Unless you’ve been living under a rock, it’s likely felt like everyone has had something to say about AI for the past 18 months. This week was no exception as I had to pleasure of meeting some of those at the frontier of the burgeoning sector. The summit featured a range of industry-specific sessions, showcasing the diverse applications of AI across industries. But beyond the keynote speeches and panel discussions, what were my three takeaways from the AI Summit that may have flown under the radar?

The Robot Revolution

Mass automation isn’t going to happen to the degree many are scaremongering about. Society has, to a degree, already resisted mass automation. Most jobs in developed economies like the UK are in the non-tradable service sector, which are inherently difficult to automate. These jobs include waiters, yoga instructors, nurses, and teachers. These jobs are resistant to automation, partly because of the market demand for human connection and empathy – consumers don’t want to talk to robots. It seemed that industries where knowledge and know-how are the cost for businesses are more likely to be safe as these sectors will adopt widespread AI faster to augment their value. However, industries where knowledge is the product, such as professional services, seem much more at risk as AI may democratise their knowledge for those who don’t currently have access to it.

Tomorrow’s models need more than memory

Current large language model (LLMs) architecture will only take us so far. Simply scaling an ‘inference machine’ – something that can memorise all known data – can’t teach a model to adapt to new forms of problems they’ve never been trained on. In the real world, where humans are confronted with new problems every day, we continuously adapt to novelty and then conjecture explanations for how to solve them. The current crop of LLMs’ strengths relies on memory and recall. While this is unreasonably effective for some problems, they fail consistently at others – as anyone who’s used ChatGPT will know first-hand! LLMs released in the next few years will continue to improve their memory-and-recall by holding vast swathes of knowledge in their “brains”, and pattern match across their memory in a way no human could do. But until we get another algorithmic breakthrough, maybe a way of programming that allows the systems to reason through problems, there’ll likely still be a need for a human in the loop.

Where are we heading?

There seemed to be an omertà on mentioning the direction of the field of AI. Call it a case of missing the forest for the trees. There was little to no mention of the acceleration toward artificial general intelligence (loosely defined as a piece of software that can do any cognitive task a human can do), long the dream of computer scientists since Alan Turing in the 1950s. What many seemed to be avoiding is that the frontier systems being grown in data centres across the East cost of America more likely resemble a new species than they do a ‘bicycle for the mind.’ Interestingly, the quantity of companies at this week’s event, backed with huge capital that are building on top of multimodal chatbots, are implicitly betting that the emergence of a silicon intelligence many orders of magnitude smarter than homo sapiens will not happen in the next decade. Artificial general intelligence could be the most influential technology of this century, above the scale of the nuclear breakthroughs in the 1940s. But at a summit specifically dedicated to the field, you wouldn’t have thought it.

Grayling have worked with some of the biggest technology clients for the last decade to help achieve their business goals. Get in touch to find out more.