Abstract: We divided innovation diffusion in AI into two situations: whether adopted(in ten years) and continuous adoption(rapid growth period, slow growth period, and adoption saturation period) from the perspective of citation. Then, we explored the causal effects of team diversity on innovation diffusion at different stages by analyzing approximately 1.7 million AI co-authored papers using causal inference methods. We discovered that: ①the national diversity of co-author teams has positive causal effect on both whether adopted and adoption and continuous adoption. It would increase the likelihood of adoption by 7.041% to 9.818% over ten years and increase the continuous adoption by 8.082% to 16.834%. ②the team diversity across disciplines and topics has a positive causal effect on adoption and it would increase the likelihood of adoption by 3.651% to 4.697%, and by 3.052% to 5.095% over ten years, respectively. ③ the team diversity across affiliation has positive causal effect on continuous adoption and it would increase the continuous adoption by 13.506% to 18.679%.

Key words: Team diversity；Artificial intelligence；Causal inference；Knowledge diffusion, Continuous adoption