CRT does not only critique. It also asks: what would justice look like? What would AI systems designed with racial justice as a central commitment actually look like, in practice, not just in principle?

This is a genuinely difficult question, and CRT scholars offer different answers. But several commitments emerge consistently.

Moving beyond bias mitigation to structural change. Technical bias mitigation, adjusting algorithms to produce more equal outputs, is necessary but insufficient. It addresses the symptoms of structural inequality without addressing the structure. Racially just AI requires attending to the social conditions that produce biased data, which means attending to housing policy, education policy, healthcare policy, criminal justice policy, the full range of structural factors that produce the racially unequal world that AI systems learn from.

Reparative approaches. Some CRT scholars argue that because AI systems have encoded and amplified historical discrimination, their reform requires not merely neutrality but repair, positive steps to remedy accumulated harm. This might mean designing AI systems that actively work to correct historical inequities rather than merely not reproducing them.

Community ownership and control. If AI systems built by and for dominant groups systematically disadvantage marginalised communities, then racially just AI requires shifting who owns and controls those systems. Community-owned data trusts, public AI infrastructure, participatory governance mechanisms, these are structural alternatives to the current concentration of AI power.

Transparency that is genuinely accessible. Technical transparency, publishing model cards, algorithmic audits, impact assessments, is valuable only if the communities most affected can access and interpret it. Racially just AI transparency is designed for affected communities, not for technical specialists.

Accountability with teeth. The communities most harmed by AI systems currently have the fewest mechanisms for seeking redress. Racially just AI governance creates meaningful accountability, not just regulatory compliance, but genuine remedy when AI systems cause harm.

These are ambitious commitments. CRT does not promise that they will be easily achieved in societies built on racial hierarchy. But it insists that the question must be asked, and that asking it changes what we see when we look at AI systems, and what we demand from the people who build them.

One mind cannot see it all. But every mind that AI systems fail to see represents a failure of imagination, of data, and of justice that we can choose, if we choose, to remedy.