A new hypothesis challenges the notion of AI's impenetrable 'black box,' potentially simplifying AI safety. South African experts weigh in on the unverified yet promising findings.