The problem with AI

Episode Summary

Executive Summary: This Tech Life episode examines three ways technology is reshaping trust and care: AI’s accuracy and bias problems, digital twin technology preserving Blenheim Palace, and the surprising popularity of mundane job-simulation games. Experts argue AI needs more reliable, explainable systems and wider adoption to avoid bias, while Blenheim shows how data and AI can support heritage conservation, and “boring” games can meaningfully improve wellbeing.

Main Topics: AI trust, accuracy, and hallucinations (Priority: 5/5): The episode opens with concerns that chatbots and large language models can sound confident while being wrong, forcing workers and business leaders to verify outputs and limiting adoption in high-stakes settings. Neurosymbolic AI as a trust solution (Priority: 5/5): William Tunstall-Pedoe explains Unlikely AI’s approach: combine machine learning’s capabilities with symbolic software’s reliability, explainability, and ability to say 'I don't know.' Digital twins in heritage restoration (Priority: 4/5): Blenheim Palace is using a digital twin to integrate archives, sensors, digitized records, and AI to manage restoration, track damage, and plan for climate resilience. Mundane simulation games and wellbeing (Priority: 4/5): Power Wash Simulator and similar games are presented as calming, repetitive experiences that offer stress relief, with research and player testimony suggesting mental health benefits. AI bias and gender adoption gap (Priority: 5/5): Zaira Chatou argues that women are adopting AI tools at lower rates than men, which can shape outputs and reinforce bias unless adoption, trust, and inclusive culture are intentionally addressed. Satellite internet vs undersea cables (Priority: 3/5): A listener question is answered with an explanation that most international internet traffic still travels via undersea cables, which remain essential and vulnerable, even as satellite internet expands.

Key Arguments: AI is not fully trustworthy because statistical models can be wrong, opaque, and unable to indicate uncertainty; this creates risk in regulated and high-stakes industries. Neurosymbolic systems aim to blend machine learning with traditional software to improve accuracy, explainability, and reliability while retaining advanced AI capabilities. Business adoption of AI stalls when errors can cause financial loss, reputational harm, or regulatory breaches, so trust is a practical deployment barrier. AI often rewards confident-sounding answers rather than correct ones, which increases the damage caused by hallucinations and invented facts. A stronger long-term response to AI distrust is technological design, not just user caution: systems should be able to explain themselves and refuse to guess. Blenheim Palace shows that AI can support preservation by unifying building records, sensor data, and archives into a digital twin for maintenance and planning. Mundane games provide low-pressure, repetitive tasks that can reduce stress and improve mood, giving players a sense of completion and calm. AI bias is amplified when one group adopts tools less often; broader participation and non-technical voices are needed to build trust and fairness. Women’s lower AI adoption is framed as a trust and culture issue, not a skill deficit, requiring intentional inclusion and experimentation-friendly workplaces.

Data Points: Employees checking AI answers: Almost as much time as they spend using the tools - Reported in a survey for Unlikely AI, suggesting trust and verification burdens are high. Business leaders who trust AI completely: 41% - From the Unlikely AI survey cited in the opening segment. Blenheim Palace age: More than 300 years old - The palace’s age is cited in the heritage restoration segment. Annual visitors to Blenheim Palace: Nearly 1 million - Used to show the scale and public importance of the site. Roof size: An acre - The palace roof is described as massive and central to the restoration challenge. World Heritage status: UNESCO World Heritage Site - Blenheim Palace’s designation underscores the significance of the restoration. Power Wash Simulator sales: Over 17 million copies - Shows the commercial success of mundane simulation games. BAFTA nominations: 2 - Power Wash Simulator sequel received two BAFTA Game Award nominations. Oxford study participants: Over 8,600 players - Players tracked mood while using a custom build of Power Wash Simulator. Mood uplift in study: 72% - Oxford University study found most players experienced improved mood while playing. World record duration: 24 hours, 6 minutes and 33 seconds - Ellie Gibson’s record for longest time spent playing Power Wash Simulator. AI adoption gap for women: 20% to 22% lower than men - Zaira Chatou cites this as evidence of unequal adoption affecting outputs and bias.

Pivotal Quotes: "I think we've accidentally rewarded AI for sounding confident rather than being correct." — William Tunstall-Pedoe: On why chatbots can be misleading and why hallucinations are damaging to trust. "It's just so peaceful. It's just delightful." — Ellie Gibson: Describing why mundane job-simulation games like Power Wash Simulator are relaxing. "AI can scale bias fast." — Zaira Chatou: Explaining why adoption gaps and lack of inclusion matter for fairness in AI outputs.

Implications: Listeners are urged to treat AI as useful but fallible, demand explainability, and watch for bias shaped by who uses the tools. Meanwhile, data-driven preservation and calming simulation games show tech can also protect culture and support wellbeing.