Correlations: Voting Between AI Models Doesn't Ensure Diversity
Correlated errors between AI models can negate the value of a vote among systems. A mathematical formula has been used to measure this effect across six open models and 1,500 exam questions.
Multi-Model Voting Does Not Guarantee Diversity
Multiplying AI models to reach a decision by vote is not enough to obtain varied opinions. The aggregation of systems can, on the contrary, lead to a convergence of errors. This finding calls into question the effectiveness of this method for decision-making.
Similar Errors Among Models Reduce Independence
Systems that appear distinct can provide similar answers because their errors are correlated. The observed results show a diversity of opinions lower than expected. Under these conditions, asking the same question to multiple models and counting the votes may reduce the number of truly independent opinions.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
A Formula Tested on Six Open Models and 1,500 Questions
A mathematical formula has been proposed to quantify the loss of independence related to these correlations. Its application has been verified on six open AI models, using 1,500 exam questions. The approach involves querying several systems with the same question and then aggregating the responses through a vote.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.