for AI adoption & resistance, w/ @dralaaclimate.bsky.social @shinasayama.bsky.social and Oliver Geden

Reflections welcome!

www.swp-berlin.org/publications...
www.swp-berlin.org
Reposted by Warren Pearce
olufemiotaiwo.bsky.social
"Common decency stigmatizes people that do not participate in it—removes them from voluntary association. We indeed have to live with one another, but terms and conditions apply."

me on why Ezra Klein should be ashamed / why shame is Good Actually

www.bostonreview.net/articles/how...
How Can We Live Together? - Boston Review
Ezra Klein is wrong: shame is essential.
www.bostonreview.net
Reposted by Warren Pearce
shobitap.org
I am a new Expert Voices columnist @science.org and my first article just came out! In it, I argue that scaling technology has a dark side, particularly when it comes to alleviating inequality and injustice. And I suggest what scientists, engineers, and policymakers can do about it.
Beware the drive to scale technology
When it comes to technology, scalability—its capacity to be standardized and then distributed en masse and across contexts—is the ultimate goal. Scientists and engineers view scalability as the route ...
www.science.org
warrenpearce.bsky.social
Yes, sounds very arm-waving by Smith. Whether or not LLMs get close to panel scores, there is the usual underlying problem with accountability that we saw during 'fuck the algorithm'.

Ironically(?) IMO REF also has an accountability problem, with gnomic feedback & panel members sworn to secrecy
warrenpearce.bsky.social
Ironically, if Smith had asked a LLM chatbot about the prospect of AI replacing humans in the REF, he would have received a very sceptical answer!
warrenpearce.bsky.social
A third of survey respondents *had not heard* the argument that the richest 1% emit twice as much as carbon as the world's poorest 50% 😳

Communicating climate change as apolitical is good for some....

Data -> climatebarometer.org/climate-baro...

#climatechange
A chart showing public awareness of media narratives about climate change, specifically that the richest 1% of the world emit more than twice as much carbon as the world's poorest 50%. Approximately one-third have heard and agreed with the claim, approximately one-third have not heard the claim
warrenpearce.bsky.social
Scraping tools like @digitalmethods.net Zeeschuimer, @lescarr.bsky.social WDRA are *vital* for social media research but how widely are they used? 🤔

If you use these tools or similar in your work, please reply 'teaching' 'research' 'both' 👇🏽 or via email if preferred 🤗

#AcademicSky
Reposted by Warren Pearce
penders.bsky.social
Our data stewards have started recommending that we no longer use US-based infrastructure for #openscience practices, given the risk of (near-future) censorship, from pre-print and data hosting to preregistration and more. That includes OSF.
warrenpearce.bsky.social
Don't worry boss, it will be all exams by the time you get back
Reposted by Warren Pearce
ltothewang.bsky.social
Please read our amazing interview with @olufemiotaiwo.bsky.social in @hammerandhope.bsky.social where I share brilliant, earth-shattering, once-in-a-generation comms insights like the following:
Reposted by Warren Pearce
tomstafford.mastodon.online.ap.brid.gy
The future of AI superintelligence is already here, it is just not evenly distributed yet

#genai #llms #cogsci #chess

https://tomstafford.substack.com/p/superhuman-intelligence-already-exists
Text: superhuman intelligence already exists, and how it has affected the game of chess tells us something
Reposted by Warren Pearce
shobitap.org
On Sunday @npr.org the UK's independent reviewer of international terrorism laws said about Palestine Action, the law applies whether you are a "hairy foreigner, or you're some nice white old lady sitting in the road". I can't stop thinking about the casual racism coming from the King's Counsel.
The U.K. is arresting pro-Palestinian protestors on terror charges. They're not deterred
After adding a pro-Palestinian protest group to a list of terrorist organizations, the U.K. has been arresting hundreds of demonstrators at weekly rallies. Critics say the move endangers free speech.
www.npr.org
Reposted by Warren Pearce
beatricecointe.bsky.social
To concur, I think LLMs in their current design & use are inherently flawed because they rely on a confusion of information with text.

(& specifically with a specific type of text that mimics written human language)

But information is very much not reducible to text.
anthonymoser.com
A key difference here is that while either can be incorrect, the structure of Wikipedia *creates context* and the structure of LLMs *destroys context*

Wikipedia has linked sources and an edit history showing where information came from and who added it when

An LLM just generates text
eugenevinitsky.bsky.social
Some of the anti-AI stuff feels a bit like when people would say "don't use Wikipedia as a source." It's just like anything else, a piece of information that you weigh against multiple sources and your own understanding of its likely failure modes