How GenAI Mentoring Can Support Underserved Students in STEM

This paper examines the potential of Generative Artificial Intelligence (GenAI) mentoring systems to address persistent underrepresentation of women, Black, Latino, and Indigenous students in STEM fields. Despite national efforts to diversify the STEM workforce, traditional mentoring approaches face significant constraints, including time limitations, scalability issues, and limited availability of mentors from similar backgrounds for underrepresented minorities. Drawing on attachment theory and social cognitive theory, we argue that AI-enhanced mentoring can complement human mentorship by providing 24/7 accessibility, personalized guidance, and cost-effective scalability. The analysis reviews existing GenAI mentoring platforms, including MentorAI and Thunai, demonstrating their capabilities in delivering personalized career guidance and overcoming geographical and resource barriers. Evidence from studies using Google Gemini as an AI mentor shows positive outcomes for underrepresented STEM graduate students, including improved time efficiency and empowering career guidance. However, our examination reveals critical challenges that must be addressed: algorithmic bias that can reinforce gender and racial stereotypes, a lack of emotional intelligence compared to human mentors, and insufficient transparency in AI decision-making processes. We identify significant gaps in AI literacy among both mentors and mentees, particularly concerning prompt engineering, bias recognition, and critical evaluation of AI outputs. For underrepresented populations who may have limited technology access, these literacy gaps pose additional barriers to equitable engagement with AI mentoring systems. The paper emphasizes that GenAI tools should augment rather than replace human mentors, as authentic mentoring relationships require emotional depth, cultural sensitivity, and psychosocial support that current AI systems cannot adequately provide. Our recommendations include comprehensive AI literacy training focused on prompt engineering and bias recognition, the development of explainable AI systems tailored specifically for mentoring contexts, and the creation of culturally responsive AI designs. We call for longitudinal research examining the long-term impacts of AI-enhanced mentoring across diverse cultural contexts and emphasize the need for transparent, domain-specific GenAI tools developed in collaboration with mentoring and education experts. Keywords: generative AI, STEM mentoring, underrepresented minorities, algorithmic bias, AI literacy, educational equity, artificial intelligence in education