After a decade in which progress in AI has meant progress in capability, what now decides whether these systems are allowed to make high-stakes decisions is not how much they can do, but whether they can be trusted. That shift is the starting point of a new two-part essay series from our founder and CEO, William Tunstall-Pedoe.

The essays argue that the trust problem is not a transitional phase that better models will resolve, but a consequence of the architecture itself: scaling today’s systems reduces the problem without ever eliminating it.

William’s journey in AI is a long one: he founded Evi (originally True Knowledge), the AI company acquired by Amazon, and then worked within Amazon with his team building Alexa. With UnlikelyAI, he is taking on what he argues is now the central blocker in AI: trust. The essays set out that argument in two parts.

The first, The Trust Ceiling, explores why trust, rather than raw capability, has become the constraint that decides what AI is actually allowed to do, and why so much value stays out of production as a result. The second, Breaking Through AI’s Trust Ceiling, sets out the architectural approach he believes can move the field beyond it: keeping neural networks for what they do best, and moving the decisions that demand certainty into a symbolic domain where reasoning can be exact, consistent and explainable.

That approach is not offered as theory alone. It is the conviction UnlikelyAI was built on and has been developing for years, since before ChatGPT. It now underpins the work we do in the industries where being wrong carries a real cost, across finance, insurance, accountancy and regulation. The essays make the reasoning behind that work public, and open it to scrutiny.

That scrutiny is the point. William’s interest is in finding where the argument holds and where it does not, and he would welcome the conversation - comments are open on both essays.

Read The Trust Ceiling and Breaking Through AI’s Trust Ceiling, and follow us on LinkedIn to hear more from the team.