The use of Intelligent Machines and Artificial Intelligence (AI) has increasingly become an important aspect of human life. One of the concerns arising from their fast development is that the AI that helps humans improve efficiency may also compete with and replace humans in the labor market and social life. However, if altruism can evolve in humans, can “altruistic” AI that helps humans learn and perform be selected for and outcompete “self-interested” AI? Multi-level selection theory proposes that altruism is selected for because groups with altruistic individuals perform better than groups with only self-interested individuals. In this research, we design an agent-based model to explore the conditions where AI Helpers, rather than AI Competitors, are favored through a parallel process of group-level competition. AI Helpers with human inputs produce more variances in their learning outcomes compared to AI Competitors that replace human inputs entirely but learn more accurately. We show that in groups that use AI Helpers, the variance generated by human exploration is preserved within the group. In the long run, this variance can lead to more cumulative advancement in these groups, which drives the use of AI Helpers to be the dominant strategy across groups.