This paper challenges the label of AI hallucinations – arguing instead that these falsehoods better represent bullshit. That is, bullshit, in the Frankfurtian sense (‘On Bullshit’ published in 2005), the models are “in an important way indifferent to the truth of their outputs”. This isn’t BS in the sense of junk data or analysis, but… Continue reading ChatGPT is bullshit
Tag: artificial-intelligence
ChatGPT in complex adaptive healthcare systems: embrace with caution
This discussion paper explored the introduction of AI systems into healthcare. It covers A LOT of ground, so just a few extracts. Extracts: · “This article advocates an ‘embrace with caution’ stance, calling for reflexive governance, heightened ethical oversight, and a nuanced appreciation of systemic complexity to harness generative AI’s benefits while preserving the integrity of… Continue reading ChatGPT in complex adaptive healthcare systems: embrace with caution
An Empirical Study of the Anchoring Effect in LLMs: Existence, Mechanism, and Potential Mitigations
This study found that several LLMs are fairly easily influenced by anchoring effects, consistent with human anchoring bias. Extracts: · “Although LLMs surpass humans in standard benchmarks, their psychological traits remain understudied despite their growing importance” · “The anchoring effect is a ubiquitous cognitive bias (Furnham and Boo, 2011) and influences decisions in many fields” · “Under uncertainty,… Continue reading An Empirical Study of the Anchoring Effect in LLMs: Existence, Mechanism, and Potential Mitigations
LLMs Are Not Reliable Human Proxies to Study Affordances in Data Visualizations
This was pretty interesting – it compared GPT-4o to people in extracting takeaways from visualised data. They were also interested in how well the LLM could simulate human respondents/responses. Note that the researchers are primarily interested in whether the GPT-4o model acts as a suitable proxy for human responses – they recognise there are other… Continue reading LLMs Are Not Reliable Human Proxies to Study Affordances in Data Visualizations
Leverage points to intervene in a system – Donella Meadows
In the lead up to next week’s compendium on systems thinking, here’s a banger from Donella Meadows. She explores system leverage points. Not a summary, but some extracts: · Leverage points are “places within a complex system .. where a small shift in one thing can produce big changes in everything” · The “state of the system”… Continue reading Leverage points to intervene in a system – Donella Meadows
Avoiding ‘second victims’ in healthcare: what support do staff want for coping with patient safety incidents, what do they get and is it effective? A systematic review
This systematic review evaluated evidence for what support staff want vs what they receive, and whether the support is effective. 99 studies were included. Some extracts: · PSI (patient safety incident) lead to emotional shame, guilt, anger, shock, depression, fear, flashbacks, helplessness, fatigue, withdrawal and more · The three most desired support types staff want before and… Continue reading Avoiding ‘second victims’ in healthcare: what support do staff want for coping with patient safety incidents, what do they get and is it effective? A systematic review
ChatGPT for analysing investigations
I think this is one of the better uses of LLMs regarding investigations – they trained their model to evaluate accident reports and extract key details from the reports. They found: · It could extract key information from unstructured data and “significantly reduce the manual effort involved in accident investigation report analysis and enhance the overall… Continue reading ChatGPT for analysing investigations
Human factor analysis of cockpit work incidents in high-speed workboats: the mystery hidden between the lines
This study unpacked what investigators look at and how they construct causes in high-speed workboats. It employed a Safety-II / HOP / HF perspective. Tl;dr: human factors are poorly evaluated and largely seen as individual-level factors. Some extracts: · “Although the analysis focused on negative observations, it also identified HFs that supported the activity” · “Many pivotal… Continue reading Human factor analysis of cockpit work incidents in high-speed workboats: the mystery hidden between the lines
The mixed blessing of risk defences and redundancy: James Reason
A few random extracts from James Reason’s timelessly awesome Managing the Risks of Organizational accidents. (Note: This isn’t an endorsement of the somewhat linearity of defences-in-depth, since we have evidence that emergent behaviour can playout in reality and with equifinality etc) There’s hundreds of things I could extract (and maybe will in time), but here’s… Continue reading The mixed blessing of risk defences and redundancy: James Reason
“I think, therefore I err”: An article about ‘good errors’, heuristics and intelligent systems
“Every intelligent system makes errors”, so said Gerd Gigerenzer. Here’s a couple of page extracts from a 2005 paper. Not sure if I’ll summarise it or not (it’s really interesting, but tough to capture in a summary…) The paper: · Challenges the rationalistic and normative ideal as cognition as purely a logical and rational one, ignoring… Continue reading “I think, therefore I err”: An article about ‘good errors’, heuristics and intelligent systems
SafetyInsights.org 