We started UnlikelyAI based on a simple idea: AI has a fundamental trust problem that critically needs solving.
This is because the technology underpinning most AI — what we refer to as ‘Large Language Models’, or ‘LLMs’ for short — is inherently flawed.Yes, LLMs are powerful in many circumstances. But they’re also prone to biases, hallucinations and inaccuracies.
This is all very well for, say, casually using ChatGPT in your personal life; but apply it to a business setting and it becomes a huge issue. In regulated industries, where in many respects AI has the most potential, these flaws present insurmountable barriers to adoption.
Because of the way LLMs are designed, these problems won’t simply improve over time. In fact, large LLM developers like OpenAI admit that they’re becoming worse.
Businesses need a solution. So we set out to create it.
Tell us about the business – what it is, what it aims to achieve, who you work with, how you reach customers, etc.
UnlikelyAI offers a solution to the problems preventing businesses from exploiting AI, especially in high-stakes fields where wrong decisions have huge consequences — financially, reputationally, or legally. We want these firms to be able to benefit from the power of AI with certainty and confidence; without having to worry about the possibility of disastrous consequences, however remote
UnlikelyAI’s new form of AI, which we refer to as ‘neurosymbolic’, combines the best aspects of LLMs with a model capable of applying meaning to the information it processes. This allows it to perform tasks to a much higher degree of accuracy and with no hallucinations. Importantly, it can also explain the decisions it makes and alert users when unable to find a satisfactory answer, enabling a high degree of trust.
How has the business evolved since its launch?
We have put years of research and development into creating our groundbreaking new AI technology since we first set up UnlikelyAI, around 2019.
Over the past year, the company has undergone a steep transition towards the introduction of this product to market, ensuring that it effectively serves our clients’ actual business needs.
We’re now working with leading firms in high-stakes environments to implement UnlikelyAI’s solutions, including recently-announced pilot programmes with Lloyds Banking Group and SBS Insurance Services. We continue to work with our partners to refine our product while combining those insights with the latest work from the R&D team.
Tell us about the working culture at UnlikelyAI
Our working culture is codified in nine core principles that define who we want to be, guiding everything from what we regard as best practice to how we hire and assess performance. These are: product magic, invention and obsession; exceptional people, coaching and growth; impact and big thinking; truth and dissent; action; trust and humility; highest standards; ownership; and delivery and grit.
It’s also vital to recognise that what we’re trying to do at UnlikelyAI is exceptionally challenging. We want to push the very boundaries of what people believe is possible with AI. So we only hire absurdly ambitious people willing to do what it takes to achieve that.
Equally important is an environment where people feel confident and comfortable experimenting. We’re not bound by convention here, and rather aim to bring science-fiction to life. This requires a culture where people are dedicated to finding new ways of doing things.
How are you funded?
UnlikelyAI is a venture-funded startup. Our last round was an oversubscribed $20M seed led by Amadeus Capital Partners and Octopus Ventures.
What has been your biggest challenge so far, and how have you overcome it?
It was a technical challenge during my time at Evi Technologies. We were trying to build natural language interfaces when the world wasn’t ready for them; back then, computers relied on menus and keywords, but we were aiming for what at the time was a sci-fi vision — simply talking to machines. We stayed persistent, though, driven by a strong belief that language is the most natural interface. Fast forward to today, where we’re seeing this huge rise of large language models, and that vision is finally becoming reality.
How does UnlikelyAI answer an unmet need?
AI has the potential to radically improve the operations of organisations in regulated fields. Yet, attempts to integrate it have, so far, largely failed.
This, in essence, comes down to inherent flaws in the design of the LLMs that organisations in these fields are trying to use. They’re prone to bias, inaccuracies and outright hallucinations, which no amount of development is going to eliminate.
To exploit AI, organisations in these fields require a different type of AI — an AI that will comply with regulations. They need an AI they can trust to make the right decisions, explain said decisions and alert the user when it’s uncertain.
UnlikelyAI’s neurosymbolic AI is that AI. Using an innovative approach capable of clearly showing its workings, it’s able to facilitate adoption within regulated industries so that all organisations can exploit its benefits at scale.
What’s in store for the future?
We’re building a trustworthy AI platform designed to solve some of the most pressing problems with AI today. Issues like hallucinations are serious barriers to safe adoption in any high-stakes industry, whether that’s healthcare, insurance or finance.
Our platform is horizontal by design. What that means is — instead of being built for a single niche or industry — it’s expressly engineered to generalise across many different use cases, which makes it scalable and adaptable.
Our goal is to provide the foundational layer for responsible AI adoption at scale; an AI system that enterprises can rely on with full explainability, control, and compliance baked in from day one.
What one piece of advice would you give other founders or future founders?
Do it. Ten years from now, we’d much rather look back on this time with the knowledge that we tried to solve an extremely challenging problem, yet didn’t succeed, than to look back with the regret of never having tried at all.
And finally, a more personal question: What’s your daily routine, and what rules are you living by at the moment?
I wake up at 6:45am. Between then and 8:00am, I go through a list of health and planning activities to set me up for the day. I then exercise — either a heavy weight session, pilates, or running. — except for one rest day.
I then start my working day with a morning stand-up with our leadership team. From there I rotate through a mixture of activities — including deep-work sessions for strategy and critical decision-making; one-to-one conversations with the team to stay connected; and project and technology meetings to ensure I’m always updated on progress and blockers. And in the background, I’m keeping up with AI and tech news to stay current in what is an incredibly fast-evolving space.
As for rules, I’m alcohol-free on three days of the week at the moment, and do not eat or drink anything after 9pm.
William Tunstall-Pedoe, CEO and Founder of UnlikelyAI
