The original message passing detection (MPD) algorithm, which exploits channel hardening theory, achieves very good performance in large-scale multi-input multi-output (MIMO) systems. However, the complexity of the MPD algorithm grows rapidly when the number of users or the order of modulation increases, presenting a great implementation challenge for practical massive MIMO systems. In this letter, based on a novel approximate probability updating (APU) scheme, a low complexity MPD (LCMPD) algorithm is proposed. The developed APU scheme considers only the most reliable constellation point during the message passing process. Combined with a serial message updating schedule, no exponential operation or division is required in the LCMPD algorithm. Compared to the existing MPD algorithm, the overall computational complexity is reduced by 11.2 to 29.4 times under various antenna configurations with almost no degradation in the bit error rate performance. Hence, the proposed algorithm is very suitable for efficient hardware implementation.