Tech Life The problem with AI
9 snips
Apr 7, 2026 Guest
Various Field Reporters and Interviewees (Clara Saliba, Dr Sterling McKinnon, Laura Kress, Ellie Gibson, Kirsty Rigdon, Zaira Chatou)
Guest
William Tunstall-Pedoe
Clara Saliba, AI/data analyst at Blenheim staff; Dr Sterling McKinnon, heritage project manager; Laura Kress, tech reporter; Ellie Gibson, gamer and streamer; Kirsty Rigdon, Power Wash Simulator lead; Zaira Chatou, AI bias consultant; William Tunstall-Pedoe, founder of Unlikely AI. They discuss AI trust and neurosymbolic fixes. Digital twinning for palace restoration. The calming appeal and community of mundane simulator games. Gender gaps and bias in AI adoption.
AI Snips
Chapters
Transcript
Episode notes
Neurosymbolic AI Bridges Power And Trust
- Neurosymbolic AI blends statistical ML with traditional symbolic rules to gain capabilities and explainability.
- William Tunstall-Pedoe explains this lets systems use LLM strengths while providing reliable, auditable answers for high-stakes industries.
Statistical Models Can't Always Admit Uncertainty
- Pure machine learning models always make statistical guesses and lack inherent mechanisms to say 'I don't know'.
- Tunstall-Pedoe contrasts that with symbolic systems which can explicitly indicate uncertainty or inability to calculate.
Match AI Use Cases To Tolerable Error Rates
- Set realistic expectations about AI accuracy and choose applications that can tolerate some error.
- William Tunstall-Pedoe advises businesses to assess whether a use case can accept non-100% performance and plan accordingly.