Artificial intelligence (AI) technologies selection drives innovation and boosts efficiency across industries. It involves multi-criteria group decision-making (MCGDM) with evaluations from different decision-makers (DMs). Probabilistic uncertain linguistic term sets (PULTSs) help manage uncertainty and provide flexible decision-making. This paper develops an innovative MCGM method with PULTSs for AI technologies selection. First, novel definitions of subtraction, preference degree and distance of PULTSs are defined. Considering DMs’ consensus level, evaluation similarity and uncertainty degree, this paper erects a tri-objective optimization model to derive DMs’ weights. A three-layer Quantum-like Bayesian Network model is constructed. Then, an interference effect calculation approach is erected based on Kullback-Leibler divergence and Jensen-Shannon divergence, along with minimum constraints. In the first layer, the DMs’ weights are computed by the tri-objective optimization model. In the second layer, the probability of criterion is aggregated by DMs’ criteria scores and interference effects among DMs. In the third layer, the extended ELECTRE III method is utilized for calculating the conditional probabilities of alternatives on each criterion. The probability of alternative is aggregated by the normalized net credibility and interference effect among criteria. A real case of AI technologies selection combined with comparative analyses is used to illustrate the effectiveness of proposed method.