AI could transform fleet management

Call it professional curiosity, but I’m fascinated by the almost day-by-day evolution in artificial intelligence during the last 18 months. Science fiction promised us the likes of Hal and Skynet, but the reality seems less threatening. It’s materialised as friendly chat-bots who’ll answer questions, draft scripts for imaginary film sequels and draw strikingly colourful landscapes within seconds. I have no doubt that it’ll find increasingly productive roles across all sorts of businesses, but I’d also advise caution for the early adopters.

Yes, I would say that. As a writer, I’ve had countless people asking me whether I see it as a threat – after all, why would you pay a human to spend hours working on something a ‘bot can turn around for free, within seconds? However, with a bit of experimentation, I’m getting a clearer view of the capabilities, pitfalls and limitations it presents. Understanding this technology is important.

Let’s start with the terminology. Most consumer-facing AI is based on what’s called a large language model (LLM), where developers feed vast amounts of text and images into the software, which then learns to spot patterns and figure out how humans tend to combine them. It’s compiling influences, rather than creating – and the results are always limited by the quality and quantity of data it’s been exposed to. If it doesn’t know, it’ll guess with expert-level confidence.

As a tinkerer, the results are amusing. Bots struggle with detailed analysis of specialist topics, add extra fingers to photo-realistic human portraits and garble text in images. I’ve seen them write sample buyers’ guides for classic cars that wax lyrical about unimaginable-at-the-time infotainment systems. I’ve also seen image generators repeatedly drawing the wrong car, despite very specific prompts. Impressive, but flawed.

There’s a dark side to this. I spent some time last year writing about facial recognition software and was repeatedly told that it’s prone to racial bias, based on the sample size of the different ethnic groups in the training data. Using bots for research is prone to the same issues – the answers are prone to being skewed by political or cultural biases, inaccuracies and manipulation within the source material. Just ask the New Zealand supermarket chain whose ‘Savey Meal-Bot’ cheerily provided users with the recipe for chlorine gas, a chemical weapon. Garbage in, garbage out.

The source material is riddled with unsolved issues too. The New York Times is currently suing several tech companies alleging that scraping training data amounts to copyright theft. Although the World Health Organisation says it cautiously supports using LLMs for health-care, it’s also raised privacy concerns about bots being fed patient data and the potential for this software to generate misleading and harmful advice. That certainly isn’t amusing for anyone who follows it.

Perhaps these are wrinkles that’ll be ironed out as the technology matures, but they’re also a reason to be cautious today and this is something fleets need to approach with their eyes open. AI was the automotive bandwagon at this year’s Consumer Electronics Show, promising conversational in-car assistance and more accurate results. Given how much our cars already know – and are learning – about us, it’s worth getting drivers to study the small print before they dive in.

However, I’d also advise some professional curiosity among fleet managers, for whom AI’s inherent trend-spotting talent could be a very useful asset. There are already tools promising to optimise routes, vehicle selections and maintenance based on multiple data sources, so I see it transforming fleet management before it makes any meaningful difference to my career. Just make sure you avoid the ‘garbage in’.