Last week, we looked at some of the more troubling stories emerging as AI becomes more powerful and more widely used.
But that is only one side of the picture.
Because while there are very good reasons to question how AI is being used, there are also examples of the technology doing things that are genuinely useful – from helping scientists uncover previously hidden patterns in cancer cells to giving surgeons an extra pair of eyes during brain surgery.
Perhaps what is particularly interesting about these examples is that, for the most part, AI isn’t replacing human expertise. It’s helping people work through information, spot patterns and make better-informed decisions.
And that may be where some of its greatest potential lies.
Anyone who has ever had to research an unfamiliar or complicated subject will recognise the problem: finding information is one thing; working out which information is relevant and trustworthy is quite another.
AI research tools such as Anara are attempting to make that process considerably easier.
Designed specifically for academic and scientific research, Anara can search across large collections of papers and documents, pull together relevant information and help researchers compare and synthesise their findings. Crucially, its answers include citations linking back to the passages they came from, allowing the researcher to check the original source rather than simply trusting an AI-generated response.
That distinction matters.
One of the biggest criticisms of generative AI is its ability to produce a confident-sounding answer that isn’t necessarily correct. A research tool that helps people find and understand information while also making it easier to verify the evidence behind it is a rather different proposition.
For businesses, there is a useful wider lesson here too. AI doesn’t always have to produce the finished answer. Sometimes its greatest value may be in helping us get to the right information more quickly – while leaving us to decide what to do with it.
At the University of Southampton, researchers have been using AI for something considerably more significant: analysing breast cancer cells.
The team developed an AI platform called CenSegNet and used it to analyse more than 330,000 cells from 127 patients. The technology identified previously unseen patterns in abnormalities within cancer cells – including differences that had previously been treated as though they were the same.
Researchers found that some of those patterns appeared to be associated with more aggressive disease, while others were linked with better overall survival.
The technology isn’t ready for routine hospital use yet, but researchers believe it could eventually help doctors understand which cancers are more likely to grow, spread or resist treatment, potentially contributing to more personalised treatment.
This is a very different use of AI from asking a chatbot to write an email or summarise a document.
Humans are exceptionally good at recognising patterns, but there are limits to the amount and complexity of information we can analyse. AI can examine enormous quantities of data and identify relationships that might otherwise be extremely difficult to spot.
And when the information being analysed could ultimately help doctors understand a patient’s cancer more precisely, that ability becomes much more than a convenient time-saver.
The Scottish FA has provided another interesting example – this time in recruitment.
When looking for a successor to Scotland manager Steve Clarke, the organisation used AI and data analysis to help create a shortlist.
Rather than simply starting with the obvious names, the Scottish FA worked with sporting intelligence specialists to analyse what successful international managers actually have in common. The process challenged some conventional assumptions – including the idea that previous international management experience, age or nationality necessarily makes someone more likely to succeed.
Around 40 potential candidates emerged from the analysis, including Sébastien Pocognoli, who ultimately got the job.
But perhaps the most important part of the story is what happened next.
The computer didn’t appoint him. People did.
The data helped broaden the search and challenge assumptions; human interviews and judgement were still used to decide whether he had the personality, leadership style and qualities required for the role.
That’s an interesting model for businesses considering AI in recruitment more generally.
Used badly, AI recruitment systems could simply introduce another layer of bias or remove too much human judgement from an inherently human decision. Used thoughtfully, however, data and AI could potentially help employers question their assumptions and identify good candidates they might otherwise never have considered.
Perhaps the most remarkable of these stories comes from University College London Hospital, where a patient became the first in the world to undergo brain surgery with live AI assistance.
Rhys Hibbert had a tumour located close to important blood vessels and nerves, including structures connected with his sight.
During the operation, an AI system analysed the live video feed from the surgeon’s camera. It tracked the surgical instruments and highlighted where crucial – but sometimes hidden – vessels and nerves were likely to be, helping the surgical team identify safer areas in which to operate.
The AI hadn’t simply been shown diagrams of what human anatomy should look like. Researchers had trained and evaluated it using hundreds of videos from previous tumour operations, teaching it to recognise important structures in real surgical footage.
Again, the important point is that the AI wasn’t performing the surgery.
The surgeon remained in control, bringing years of training, experience and judgement to the procedure. AI provided additional information at precisely the moment it was useful.
It’s difficult to imagine a clearer example of the potential of human expertise supported by AI, rather than replaced by it.
Not every useful application of AI needs to involve groundbreaking medical research.
Morrisons is among the supermarkets testing AI-enabled shopping trolleys designed to make the weekly shop a little easier.
Customers scan their loyalty card and add products to the trolley, which uses cameras to read barcodes and keeps a running total of the cost. Loose produce can be weighed in the trolley, while the screen could potentially alert shoppers to offers they have missed.
There are still some wrinkles to iron out – during the BBC’s test, the trolley didn’t spot an attempt to put five butternut squashes into the basket while declaring only one.
But that is precisely what trials are for.
The more interesting question is what technology like this could eventually do. Could a trolley warn someone that a product contains an allergen? Help shoppers keep within a budget? Suggest an offer that genuinely saves them money? Or simply remove some of the hassle from checking out?
None of those things will change the world.
But useful technology doesn’t always have to.
Taken together, these stories present a rather different picture from the examples we looked at last week.
AI is helping researchers navigate vast quantities of information. It’s finding previously hidden patterns in cancer cells. It’s helping recruiters challenge assumptions, assisting surgeons during extraordinarily complex procedures and even experimenting with ways to make the weekly shop a little easier.
There is also a common thread running through many of them.
The AI provides information, analysis or assistance. A human remains involved.
The cancer researchers still need to understand what the patterns mean. The Scottish FA still had to decide who should manage the national team. The surgeon was still performing the operation. And even Anara’s approach to research is built around making it easier for people to check the evidence behind an AI-generated answer.
Perhaps that gives us a useful way to think about AI more broadly.
The question doesn’t always have to be, “What can we get AI to do instead of us?”
A better question might sometimes be, “What could we do better if AI helped us?”
And that feels like a rather more optimistic place to start.