The Elephant in the Room - We are racing toward “super intelligence” without agreeing on what we want all that intelligence for
In September 2026, some of the people building the world's most powerful AI systems started warning publicly that the technology could become too powerful for humans to control. A senior researcher resigned from Anthropic citing existential risk. Anthropic's CEO wrote a long essay calling for the industry to pace the frontier. OpenAI's Sam Altman, DeepMind's Demis Hassabis, and even Elon Musk expressed varying degrees of alarm. And yet everyone kept building, faster than before. The question now isn't how fast we should run the race. It's whether anyone has actually defined the finish line.
About This Episode
This is a host-only episode - with Dean van Leeuwen and Graeme Codrington thinking out loud about the biggest question in technology right now. And they think almost everyone is asking the wrong one.
With some scepticism, Graeme notes that the timing of the AI safety warnings coincides neatly with Anthropic and OpenAI's IPO preparation cycles - and that the spectacle of a company saying "I have the most amazing product but it's too powerful and I don't want to sell it to you" is historically unprecedented and probably strategic. His read on what large language models actually are - auto-regressive token predictors that hallucinate rather than think - has led him to a simple conclusion: the danger isn't an AI system becoming autonomous. It's humans building over-equipped, under-monitored systems, removing all guardrails, and then claiming they couldn't have stopped it.
Listeners to the podcast may remember Dean’s experience of riding in a driverless taxi in Nashville, ending up in a standoff with a road foreman whose hand signals the algorithm couldn't interpret, proves that it’s not humans in the loop that matters - it's humans in the lead. A Waymo specialist hundreds of miles away had to take remote control to resolve it. The machine is a tool. It reaches a limit. At that limit, a human must be able to step in, redirect, and take responsibility.
The episode also explores what AI should actually be for. Is it to advance human intelligence? Not smarter machines - more capable people. Better decisions, faster learning, problems that couldn't be solved before. Or should we consider retiring the label "artificial" entirely? Should AI actually stand for Artificial Intelligence? Perhaps instead it should stand for Advancing Intelligence - machines and humans together, each contributing the type of intelligence they're best at.
In 1961, America was in a technological race with a powerful geopolitical rival, under enormous uncertainty, with rockets that could carry nuclear weapons. President Kennedy didn't get caught up in the technology debate. He defined the end game: put a man on the moon and returned him safely to earth, before the decade was out. Does the AI race need the same thing? Not a benchmark. Not a regulatory framework. Not "don't kill anyone." A vision of what ten years of extraordinary AI could make humanity capable of doing.
Meet Our Guest
3 Things You'll Take Away
1. The question isn't how fast - it's where to. The entire AI safety debate is focused on speed and risk. Almost nobody is asking what the end game actually is. Leaders who define their own AI end game - what they want AI to make their people and organisations capable of - will make far better decisions than those chasing the most powerful tool for its own sake.
2. Humans in the lead, not just in the loop. There's a critical difference. Humans in the loop means passive oversight. Humans in the lead means active responsibility - the ability and obligation to redirect the machine when it reaches the limits of its capability. And if you let it run anyway, you are responsible for what it does.
3. AI should stand for Advancing Intelligence. The reframe that matters isn't technical - it's purposive. If the goal is advancing human intelligence (more capable people, better decisions, problems solved that couldn't be solved before), the conversation about what to build, how to govern it, and who benefits becomes entirely different.
Memorable Moments
"We keep asking, will AI become intelligent enough to destroy humanity? But perhaps we should first ask, will humanity become intelligent enough to use AI wisely? That's one of the most important questions." - Dean van Leeuwen, 09:26
"Humans in the lead always, and humans taking responsibility always. And anything that's outside of those two should just be plainly illegal - because it's too dangerous." - Graeme Codrington, 20:25
"It's not what can the machine do without us. The question is: what can people, teams, and organisations now do that they could not do before because the machine exists?" - Dean van Leeuwen, 22:58
"I don't think it's AI to SI. I think it's artificial to advancing. It isn't artificial intelligence as in a machine-like human. It is advancing intelligence - with machines and humans together." - Graeme Codrington, 25:38
"Kennedy's vision is maybe the type of vision we need now. For him, it wasn't about the technology itself. It was about advancing human society. We do these things not because they are easy, but because they are hard." - Graeme Codrington, 27:21
Chapter Markers
00:26 Something extraordinary happened in September - the AI warning moment
01:30 Welcome and today's grey elephant: AI without an end game
03:21 IPO’s and why AI companies are saying what they're saying
04:04 The contradiction: warnings about danger, faster building anyway
07:14 Are we asking the wrong question?
07:58 We're having the wrong conversation about AI
08:37 Dinner with CEOs, Gordon Brown, and a question about humanity's future
09:26 The better question: will humanity be intelligent enough to use AI wisely?
10:14 The bottle of water test - AI is not autonomous
11:00 What LLMs actually are: auto-regressive token predictors, not thinking machines
11:54 Dean’s Waymo story: humans in the lead, not just in the loop
15:41 President Obama's real concern: rogue humans using AI, not AI rising up
16:35 Agents, loops, and hallucinations - what the "danger" actually is
19:40 Humans responsible, always - it should simply be illegal otherwise
21:20 AI is just like any other industry: duty of care, regulation, liability
22:06 What is the end game? Where exactly are we racing to?
22:58 Should the purpose be advancing human intelligence, not smarter machines?
25:38 From AI to SI - and why Advancing Intelligence is the better label
26:32 AI needs a moonshot end game like President Kennedy
29:06 1961 and 2026 - the parallels are extraordinary
30:02 AI's Kennedy moment: what should 10 years of AI make humanity capable of?
30:53 Coming up: Stanford neuroscientist on the human brain and AI agents
31:39 Quiz: how much spaceflight experience did America have when President Kennedy announced the moon mission?
33:28 Crazy ambitious - the right kind of crazy
Frameworks, Books & References Mentioned
Jacob Coxon - resigned from Anthropic, September 2026, warning of AI existential risk
Dario Amodei (CEO, Anthropic) - "pace the frontier" essay, September 2026
Sam Altman (CEO, OpenAI) - OpenAI IPO delayed to 2027
Demis Hassabis (CEO, DeepMind) - expressed support for pacing AI development
Gordon Brown - The Future Starts With Us (book)
Barack Obama - "non-zero chance" of AI rising against humanity; primary concern: rogue humans
Donald Trump, UN General Assembly, September 2026 - proposed renaming AI as "SI" (Super Intelligence)
Waymo - autonomous vehicle service, Nashville; the human-in-the-lead lesson
Alan Shepard - first American in space, 15-minute suborbital flight, 5 May 1961
John F. Kennedy - moon mission announcement, 25 May 1961
Stuart Brand's Pace Layers model (referenced in Ep. 20)
The seven grey elephants framework → tomorrowtoday.consulting/greyelephantsintroduction
Coming up: Stanford neuroscientist on the human brain and AI agent interaction
Related Episodes
Humans in the Lead: The Case for Advancing Intelligence, Not Just Automating It - Ep 18
AI's Power Hunger, Sovereign AI & Sand Scarcity - Lessons from South Africa - Ep. 10
The Elephant You Could Name
Ask your leadership team: what is your organisation's AI end game? Not which tools you're using, not your AI policy, not your governance framework. What do you want AI to make your people capable of doing in ten years that they cannot do today? If nobody in the room can answer that, you're running a race without a finish line.
The Seven Grey Elephants
Every episode of Elephants in the Boardroom explores one or more of the seven grey elephants - the high-impact, visible forces of change already in the room, shaping how we lead, work and learn:
-
Angry People - rising frustration, polarisation and shifting trust
-
Ageing Populations - demographic shifts reshaping economies, workforces and care
-
Angry Planet - climate, ecology and a natural world pushing back
-
Multipolarity - a fragmenting global order and the end of a single centre of power
-
Inequality - widening gaps in wealth, access and opportunity
-
The Big Squeeze - pressure on supply chains, resources, prices and capacity
-
Intelligent Advances - AI and the technologies reshaping how we think, work and decide
Learn more about the framework → tomorrowtoday.consulting/greyelephantsintroduction
The Seven Grey Elephants
Elephants in the Boardroom is the podcast from TomorrowToday Global - futures thinkers helping leaders see the grey elephants already in the room. Hosted by Dean van Leeuwen and Graeme Codrington, with guests and colleagues from around the world.
🔗 www.tomorrowtoday.consulting
This episode is proudly sponsored by Achilles - advanced risk monitoring for suppliers and third parties, helping organisations build anti-fragile, responsible and transparent supply chains. achilles.com
Subscribe & Share
If this episode helped you see something differently, the most useful thing you can do is pass it on. Forward it to one person who needs to hear it, leave us a review on Apple Podcasts, or tag us on LinkedIn with the elephant you're willing to name.
Credits:
Hosted by Dean van Leeuwen and Graeme Codrington
Sponsored by Achilles - Supply Chain Risk Management https://www.achilles.com/
Produced by Matt from It Starts with a Podcast https://www.linkedin.com/in/itsmattallen/