Artificial evolutionary intelligence (AEI): evolutionary computation evolves with large language models

Deep learning (DL) and evolutionary computation (EC), two main branches of artificial intelligence, have attracted attention in a far different way over the past decades. On the one hand, the DL area focuses on recognizing data patterns for simulating the human brain’s complex decision-making power, which has witnessed the boosting of large language models (LLMs). LLMs have shown inspiring adeptness at mastering many multimodal tasks, thanks to the big data for pre-training, the large-sized architecture for learning, and the tailored optimization strategies for fine-tuning. On the other hand, the EC community paid more attention to solving complex computational tasks, aiming to extend the global search capability of meta-heuristic methods by mimicking natural biological evolution. Nevertheless, the development of EC in real-world scenarios is far from satisfactory compared with DL; even the EC itself is seldom used in the optimization tasks within DL/LLMs. This paper provides a look at the future of EC from the perspective of artificial evolutionary intelligence (AEI), i.e., the cooperative evolution of EC and artificial general intelligence with the assistance of LLMs. A paradigm of LLM for EC has been discussed to provide some potential research topics for the interdisciplinary between optimization and learning. Specifically, three main issues of LLMs are considered for AEI, i.e., the multi-modal representation capability for encoding, reproduction, and selection, a general model for versatile learning such as surrogate, dimensionality reduction, configuration, recommendation, and generative models, and the ability to understand EC in terms of the EC concepts, EC codes, and EC behaviors. Furthermore, an open-source platform has been realized, which is expected to promote research in AEI and its applications. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2024.