95% of AI projects fail, so why do agentic AI leaders keep winning? In Episode 22 of The AI Values Podcast, Edosa Odaro and Lindley Gooden sit down with Tim Carmichael, Chief Data Officer at Chalhoub Group and a former British Army officer, for a conversation that refuses the easy version of the agentic AI story. Tim's argument is uncomfortable for anyone who has quietly assumed hybrid workforce strategy is mostly a technology problem: it's a leadership problem first, decided by servant leadership, AI risk management, and whether entry-level staff are elevated or discarded. Expect a hard look at C-suite AI literacy, board AI literacy, and why the AI project failure rate keeps climbing even as AI adoption accelerates.Tim Carmichael is Chief Data Officer and a former British Army officer who later built an airline from the ground up. He now also works as an executive coach, known across the data and AI leadership circuit for his focus on servant leadership.LinkedIn Tim Car: https://www.linkedin.com/in/tim-carmichael-link/WHAT IS COVERED► Tim breaks pitching AI initiatives into three reactions: glazy eyes, rolly eyes, or sparkly eyes. He explains why chasing the sparkly-eyed leaders first is the fastest route to agentic AI adoption that actually sticks.► "Don't throw your brain away when you're using AI." Tim's warning lands squarely on the human supervision of AI debate, and why treating agentic AI as an intern, not a replacement, keeps leaders in control.► Entry-level jobs used to build careers on repetitive, predictable work. Tim argues leaders must now deliberately elevate what entry-level jobs teach, the questioning and the nuance, instead of just automating the routine away.► Why do AI projects fail at rates as high as 95 percent? Tim pushes back on the doom-laden framing and reframes the AI project failure rate question around whether leaders understand the problem they're actually solving.► From the boardroom to the call centre, Tim walks through what board AI literacy actually requires, and why "just because you can doesn't mean you should" is the working definition of ethics behind his own AI risk management decisions.⏰ Episode timestamps00:00 - Cold Open: Which Eyes Are You Making Right Now? 01:57 - Servant Leadership Is the Root of AI Leadership 06:15 - Agentic AI Pitches: The Glazy Eyes, Rolly Eyes, Sparkly Eyes Test 08:08 - AI Leadership as Translation: The Interpreter Approach 10:42 - Hybrid Workforce: Playing to Human and Machine Strengths 15:15 - Why 95 Percent of AI Projects Still Fail 19:53 - The One Leadership Quality Agentic AI Cannot Replace 33:00 - Humans Supervised by Machines: Where AI Governance Draws the Line #TheAIValuesPodcast #AIGovernance #AIRiskManagement #ResponsibleAI #podcast──────────────────────────────────────ABOUT THE AI VALUES PODCAST:The AI Values Podcast is where leaders come to think clearly about the trade-offs behind AI adoption not just the opportunities. Hosted by Edosa Odaro (author, 'The Values of AI') and Lindley Gooden (author, 'The Future of Truth'), with weekly conversations at the intersection of AI, trust, governance, and the future of work.🎙 SUBSCRIBE to The AI Values Podcast for honest, rigorous conversations at the intersection of AI ethics, AI governance, and business leadership.◼ Find out more about the Institute ◼ Co-Host: Edosa Odaro & Lindley Gooden

AI governance means little if judgement cannot keep up. Can a boardroom trust a deepfake that only needs to create doubt, not fool anyone completely?

In Episode 21 of The AI Values Podcast, Edosa Odaro and Lindley Gooden sit down with Professor Julie Wall to ask an uncomfortable question for anyone rolling out AI at scale: if a model can predict an outcome, does that mean it should be used to decide one? Wall researches deepfake speech detection and explainable AI, and contributes to AI standards at the British Standards Institution. Her argument: responsible AI is a leadership discipline, not a compliance checkbox. It's a timely conversation. Recent research has linked heavy AI reliance to a measurable decline in critical thinking at work, and Gartner has warned that treating AI agents as uniformly governable is a fast route to enterprise failure.

Wall's episode moves through the future of work, deepfakes and public trust, algorithmic bias against neurodivergent users, and who actually carries accountability when an AI system gets it wrong.

WHAT IS COVERED:

► Why judgement beats automation. Routine work is already automating: research, drafting, first-pass code. Wall's own words: "Our judgement becomes important. That will become the core value."

► Deepfakes do not need to convince you. They need to make you doubt. Wall points to a real case: a US school principal deepfaked into "saying" comments he never made.

► Accuracy is not accountability. Wall built an AI that detects credibility in voice patterns, and it worked. "Just because something can be predicted doesn't mean that it should be used."

► Bias hides inside "typical" behaviour. A tool built for a general cohort reads gaze aversion as disengagement. For neurodivergent students, it means the opposite.

► The one question every board should ask. Not whether the AI is accurate enough. Wall's answer: organisations should be prepared to use it responsibly.

⏰ Episode Timestamps

00:00 - AI Safety Begins With The Deepfake Judgement Question

05:41 - Responsible AI At Work: Judgement Becomes The Core Skill

14:05 - AI Ethics And Critical Thinking: Asking The Right Question

21:22 - Deepfakes Don't Need To Fool You, Just Create Uncertainty

25:16 - AI Governance And Accountability: Who Owns The Decision

27:22 - Algorithmic Bias And Neurodivergent Users: Measuring vs Understanding 32:26 - The Board's One Must-Ask Question On Responsible AI

34:24 - Why AI Literacy Makes The Future Look Brighter

#TheAIValuesPodcast #AIGovernance #AIRiskManagement #ResponsibleAI #podcast

─────────────ABOUT THE AI VALUES PODCAST:

The AI Values Podcast is where leaders come to think clearly about the trade-offs behind AI adoption not just the opportunities. Hosted by Edosa Odaro (author, 'The Values of AI') and Lindley Gooden (author, 'The Future of Truth'), with weekly conversations at the intersection of AI, trust, governance, and the future of work.

🎙 SUBSCRIBE to The AI Values Podcast for honest, rigorous conversations at the intersection of AI ethics, AI governance, and business leadership.

◼ Find out more about the Institute

◼ Co-Hosts: Edosa Odaro & Lindley Gooden

AI governance is usually treated as a policy problem. Mat Mallett, who has led technology at board level from the UK Space Agency to a cancer charity he founded himself, says the real problem is that even he can't control generative AI once it's out in the world.In this episode of The AI Values Podcast, Edosa Odaro and Lindley Gooden sit down with Mat Mallett, a board-level technology leader whose career runs from the UK Space Agency to CTO/COO of Cancer Central, the AI health tech charity he founded, where he also led development of Scotland's NHS 111 app, for a conversation about what actually happens when AI moves from theory into the boardroom. Mat's argument is uncomfortable for anyone treating generative AI as a solved problem: these systems are still teenagers, prone to hallucinating with total confidence, and the only real safeguard is human in the loop oversight, clear AI compliance expectations, and holding software vendors to account. Along the way the conversation covers AI-powered cyberattacks, deepfakes eroding trust on social media, what happens to personal data once AI can map it at scale, and why boardroom critical thinking matters more than ever as agentic AI edges toward autonomy.◼ Guest LinkedInWHAT IS COVERED: ► Mat Mallett says today's generative AI models are still teenagers. "They think they know everything… but they hallucinate badly," he warns, and explains why over-reliance on their outputs is his biggest AI safety concern right now.► AI-powered cyberattacks and deepfakes on social media: Mat on the "whack-a-mole" battle hackers are now winning with AI, and why AI-generated fakes are quietly reshaping public opinion.► What happens to your personal data once AI can map it at global scale. Mat explains the upside of AI-personalised recommendations, and the less obvious cost: a shrinking diversity of thought as algorithms cluster people who already think alike.► "We can't control generative AI," Mat says, and it's exactly why human in the loop oversight and software vendor accountability matter more as agentic AI edges toward full autonomy. Guardrails, he argues, look different for every business.► Inside the boardroom: how Mat reads the room when a board paper doesn't feel right, why critical thinking matters more than technical fluency, and his closing take on why quantum computing, not generative AI, is what genuinely excites him next.⏰ EPISODE TIMESTAMPS:00:00 - AI Governance in the Boardroom: Why We Can't Control Generative AI03:38 - AI Ethics From the UK Space Agency to Cancer Central05:01 - AI Safety Risk: Generative AI Hallucinates Like a Teenager06:20 - AI Deepfakes and Cyberattacks: The Governance Blind Spot14:08 - Data Privacy in a Fully Digital World 17:07 - Agentic AI vs Generative AI: Why Vendors Must Be Held to Account 19:44 - Responsible AI Leadership: Critical Thinking in the Boardroom 25:23 - Quantum Computing: The Next AI Breakthrough Mat Mallett Is Chasing #TheAIValuesPodcast #AIGovernance #ResponsibleAI #podcast #AgenticAI──────────────────────────────────────ABOUT THE AI VALUES PODCAST:The AI Values Podcast is where leaders come to think clearly about the trade-offs behind AI adoption not just the opportunities. Hosted by Edosa Odaro (author, 'The Values of AI') and Lindley Gooden (author, 'The Future of Truth'), with weekly conversations at the intersection of AI, trust, governance, and the future of work.🎙 SUBSCRIBE to The AI Values Podcast for honest, rigorous conversations at the intersection of AI ethics, AI governance, and business leadership.◼ Find out more about the Institute ◼ Co-Hosts: Edosa Odaro & Lindley Gooden