Thursday, August 14, 2025

Quality over categories

From Scientists as Experts in Public Debates Characterised by Scientific Uncertainty: The Swedish COVID-19 Debate by P. Mattsson and E.P. Vico.  From the Abstract:

This study explores how academic scientists engage as experts in public debates characterised by scientific uncertainty and societal urgency, focusing on rhetorical positioning and communicative intentions. The research centres around the debate concerning COVID-19 measures in Sweden and analyses 109 opinion pieces written by scientists in various newspapers. The analysis identifies four ideal-typical expert roles: Reformers, Advisors, Informers, and Evaluators. These roles illustrate how scientific expertise can serve multiple purposes in societal crises marked by uncertainty. Reformers take a critical stance, questioning foundational assumptions and advocating for systemic change. Advisors offer actionable recommendations in the face of uncertainty, while Informers contribute by clarifying facts and providing context; Evaluators look back to assess what has worked, guiding future improvements. The typology responds to calls for greater transparency and reflexivity among experts by illustrating the diverse ways scientists assume expert roles in public debate. Recognising the variety and complementarity of these roles and promoting awareness and openness about them can play an important role in sustaining science’s legitimacy amid uncertainty. By shedding light on scientists’ rhetorical positioning and communicative intentions, our framework supports a more structured and nuanced reflection on public engagement. Such awareness is necessary for building and maintaining public trust, particularly during times of crisis.

Four types of experts:

Reformers take a critical stance, questioning foundational assumptions and advocating for systemic change.

Advisors offer actionable recommendations in the face of uncertainty.

Informers contribute by clarifying facts and providing context.  

Evaluators look back to assess what has worked, guiding future improvements.

 I accept the categories as possibly useful.  I wonder, however, how useful the categories are, independent of the the quality of critical questions, recommendations, factual accuracy, and comprehensiveness of what has worked.  

With Covid-19 in the US, we had many in CDC, FDA, Academia, Pharmaceutical industry and in Government who putatively ought to have been experts, and certainly took on various combinations of the four categories but who were at the time demonstrably uncritical, made recommendations without justification, advanced "facts" which were known to be false, and were either ignorant of or chose to ignore what had worked in the past.  

If the "expert" is gullible, unreasoning, untrustworthy, and ignorant, the problem is not the categories of expert but the quality of expert.

Success comes from active and conscientious curation of data/information/values inputs.

A thought for later development.

The AI models function based on the Large Language Models (LLM).  From hallucinations, to clearly wrong answer, to inexplicable answers, there are a range of results from AI which are at least puzzling and sometimes alarming.  No one yet understands why these things occur.

My suspicion is that some of the issue reflects inherent errors and biases in the text ingested by the LLMs in the first place.  If a disproportionate percentage of input is from Legacy Mainstream Media, then of course there will be a pattern to errors.  If the LLM has an excess reliance on Reddit, likewise.  

My further suspicion is that the improvement of AI will necessarily come from an improvement in the data constituting the LLMs.  The sources must be curated so that the strongest and most useful only are contributors.  The lower quality are gradually filtered out.  

The further thought is that this, if true, is analogous to humans.  Independent of the inherited attributes (Gender, height, health, IQ, etc.) an individuals degree of success is largely shaped by the values and behaviors cultivated by themselves, their class, their religion, and their culture.  The combination of the three are not equally effective everywhere in the world.  The successful person curates their culture as a predicate to their future success.

For AI, effectiveness will come from better curated input into LLMs.

For individuals, it is already the case that effectiveness arises from the quality of behaviors and values is a function of selected personal, social class, religion and cultural attributes.  

Success comes from active and conscientious curation of data/information/values inputs.  

History

 

An Insight

 

I see wonderful things

 

Offbeat Humor

Data Talks

 

Bouquet de Fleurs by Gustaf Ericsson

Bouquet de Fleurs by Gustaf Ericsson (Sweden, 1846 - 1920)






















Click to enlarge.

Wednesday, August 13, 2025

History

 

An Insight