Process parameters of rotating velocity, welding speed, Zn interlayer thickness, and ultrasound power are optimized by the hybrid of back propagation neural network (BPNN) and gray wolf optimization algorithm (GWOA) to obtain a high‐quality Zn‐added ultrasound‐assisted friction stir lap welding joint of 7075‐T6 Al/AZ31B Mg dissimilar alloys. The results state that the prediction accuracy of the trained BPNN model is acceptable. The optimal process parameters combination is obtained by the GWOA which is combined with the trained BPNN. The verification tests are performed under the executable optimal solution, which consists of the rotating velocity of 1054 rpm, the welding speed of 54 mm min−1, the Zn interlayer thickness of 0.05 mm, and the ultrasound power of 1568 W. The tensile shear load of the joint reaches 9.05 kN, and the strength is 11.8% larger than that of the reported optimal joint. The artificial intelligence optimization method of GWOA combined with BPNN can accurately predict and optimize the joint strength, which has great time and economic advantages.