I recently sat down with Temiloluwa Gbadebo, a leading expert in artificial intelligence, to discuss proposed regulations for AI development and protecting innovation in the field. Temiloluwa Gbadebo is the author of “Designing Machine Learning Systems at Scale: Architecting Robust and Scalable AI Solutions” and the founder of the data accelerator initiative Techlifta.
To start, I asked Temiloluwa Gbadebo about the challenges policymakers face in regulating a technology as complex and fast-moving as AI. He explained, “AI systems are extremely diverse, with different capabilities and applications across industries. Creating a ‘one-size fits-all’ regulatory approach would likely stall innovation. The key is striking the right balance between providing guidance to steer development responsibly while allowing room for ongoing progress.”
I also asked Temiloluwa Gbadebo about the risks of overregulation, given the diversity of AI systems and the pace of advancement. “Policymakers walk a delicate line between unleashing innovation and preventing harm, requiring regulatory humility,” He noted. “Prescriptive mandates broadly applied risk severely chilling R&D investments under the banner of precaution when evidence is lacking on dangers relative to benefits.” He advocated that oversight concentrate only on functionality proven high-risk today, like autonomous weapons, not unfounded speculation
One set of guidelines gaining traction is the EU’s Artificial Intelligence Act. I asked Temiloluwa’s opinion on these proposed rules for high-risk AI systems. “The EU’s risk-based approach has merit in focusing regulatory oversight on AI applications posing significant public harms,” he commented. “Mandating risk management for uses of AI in areas like critical infrastructure makes sense. However, policymakers must not define ‘high-risk’ overly broadly at the expense of progress on AI for social good.”
Given AI’s complexity, how can policymakers scope effective oversight? Temiloluwa Gbadebo discussed a measured, context-specific approach assessing functionality by intended use and risk factors. “For example, predictive healthcare algorithms require different appraisal metrics than self-driving vehicles or stock trading systems – one standard fails to capture relevant variances.” He also called for concentrating governance only where risks demonstrably outweigh benefits rather than theoretical possibilities of harm, citing autonomous weapons as the most straightforward case warranting bounds today.
Shifting to industry, I questioned self-regulation’s promise and pitfalls. Temiloluwa Gbadebo noted information sharing and coordination through technical standards can greatly support trust and accountability. But cautioned that “Voluntary efforts demand regulatory backstops – independent auditing provides invaluable signals on gaps for issues like algorithmic bias with due diligence requirements.”
Given global outlook, I asked if any AI policies worldwide offered valuable models. Mr Temiloluwa Gbadebo highlighted Singapore’s voluntary certification scheme for determining if AI systems meet safety and performance benchmarks as one creative approach. “These alternative regulatory tools provide guidance without the costs of rigid mandates applied too early,” he commented. “But there are some companies following our unique path to develop AI for social progress and sustainable growth.”
I deeply appreciate Temiloluwa Gbadebo lending his expertise on the questions before us in steering AI’s marvellous potential to benefit humanity. Our discussion spotlighted that while governance innovations must keep pace with technology transformations, we have promising frameworks to enable equitable, inclusive progress if stakeholders come together. The global challenges ahead demand nothing less than our best collective thinking.
