Artificial intelligence has become such a familiar part of our daily lives that many of us barely pause to wonder how it actually works. From photo apps that recognise faces to the large language models embedded in our productivity tools, AI is no longer a futuristic concept, it’s simply the landscape we live in.
Last October, the IET Sussex Network set out to peel back some of that familiarity with a talk titled “How DO Machines Learn?” It offered a rare chance to go back to basics and explore what actually sits beneath the buzzwords. Rather than diving into complex, abstract theory, the session focused on one of the simplest, and most classic, examples of machine learning: teaching a computer to recognise hand‑written numbers.
It’s a perfect illustration because it’s something humans find almost trivially easy. We glance at a scrawled digit and instantly know it’s a five or an eight, no matter how messy the handwriting. But to a computer, this is a surprisingly difficult task. Unlike typed text, there’s no consistent pattern, no standard shape, and no guarantee that two people will write a number the same way.
To train the model, the speaker used a dataset containing 50,000 hand‑written digits, each one labelled with the correct answer. This was the model’s “education,” the same way a child might practise with worksheets. Once the system had completed its training, another set of 10,000 images was used to test how well it had learned.
The result was a 97 percent success rate, an impressive figure for such a lightweight model.
But as the speaker pointed out, that three percent error rate looks rather different depending on the application. If an app that sorts scanned paperwork misclassifies three out of every hundred digits, most of us can live with that. If a self‑driving car misclassifies three out of every hundred stop signs… well, that’s a different story entirely. As he joked, he’d be “quite miffed (if not dead)” if the technology behind autonomous vehicles had that margin of error.
It was a playful reminder that success in machine learning isn’t just about accuracy. Context, consequences and confidence matter too.
For anyone who couldn’t make it, the Sussex Network has made the full talk available on its YouTube channel. And building on that strong foundation, the Network is now ready to shift gears and explore another big question, one that feels especially relevant in a world where almost everyone has experimented with ChatGPT, Copilot or another AI assistant:
How do Large Language Models think?
On the evening of Tuesday 10 February, Dr Jeff Mitchell, Assistant Professor in Computer Science & AI at the University of Sussex, will shine a light on exactly that. His talk, “How Do Large Language Models Think?”, will take us into the architecture of today’s most talked‑about AI systems. LLMs have shown extraordinary abilities to generate explanations, summarise complex topics and hold surprisingly human‑like conversations, but the mechanisms behind those abilities remain mysterious to many.
Dr Mitchell plans to explore how these systems are constructed, what their internal structure tells us about intelligence, and how their abilities compare with our own. He’ll discuss why training a model to predict text, something that sounds simple on the surface, can create behaviour that appears thoughtful or insightful. And he’ll examine whether that appearance has anything to do with genuine reasoning.
Whether you’re curious about the inner workings of AI, supporting organisational discussions about responsible use, or simply fascinated by the way machines seem to mimic human communication, this promises to be a lively and informative evening.
Join Us at Worthing College
As always, Sussex Network events are free and open to everyone, whether or not you’re an IET member, so feel free to bring colleagues, friends or anyone who’s curious about the future of technology.
People will gather at Worthing College at 6:30pm for refreshments and a friendly chat, with the talk beginning at 7pm. You can register using the link provided in the event listing, and if you can’t make it on the night, the session will be recorded and uploaded to the Sussex Network YouTube channel.
Travel information for Worthing College is available on the Sussex Network homepage.
We’d love to see a great mix of attendees, from seasoned engineers to those just dipping their toes into the world of AI. After all, understanding these systems isn’t just valuable for experts, it’s becoming part of everyday digital literacy. So come along, ask questions and join the conversation.
Share your thoughts
If you attended the October session, it would be wonderful to hear how your thinking has evolved since then. Do LLMs seem more mysterious than hand‑written digits, or perhaps less so? Has your organisation begun adopting AI tools in new ways? These kinds of discussions are what make the Sussex Network such a welcoming and collaborative space.
I hope to see you there.