Are We Building AI on Shaky Foundations?

What if the path we’re on, the relentless pursuit of bigger, more powerful neural networks, is a blind alley? What if the very foundation of the current AI revolution, with its vast energy consumption and probabilistic nature, is not the only way, or even the best way, forward? It’s a question that challenges the prevailing narrative, a narrative that has captivated the world and fuelled a technological gold rush. But in a small corner of the UK, a team of researchers is quietly pioneering a different path, one that harks back to the very origins of artificial intelligence and offers a glimpse of a more efficient, more logical, and perhaps more trustworthy future.

At the heart of this story is Noel Hurley, a veteran of the British technology scene. His journey is a testament to the cyclical nature of innovation, a career that has taken him from the heart of a fledgling startup that would grow into a global titan, to the helm of a new venture that seeks to redefine the very essence of AI. In a recent conversation on the Project Flux podcast, Noel shared his story, a narrative that is as much about personal conviction as it is about technological disruption.

Noel’s career began at ARM, the Cambridge-based company that would go on to design the chips that power the vast majority of the world’s smartphones. He joined when it was a small team of just 40 people, a time when the American and Japanese giants dominated the semiconductor industry. It was a period defined by a fierce competitive spirit and a clear, audacious goal: to become the number one risk processor company in the world. As Noel recounted, this clarity of purpose, championed by leaders like Sir Robin Saxby, was the driving force that propelled ARM from a small British upstart to a global powerhouse.

After a decade and a half, Noel left to co-found his own startup, a venture that, like many, faced the harsh realities of the 2008 financial crisis. But the pull of his former employer remained strong. In a twist of fate, a former boss called with an irresistible proposition: “We’re going to take on Intel in servers. Do you want to come back for the fight?” It was an offer Noel couldn’t refuse, a chance to once again be part of a team taking on the biggest player in the game. He returned to a much larger ARM, a corporation of over a thousand people, and played a key role in its continued growth, eventually leaving in 2022 when the company had swelled to 8,000 employees.

This journey, from a 40-person startup to a global corporation and back to a blank sheet of paper with his new venture, Literal Labs, has given Noel a unique perspective on the technology industry. It’s a perspective that is deeply rooted in the principles of efficiency, a relentless drive to do more with less. And it’s this principle that lies at the heart of Literal Labs and its radical approach to AI.

For many, the term ‘AI’ is synonymous with neural networks, the complex, brain-inspired systems that have given us everything from image recognition to large language models like ChatGPT. These models are incredibly powerful, but they come at a cost. They are, by their very nature, probabilistic. This is a concept that I’ve spent a lot of time exploring in my own work teaching AI in project delivery. To put it in simple terms for those less familiar with the jargon, a probabilistic system is like a weather forecast; it gives you a likelihood, a percentage chance of a particular outcome. It might tell you there’s an 80% chance of rain, but it can’t tell you with absolute certainty whether you’ll need an umbrella.

In many applications, this is perfectly acceptable. But in the world of project delivery, particularly in disciplines like cost management, we need something more concrete. We need a calculator, not a weather forecast. We need a system that is deterministic, one that gives you a definite, verifiable answer every single time. The probabilistic nature of neural networks can be a significant limitation when the stakes are high and precision is paramount. This is where the work of Literal Labs becomes so fascinating.

Noel and his team are revisiting a different branch of AI, one based on logic. Instead of a complex web of interconnected ‘neurons’ that learn by adjusting statistical weights, their approach is built on what’s known as propositional logic – a vast network of ‘if-then’ statements. This method, which has its roots in the early days of AI research, is fundamentally deterministic. It’s a system of pure logic, where one and one will always equal two. The ‘learning’ process involves a voting algorithm, known as a Tsetlin Machine, which determines which of these if-then statements to include, which to exclude, and which to ignore. The result is a model that is not only more transparent and traceable but also incredibly efficient.

This efficiency is not just a minor improvement; it’s a game-changer. As Noel explained, the immense computational power required by today’s leading AI models is largely due to the sheer volume of multiplication operations they perform. Multiplication is an expensive process in computing terms, both in terms of the physical space it takes up on a chip and the energy it consumes. The logic-based approach of Literal Labs, by largely removing this multiplication, creates models that are orders of magnitude more energy-efficient. In a world increasingly concerned with the environmental impact of technology, this is a profound advantage.

This brings us to a crucial point about the future of AI, a perspective that I believe is vital for anyone involved in technology and project delivery. The future is not about a single, monolithic AI that will solve all our problems. It’s about a combination of different AI architectures, a toolkit of diverse approaches that we can deploy for specific purposes. As Noel eloquently put it, even the human brain is not just a neural network; it’s a complex interplay of different structures. The future of AI will mirror this, with logic-based systems sitting alongside neural networks, each playing to its strengths. We will use the right tool for the right job, choosing the deterministic precision of a logic-based system for our financial models, while perhaps using a probabilistic neural network for more creative or open-ended tasks

This vision of a more diverse and specialised AI landscape is not just a technical point; it has significant implications for the current geopolitical AI race. The UK, as Noel points out, has a history of punching above its weight in the technology sector, with ARM being a prime example. By fostering innovation in alternative AI architectures, the UK has an opportunity to carve out a niche for itself, to lead in the development of more efficient, more trustworthy, and more specialised AI systems. It’s a chance to move beyond the current narrative of an arms race for ever-larger models and to focus on building a more sustainable and responsible AI future.

Noel’s journey and the work of Literal Labs are a powerful reminder that in the world of technology, the path to innovation is rarely a straight line. Sometimes, the most revolutionary ideas are not found in the relentless pursuit of the new, but in the rediscovery and reimagining of the old. The story of the Tsetlin Machine, a concept that dates back to the 1960s, being resurrected to solve the challenges of 21st-century AI is a compelling narrative of how the past can inform the future.

Our conversation with Noel Hurley covered a vast and fascinating territory, far more than can be captured in a single article. We delved deeper into his experiences at ARM, his thoughts on building trust in a world of deepfakes and misinformation, and the delicate balance between regulation and innovation.

We even found out which three apps he would keep on his phone if he had to delete all the others. To hear these stories and more, and to gain a deeper understanding of the alternative future of AI that Noel and his team are building, I encourage you to listen to the full episode of the Project Flux podcast. It's a conversation that will leave you questioning the assumptions we make about artificial intelligence and feeling optimistic about the potential for human ingenuity to forge a better path forward.

Listen on

• Apple

• Spotify

• YouTube

  Read the Blog

That’s it for today!

See you soon,

James, Yoshi and Aaron—Project Flux