Experts versus algorithms

I’ve been revisiting the post from 2 November 2013 about algorithms, drawing impetus from the book by Daniel Kahneman Thinking fast and slow. I titled my post Why experts are better than algorithms (post #168).

I’ve just listened to the post recited slowly and with gravitas by the synthetic voice of John Rhys Davies (attached). My take on the theme hasn’t changed. But the social context is different.

It seems that “algorithm” has become a by-word for the worse aspects of social media. Algorithms have become culprits in gathering up people’s online interactions to deliver content and advertising that reinforces biases and prejudices, filtering out contrary views.

More insidiously, would-be influencers recruit such algorithms to change the social landscape and even influence political outcomes. Today, “algorithm” circulates publicly as a shorthand for opaque systems of recommendation, ranking, and behavioural nudging, often operating at scale and with minimal accountability. (ChatGPT crafted that sentence.)

In other writings I’ve sought to rehabilitate “algorithm” from this automatic association with manipulation. After all, any computer operation, from presenting pixels on a display screen to searching the internet, is algorithmic.

In a more recent post, I discussed Kahneman’s writing in the context of AI. See recent post (April 2025): Can AI only think fast? In so far as we can ascribe “thinking” to these AIs, the answer is primarily “yes” in Kahneman’s terms. In that post I also considered how groups think, discuss and act, invoking the idea that thought is communal, social, cultural and bound up with language — before it is individual.

Scoring systems

ChatGPT helped me further summarise. At the time, Kahneman was largely concerned with relatively modest decision procedures — recruitment, diagnosis, profitability, student examinations — served by simple heuristics, scoring systems, and rule-based protocols designed to counter well-documented cognitive biases.

The kind of algorithms attracting Kahneman’s attention were transparent, deliberately reductive, and explicitly under human design and control. Scoring job applicants on a simple attribute matrix comes to mind. On the other hand, contemporary platform algorithms are adaptive, data-hungry, and frequently inscrutable even to their designers.

Platform algorithms

Kahneman’s scoring algorithms are meant to restrain hasty decision making; platform algorithms often amplify the haste, accentuating unchallenged emotional responses, arbitrary preferences, outrage, and confirmation bias in accelerated loops.

Aspects of Kahneman’s basic account of human judgement seems to fit the current social media context, even as the technologies have undergone transformation, especially with the wide use of AI.

We are still overconfident, still prone to narrative coherence, still inclined to mistake familiarity for understanding. What has changed is that these tendencies are now entangled with large-scale socio-technical systems that operate continuously, automatically, and commercially.

I argued that expertise is not simply a matter of outperforming procedures on narrowly defined tasks. It is bound up with responsibility, interpretive labour, and participation in shared practices.

Collective reasoning

I think Kahneman missed some important aspects of decision making. As I outlined in my post, experts do not merely decide; they explain, justify, listen, and revise. They operate within institutional, ethical, and conversational contexts that algorithms, however sophisticated, don’t process in the same way.

In this respect, the older debate between clinical judgement and statistical prediction, rehearsed by Meehl and popularised by Kahneman, now sits within a broader ecology of collective reasoning. Decisions are rarely isolated events. They unfold over time, across groups, infrastructures, and feedback loops. The apparent superiority of an algorithm at a single decision point tells us little about how meaning, trust, and legitimacy are sustained in practice.

If anything, the present moment reinforces a conclusion already implicit in the 2013 post: that optimisation is a thin proxy for judgement, and that solidarity, accountability, and interpretive openness remain central to how societies make decisions under conditions of uncertainty. Algorithms may be indispensable, but they do not replace expertise so much as redistribute it — often unevenly — across technical, institutional, and social domains

I think these last two paragraphs from ChatGPT accurately restate my sentiments in the 1913 blog. Dare I say, they also exceed them. I couldn’t have put it better!

Reference

• Kahneman, Daniel. Thinking, Fast and Slow. London: Penguin, 2011. 
• Meehl, Paul E. Clinical vs. Statistical Prediction: A Theoretical Analysis and a Review of the Evidence. University of Minnesota Press, 1954. 

Note

• Featured image is by ChatGPT: Please generate an image of some defunct measuring instruments in a postapocalyptic clinical setting.