Can algorithms replace human discussion in group decisions?
Simple decision-making algorithms cannot replace actual human interaction when group members have different levels of skill.
Source
Does interaction matter? Testing whether a confidence heuristic can replace interaction in collective decision-making
Inspect source passages
Can algorithms replace human discussion in group decisions?
These passages occur verbatim in the stored paper text. They support the points listed below; they do not verify every statement in the explanation.
256 trials
Each session consisted of 8 blocks of 16 trials (128 trials in each session and 256 trials in total).
What is the "two-heads-better-than-one" (2HBT1) effect?
There is a growing interest in the mechanisms underlying the “two-heads-better-than-one” (2HBT1) effect, which refers to the ability of dyads to make more accurate decisions than either of their members (e.g., ).
What they did
Researchers analyzed data from fifty-eight healthy male participants, paired into 29 dyads, who performed a visual target detection task. Across 256 trials in total, participants made individual judgments and rated their confidence before resolving disagreements either verbally (15 dyads) or non-verbally (14 dyads). The researchers then tested whether computer algorithms selecting either the more confident or the faster partner's decision could match or exceed the accuracy of the actual human partnerships.
What they found
The confidence-based algorithm only improved group performance when both partners had similar perceptual skills. When partners differed in skill, the algorithm suffered from a collective loss, whereas real-world interacting pairs performed significantly better by learning to discount the opinions of confident but inaccurate partners. Additionally, selecting decisions based on speed was highly inefficient, as response times proved to be a noisy and unreliable substitute for confidence ratings.
The limits
What it doesn't show
The study only evaluated healthy young male participants with a mean age of 23.5 years who were already friends, meaning the results might not generalise to diverse groups or strangers. Because participants received immediate feedback on every decision, they could easily judge their partner's credibility, a luxury not always present in real-world environments. Finally, the findings are limited to basic sensory judgments and may not apply to complex, knowledge-based tasks like medical or financial decisions.
Key terms
- confidence heuristic
- The psychological bias of evaluating the reliability of a statement or decision based on how confidently it is expressed.
- Maximum Confidence Slating
- An algorithm that automatically selects the decision made by the partner who reports higher confidence.
- metacognitive accuracy
- An individual's ability to accurately evaluate and judge the reliability of their own decisions.
- Minimum Reaction Time Slating
- An algorithm that automatically selects the decision of the partner who responds the fastest.
- two-heads-better-than-one effect
- The phenomenon where a collaborative pair or group makes more accurate decisions than its best-performing individual member.
- normalisation
- A statistical adjustment that shifts confidence ratings so different individuals share the same average score and variance, eliminating personal biases.
Flashcards
0 of 35 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
What is the 'two-heads-better-than-one' (2HBT1) effect?
Common questions
Why does the confidence algorithm fail when partners have unequal skills?
The algorithm blindly trusts the partner with higher confidence. If one partner is much less skilled but highly overconfident, the algorithm will repeatedly select their incorrect decisions, whereas human partners eventually learn to ignore their overconfident teammate.
Can response speed be used as a shortcut for confidence?
No. Although faster decisions are generally associated with higher confidence, using reaction times to guide group choices resulted in major accuracy losses because speed is highly variable and easily corrupted by noise.
What did the researchers discover about normalizing confidence scores?
Normalizing scores (leveling out individual scale biases) helped pairs with equal skills because it aligned their confidence metrics. However, for unequal pairs, normalization removed valuable information about actual skill differences, reducing decision accuracy.
Does verbal discussion make a difference compared to non-verbal confidence sharing?
Surprisingly, the study found no significant differences in decision quality between dyads who could verbally debate their choices and those who only saw each other's confidence ratings.
More on Decision making