Interference alignment is a novel technique that aligns interference in each receiver using half of the signal space and leaving the other half for the desired signal, independent of the number of users that share the channel. Most of the research focuses on the channel's capacity, degrees of freedom or throughput. In this paper we work with the error probability. For this purpose, we consider the pairwise error probability using a particle swarm optimization to optimize the precoders of a three-user interference channel with interference alignment. Simulations show that using this technique improves the performance of the three users when optimized jointly by at least 15dB SNR at a 10-1 BER for the first user compared to the one proposed by Cadambe and Jafar, and at least 5dB SNR at a 10-2 BER when optimizing the precoder for user one only compared to user one when the three users are optimized jointly.