A novel interacted collective intelligence (ICI) algorithm was developed in this work for maximum power point tracking (MPPT) of centralized thermoelectric generation (TEG) systems at non-uniform temperature gradients (NTG), upon which waste heat recovery ability can be considerably enhanced to enhance resources utilization efficiency. As centralized TEG system often exists numerous local maximum power points (LMPPs) under NTG, ICI is adopted to effectively seek global maximum power point (GMPP), upon which energy exploitation and utilization can be efficiently improved. To achieve a higher searching efficiency, a sub-optimizer with the best solution from all sub-optimizers is dynamically selected during each iteration to guide others. Although multiple sub-optimizers based ICI might lead to a higher computation complexity, a wider global search and more stable convergence can be achieved compared with single meta-heuristic algorithm. Four case studies, e.g., start-up test, step variation of temperature, stochastic temperature variation, and sensitivity analysis, validate the effectiveness and superiority of ICI. Simulation results indicate that ICI based MPPT can produce the largest energy with minimum power fluctuation under NTG compared against other five sub-optimizers, e.g., 109.24%, 112.54%, 108.61%, 107.10% and 116.75% to that of dragonfly algorithm (DA), firefly algorithm (FA), salp swarm algorithm (SSA), moth flame optimization (MFO) and multi-verse optimization (MVO) respectively in the step variation of temperature.