Reposted by Kate Nave
Good piece from @jburnmurdoch.ft.com & @sarahoconnorft.ft.com on AI in science. It’s making some bits faster, but there is a new ‘verification tax’ taking up time. And there are heightened bottlenecks in physical experiments & data collection. AI is a workflow problem
www.ft.com/content/4508...
ft.com
The AI Shift: Is AI supercharging science?
Productivity gains in analysis are getting stuck in physical-world bottlenecks
![Poster for this semester's reading group. We will read John Dupré's "Everyone flows: A process philosophy of human life" (2025). The reading group meets every first Wednesday of the month at 16:00 CET in room NDEF04/346 at Ruhr University Bochum as well as online via Zoom. If you're interested just join our moodle group at: https://moodle.ruhr-uni-bochum.de/enrol/index.php?id=47287, password: theory. Or just write me an e-mail: vera.straetmanns[at]rub.de](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:pknrfjwd3hv36hn7ga44sgla/bafkreibkua3cvbti2trzp76ejlomozgn3e5cbi4xuvdwppcdmtv3wcdt7m)
![Knowledge Neglect in Agent Interactions (with
Chatbots)
Ambrose Brown1[0009−0009−2726−8238], Andrew Hoskins1[0000−0002−1207−2728],
and Vaishak Belle1[0000−0001−5573−8465]
The University of Edinburgh, Edinburgh, UK
s2599421@ed.ac.uk
Abstract. In 2014, Elizabeth J. Marsh and Sharda Umananth formalized the term ‘Knowledge Neglect’ in a book chapter on processing inaccurate information. The phenomenon describes the human tendency to fail to retrieve and apply stored knowledge within a given context. This effect occurs even when reading short fictional narratives that are coherent and plausible. Our primary objective is contextualizing existing literature on knowledge neglect and the acquisition and persistence
of false beliefs within the context of interactions with chatbots. While
human knowledge neglect largely stems from cognitive biases, Large Language Model (LLM) based chatbots exhibit analogous content effects on reasoning tasks, while prioritizing semantic fluency over factual verification. Knowledge neglect may occur as an intermittent failure of individual monitoring (e.g., missing an error in a short story), but here we
conceptualize it as a structural feature within agent interactions. LLM based chatbots optimize for the very conditions that facilitate knowledge neglect, while the act of "offloading" memory to the bot reduces the reader’s motivation to engage the costly "unbelieving" process. We
provide a framework for analyzing knowledge neglect in human-chatbot
interactions, positing that this mechanism risks a dangerous feedback loop where knowledge neglect is amplified in the interaction between agents. This conceptual model is intended for researchers and designers of chatbots, as individuals increasingly rely on them for information
amid growing calls for building epistemic resilience.
Keywords: Chatbots · Knowledge-Neglect · Cognitive Science · Human-
Computer Interaction · Confabulation](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:njbnwom5rkduwaenpphfyhvi/bafkreienkogbbam5g7muuzyr5jpe6dvcr6qm3m5xneulv7fvm6wzfjqm34)
