Public transport is at a crossroads. While we've seen incremental improvements in accessibility and convenience, the core issues—cost and coverage—persist. Because public transport is expensive to operate, it remains a niche option, used by a minority - especially outside inner cities.
However, artificial intelligence offers us an unprecedented opportunity to break free from constraints that have existed since the start of the era of car dominance. By reimagining how public transport is delivered and expanding its reach, AI could usher in a revolution—if we're bold enough to embrace it.
The starting point is that only a tiny proportion of people in the UK live within walking distance of high-frequency public transport (I’m lucky enough to be one of them). If you do have frequent public transport accessible locally, you choose to use it. That’s why, in the UK, most journeys in London are made by public transport but not most journeys in Leicester.
“We can deliver a better outcome for carbon by using AI to make sustainable transport superb”
The problem is that providing high-frequency services to most locations is unaffordable. The main driver of the cost is the driver. Take that out, and the position is transformed.
Now, I’m not saying that we’ll take the driver from existing services. Most services currently operate with one human and a large number of passengers (as services with few passengers are unviable). One member of staff is probably the lower acceptable limit. That’s why London’s Docklands Light Railway and Victoria underground line have operated with staff on board since they opened decades ago, even though the trains have always been automatic.
But what driverless public transport will enable is an extraordinary expansion of service to places that are unaffordable when staffed, but affordable when automated.
It’s not about cars
There’s a lot of focus on autonomous cars. However, the main problem with cars is physics: they take up too much space. The logic of a famous 1960s UK newspaper advert still holds true. In three pictures it showed, first, a street full of cars; then the same street with a number of people standing in it; and, finally, a single bus on the same street. Next to the pictures was the tagline: “These vehicles are carrying 69 people who could all be on this one bus.”
By contrast, driverless vehicles have the potential to create gridlock. If people like me who don’t drive start taking AVs, no-one will ever move. AI can optimise but it can’t solve impenetrability: the law of physics that states that only one object can occupy a physical space at a time. Only one car can occupy each space on the road. Autonomous or otherwise, cars have a percentage point of the capacity of buses.
Moreover, just as with other forms of AI, AVs will continue to make mistakes. That’s why it’s taking so long to get beyond a few robotaxis in a couple of American cities to global adoption. But public transport AVs can have the back-up of a room full of video operators. One operator can be responsible for escalations from hundreds of vehicles. If the vehicle isn’t sure what it can safely do in a particular situation, it can propose a solution to a human, and the human can check. A shuttle that stays within a defined suburb will have far fewer issues than a driverless car that needs to know every streetscape on Earth.
“My point isn’t that AI is good. My point is that AI is happening. Our job is to get the best we can out of it”
That means that each suburb, town and village can be connected by high-frequency autonomous shuttles to the core public transport network. This would be transformational. And it will become possible.
The reason it will be possible is that the cost of public transport without the cost of fuel or drivers (obviously these vehicles will be electric) is a fraction of the current cost. Imagine the places that it will be possible to profitably connect! If public transport connectivity can be provided for £1 per vehicle mile, the density of the network will transform. As the peripheral network becomes larger, the core network becomes stronger. It’s the opposite of the Beeching Cuts – the 1960s closures of thousands of miles of the UK rail network. I can’t wait.
My only slight worry is that I don’t see as much progress being made towards making it happen as I’d like. The only city that’s doing this seriously and properly is Norway’s capital, Oslo, under the leadership of Bernt Reiten Jenssen, visionary CEO of public transport authority Ruter. We all need to catch up.
But the cars are coming
However, we cannot take our eye off the ball when it comes to AVs. I have long predicted that it will take a very long time to get to Level 5 autonomy – i.e. when a car can go anywhere without anyone in the driving seat. It’s not because the technology isn’t there to make it happen most of the time. It’s because the consequences of failure are so great. But, equally, autonomous cars are already live in US cities, carrying passengers.
We need to be alert that AV operators don’t try to pull off the same trick that their automotive industry predecessors did almost exactly a century ago. Back in the 1920s, car execs realised that it would be easier to get their product off the ground if cities were better designed for cars. So they persuaded councils to put fences along pavements and to corral pedestrians onto fixed crossing points. You can imagine a similar dynamic playing out with autonomous.
AI-driven cars also create the risk that the 'taxi’ product becomes a much more significant competitor. One of the characteristics of AVs is that it becomes much less clear what is a bus, what is a taxi, what is a car and what is a private hire vehicle. If you could use your AV as a taxi when you’re at home watching TV, there is a significant risk of new competition from vehicles that are more sustainable than a petrol car but a lot less sustainable than other transport modes.
Better and cheaper
It’s therefore crucial that our core network becomes cheaper. It won’t become cheaper because we will take out the drivers from busy services - we won’t. But the network will become cheaper. AI will be able to optimise schedules both in advance and in real-time. What’s the best roster for all the buses and trains in a given region to balance cost and revenue? Like all models, AI’s answer will depend on assumptions (elasticities will remain as key as ever). But it’ll have the capability to try to answer the question. Today it’s a question we don’t even try to answer.
Some of you may doubt what I’m saying. You’d be right to. I mean, obviously, I’m wrong - but I’m not going to be a million miles wrong. The core capabilities of AI are now pretty clear. So the only question is whether they will scale to the extent I’m describing.
And I think we got the answer to that last year when Google announced an agreement to purchase nuclear energy from reactors to be developed by Kairos Power because, Google says: “The grid needs new electricity sources to support AI technologies.”
That’s how much power they think they’ll be using. That’s because they’re going to be doing… well… the stuff I’m talking about. So I’m definitely wrong in the detail. But I’m not wrong that it’s another big transition, like the internet.
What about the planet?
Many of you may worry about the sustainability implications of all this. I certainly do. After all, not every AI company is thinking about the energy they generate. I asked the founder of one AI start-up how he thought about the carbon implications of what he was doing and he literally hadn’t thought about it at all.
The problem is that this article isn’t what I want to be happening: it’s what is happening. I can’t prevent AI any more than King Canute could turn back the tide. My specialism is transport and mobility, so all I can do is try to minimise the carbon emission in our sector. And my assertion is that we can deliver a better outcome for carbon by using AI to make sustainable transport superb.
This is big
To give you a sense of the power of AI, a modern AI model uses 10 billion PetaFlops of computing power to train. (This isn’t a prediction - it’s reality today.) Flops and PetaFlops aren’t language we’re familiar with (Flops stands for floating-point operations per second). It’s a measure of computing power. Just saying that 10 billion PetaFlops is a huge amount of computing power is pretty meaningless. So let me help you try to visualise what it looks like.
Imagine doing a maths calculation on a pocket calculator. Visualise that? OK, that’s about 20 Flops. Now imagine you’re a real whizz at multi-tasking, so you’re simultaneously doing a maths calculation on 50 pocket calculators at the same time. I appreciate this might be harder to visualise. Now imagine that every human on earth is simultaneously doing a maths calculation on 50 pocket calculators each. Now imagine that every planet in the solar system is populated with as many humans as Earth, and they are all doing a maths calculation on 50 calculators each.
“Get it right, it will usher in a new golden century for public transport; get it wrong, and we’re dinosaurs”
I realise we’re getting beyond the usual “size of Manhattan”, “number of double-deck buses” or “area of football pitches” scales here. Sorry. There’s a lot of computing power in AI. And I’m afraid this visualisation is about to get worse.
You see, and bear with me, but I’m going to have to ask you not to operate 50 pocket calculators but seven trillion. If all your seven trillion pocket calculators were piled on end, they would stretch out from your hand to the sun. And back. Six times. So I’m afraid you might struggle a tad to operate them all at once. But you’re game for a go, right?
So, there we have it, if you imagine you and everyone else on Earth (plus an equivalent number of people on every other planet) all simultaneously doing a mathematical calculation on a pile of pocket calculators that each stretches from Earth to the sun and back six times, that is how much computing power is used to train a modern AI model.
I don’t know about you, but I think that’s going to have an impact on things.
Any downsides?
Are there downsides other than the fact AI requires so much energy? Oh my goodness, yes: AI has the potential to regurgitate our own biases, to build enormous dependency on hackable computers and to blur the lines of organisational accountability.
My point isn’t that AI is good. My point is that AI is happening. Our job is to get the best we can out of it. And now, I’m afraid I need to come to the tricky part. We don’t have a great track record of getting the most from technology. In fact, transportation is one of the only sectors I can think of where technology and automation make things more expensive.
When train signalling systems moved the skill first from the cab to the signal box and then to the software developer’s studio, we incurred all of the costs of modern signalling but carried on paying the drivers as if the job was as skilled as it had been before. When the e-commerce revolution hit transport, we incurred the costs of online retailing. And we kept booking offices open. And we have more ticket vending machines than ever. And dissatisfied customers.
In general, automation allows us to add new costs, retain the old costs and fail to deliver consumer benefits. Now, this is a problem given what’s about to happen. I’m not blaming anyone else for this. I’ve been a director of transport companies since I was 26. I carry as much blame as anyone, and more than most. Collectively, we haven’t fixed this. But we’ve got to. One of the reasons I created Freewheeling is that I can see this coming and feel like it needs an outsider championing the change. When I was at Transport for London, I found it surprisingly hard to have this conversation. When I was on the inside doing the job I did then, I really needed someone like me on the outside.
The AI revolution is going to transform every sector, upend customer expectations and revolutionise the workplace. But we shouldn’t see it as a threat. If we get it right, it will usher in a new golden century for public transport. Get it wrong, and we’re dinosaurs.
ABOUT THE AUTHOR
Thomas Ableman is former director of strategy and innovation at Transport for London and founder of Freewheeling
www.freewheeling.info
