Twitter's transport experts are wrong, and I know better than them

In which I unwisely pick a fight with some actual experts.

If you’d like me to write more regularly, please consider pre-pledging a subscription for if I ever launch a paid tier. I’d love to write more essays like this but they take several work days to produce. So I need to know people are willing to pay for my blazing takes if I’m ever going to make the leap.

---

If there’s one phrase that sums up all of the problems that Britain has faced since 2016, it is probably the one from Michael Gove’s infamous interview on Sky News just before the referendum, when he uttered the immortal line: “I think the people of this country have had enough of experts.”

It’s hard to think of another combination of words that could cause me such an instantaneous fury, short of “James Corden wins lifetime achievement award”.

So when I heard this, it was like a red rag to a bull. The fact that Gove didn’t exactly say this was besides the point as I was already too angry to think straight¹.

This is because I like experts. Most of my job as a journalist and writer is phoning up experts, asking them for their expert opinions, and then writing down what they say.

So in this post, I’m going to do something dangerous, and disagree with not just one expert, but two – both of whom, I actually really like.

A couple of weeks ago railway engineer and all-round transport guru Gareth Dennis highlighted a new driverless bus project in Scotland. He criticised it as a “total and complete waste of time and money”². And he was not alone. The transport author Christian Wolmar similarly argued that the project is “Another ‘driverless’ con”.

Usually, I’d defer to their superior knowledge of the field and assume that they’re correct, because they’ve spent years working on transport projects for real. Gareth’s passionate advocacy is what radicalised me as a pro-HS2 ultra³. And Christian is perhaps the best mayor that London never had⁴, plus he wrote a really great book about the history of the London Underground⁵.

So I definitely don’t have anything like their expertise. But I do have some tenuous transport nerd credentials of my own.

For example, I turned up at Abbey Wood at 6am to catch the first Elizabeth Line train on the first day of operations, I’ve visited every single Tube station, and a few years ago I actually planned a holiday around (incredible nerd boast alert) travelling on the Wuppertal Suspension Railway⁶

Now here’s where it gets awkward: I think that Gareth and Christian are wrong about driverless buses. In fact, I think driverless buses could work great, could revolutionise transport – and help deliver the O’Malleyist agenda for Britain.

So here’s why I’m right, and the people with the credentials, professional experience and years of study in the field, are wrong.

Rejecting experts? WTF? Is this the start of my populist tilt? Is it now only a matter of time until I’m railing against corrupt elites and arguing that the eggheads in Whitehall just need to believe in Brexit a little harder? Or will I just continue writing mildly contrarian columns about how building luxury flats for millionaires is good, or that we should engage properly with the intellectual ideas behind ‘wokeness’? Subscribe now (for free) and find out directly in your inbox.

Or if you really like my work, you can of course pre-pledge a paid subscription. Oh go on, it’s my birthday in a few days time (and it’ll cost you nothing unless I do finally make the leap at some indeterminate point in the future.)

Realistic expectations

Let’s start with the obvious question: We know that driverless vehicles are getting pretty good, as evidenced by the fact that many major tech and automotive companies are testing them on real roads.

I’ve actually experienced a few driverless tests myself: Way back around 2017-ish, I went to a Nissan press event and was driven around a circuit of real London roads without the driver needing to touch the controls once. I’ve also travelled on an autonomous pod-like bus that was being tested off-road in London’s Olympic Park.

Subjectively, both felt completely fine. In both cases, within seconds I was chatting to the safety driver like it was a normal car journey. And my level of alertness was about the same as when I’m sat in the passenger seat of a car, in the sense that I’m not closely watching the road – but I am paying just enough attention to absolutely infuriate my partner by second-guessing her driving decisions.

Both of these events were now several years ago. Since then, the technology, expertise and knowledge of the companies developing such systems has continued improving⁷. In fact, it appears to be getting seriously good – to an extent we might not realise.

Safety records

The first thing to remember is that autonomous vehicles do not need to be 100% perfect, 100% of the time. What’s more important is how autonomous technology performs compared to humans.

And I think there is evidence that autonomous vehicles are getting there.

Take Tesla, for example. It has (controversially) deployed what it calls “full self driving” to thousands of real world users, and whether or not you think the decision to do so is reckless or not, it’s clearly very impressive.

Check out this video of the most recent software update, which shows a 20-minute journey between two points in Los Angeles, where the driver does not need to touch the wheel once⁸.

(Try to ignore the Musk zealot providing the commentary, and remember not to let Elon break your brain.)

What makes it particularly, and I think surprisingly, impressive, is that unlike other autonomy projects, Teslas use only on-board cameras to navigate, and don’t augment navigation with expensive Lidar sensors.

But what about more rigorous safety data? Annoyingly, such data for autonomous cars – let alone buses – is actually quite hard to find. But there is some evidence we can point to.

Waymo, Google’s autonomous car project, claims to have driven a million miles with no injuries, and only two collisions that met the criteria that required them to be reported - with only 18 “minor contact events” otherwise. And Cruise, an autonomous subsidiary of General Motors, also claims to have driven a million miles with fewer incidents than humans would make in the same circumstances.

Similarly, Tesla argues that it’s ‘autopilot’ features are statistically incredibly safe⁹.

On the other hand, we do know there has already been one autonomous death. In 2018, a pedestrian in Tempe, Arizona was killed by an Uber test vehicle (the company has since abandoned the project).

So you can take the above stats with as many grains of salt as your trust of major corporations allows¹⁰. But I think drawing any grander conclusions about what this all means is hard, largely because most accident stats are measured in terms of accidents per (very long) distance driven, and autonomous vehicles have simply not yet racked up enough real world miles yet to make a more meaningful comparison¹¹.

In any case, I admit there are plenty of reasons to be sceptical of autonomy: It’s not a coincidence that both Waymo and previously Uber were both testing in Arizona, with it’s obnoxiously wide highways and year-round good weather. That’s a very different scenario to a hefty bus weaving around a narrow Victorian high-street in Scottish rain at 11:30pm on a Friday evening.

Lowering the bar

So why am I still so bullish about autonomous buses?

Essentially, while I’ve talked a lot about cars, because that’s where much of the public data is, buses are probably significantly easier to automate – thus moving the bar for “feasible technology” somewhat lower.

Why? Because buses are much more predictable than cars. They travel on the same fixed routes dozens of times every day, they move slowly, and they are managed as part of a fleet. And this means that instead of simply relying on on-board sensors and cameras like a Tesla or Waymo, buses can do much more.

For example, it’s conceivable that the bus company could go out and capture a detailed 3D map of the specific bus route (and immediate surroundings to account for potential diversions etc), and manually process it to ensure that it is correct and up-to-date, the details of which can then be fed into the autonomous driving system to make sure it is working from the best data available.

Similarly, it could use sensors in the road or street furniture along the route to augments its navigation capabilities. In fact, this is exactly what the trial in Scotland is doing – the company running it says that it will be receiving real time data from traffic lights en route. And you can imagine how in the future, if there’s a particularly tricky junction to navigate or similar, the bus company could deploy extra sensors and cameras to aid the bus passing through¹². Autonomous buses don’t need to know how to essentially drive everywhere like driverless cars will need to – they just need to know how to drive one specific set of routes really, really well.

And because the buses will be centrally managed, the autonomy features don’t have to be 100% perfect like they would be for a private car. It’s possible to imagine how the developer could reach, say, 99% reliability and then perform the last 1% of manoeuvres with a human driver connecting remotely from the office, as an emergency back-up. (This is a technique used by the ground-based delivery robots that you can see trundling through the streets of Milton Keynes, as part of a separate autonomy trial.)

So for all of the bewildering complexity, when it comes to actually driving, buses are easier to automate. And given we can already see almost flawless examples of autonomous driving in the real world, I think there is a strong case to think this really could be the moment that autonomous vehicles take off.

But I can perhaps forgive you if you don’t quite believe me. For literally decades, we’ve watched Tomorrow’s World et al show us the driverless cars of the future, only for nothing to happen. Tesla’s forays into “full self driving” are famously crappy, and dangerous too.

But I think assuming this lack of progress will continue forever is basically faulty pattern recognition, because this starting point where a new technology is terrible is also true of basically everything.

The first aeroplanes were useless – right up until they reached a tipping point of reliability and functionality, and then they transformed first the battlefield, and then the human geography of the entire planet.

The first electric cars looked like ‘The Homer’ and barely had enough range to leave the driveway, but the technology incrementally improved to the point where, finally, just in time, we have a viable alternative to the combustion engine.

And most recently, until the end of last year generative AI was crap – and then ChatGPT and Midjourney happened, and though they are not perfect, they are good enough to be extremely useful (even the image at the top of this piece is AI).

So I don’t think it’s insane to think we’re approaching a similar tipping point with autonomous driving, and that we should enthusiastically put money into tipping over it.

The Real World

However, the the technology is not, of course, the full story. The other half is the messy business of humans.

For example, even if you have a driverless system, buses still need to allow people to board and alight safely. And, as Gareth argues here, this is hard enough to do on trains which live on fixed rails¹³.

That’s one of the reasons why the Scotland trial is slightly awkwardly including not just a safety driver, poised at the steering wheel, but a second staffer as a “bus captain” who can conceivably check tickets, help wheelchair users board, and so on.

But I don’t think these sorts of “real world” problems are particularly insurmountable.

For example, what’s great about buses compared to other forms of transport is the public sector doesn’t just (directly or indirectly) manage the buses – it manages the roads too.

So to ease the boarding process, it’s conceivable that the public realm could be better adapted. Perhaps it might be useful to have pavements ramp up towards the kerb to allow for unassisted level boarding? Or perhaps roads could be adapted to make vehicle movements more predictable? Or bus stops could be better optimised to help autonomous vehicles position themselves?

Hell, we could even have special lanes that are only for buses to reduce traffic they have to interact with – we could call them “bus lanes”.

But this is just me speculating. I don’t know the details of exactly what adaptations might be needed as that’s more of a question for the likes of Gareth and Christian. What I do know, however, is that as literally fixed in stone much of our built environment appears to be, it can be easily changed. Just as we built petrol stations for what replaced horses, and are building electric chargers for what is replacing the combustion engine, we can adapt for autonomy – and other new forms of transport too.

And as for on-board staff, it’s pretty easy to imagine how most tasks can actually be completed without the need for a human on every single bus¹⁴. Even if there are worries about safety (for example, we might worry that the bus driving off too fast while passengers are standing up), it’s conceivable that concerns could be mitigated as the technology is proven in real world service¹⁵. After all, we no longer require a guy with a red flag to walk in front of cars.

Technological feasibility aside however, there are other objections – which are basically ideological but still may count as reasons not to pursue autonomy.

The most obvious one is money. The trial in Scotland is receiving public money – which could be spent on other transport priorities or even (spins wheel), NHS nurses, “our boys”, or liberating the Postcode Address File.

But I think this technology is worth trialling. Why? Because you only have to invent autonomous buses once. Of course, if we were developing a self-driving technology for this one specific bus route, it’d be crazy – but once autonomy has been figured out, there’s no reason it couldn’t deploy to, well, every other vehicle in the world eventually. The potential upside of mastering the technology seems obvious and enormous to me¹⁶. So just as it was useful for the government to place speculative bets on different vaccines just in case they turned out to be effective, it seems like a good thing for the government to place bets on different autonomous transport trials too.

The other objection is perhaps trickier: If autonomous buses are successful, then by definition a lot of bus drivers are going to lose their jobs. This would, of course, be sad for the drivers affected, but this is also why another core plank of the O’Malleyist Manifesto is that we need an extremely generous welfare state to support those affected¹⁷ by the march technological progress¹⁸. But in any case, in the short-term at least I would bet that even if buses can be fully automated, as per the above, that’s still a long way from every driving task being automated, so there will be adjacent industries that will require human drivers for a long time yet¹⁹.

The O’Malleyist Transport Revolution

There is one final argument, and this is perfectly articulated by Gareth in the above article about trains – but it’s relevant to buses too:

Whenever a technology-led proposal is put forwards, the first question everyone should ask is always: “What problem will it solve?” and if the list is very short, then that technical proposal is likely to be without merit.

And this is where I get to what I think is the strongest argument for why autonomous buses are worth pursuing.

Simply put, if you’ve ever tried to catch a bus outside of London you’ll know that they are, for the most part, shit. Thanks to a generation of austerity, meagre local authority budgets, and a completely uncoordinated approach to public transport, passengers outside the capital have to put up with irregular services, poor reliability, and disgustingly high fares.

The most expensive part of a bus’s operating costs is the labour. According to this guy, the driver typically makes up 70% of the cost of running a bus. So if we can remove the need for a driver, we can make buses cheaper, and we can have more of them: Without a driver, more routes will become more economical to operate at more times of day. And we could up the frequency of service. And maybe we could reduce fares too.

Imagine a future where we have silent, electric-powered autonomous buses running around the clock²⁰. No private bus company would pay a human to drive the 4am bus that runs empty for the entire route most nights, and no public sector would subsidise it. But if the bus can drive itself and the cost is a trivial amount of electricity (perhaps generated locally at the depot using solar panels during the day), then running the service just in case is more viable. And it is these marginal gains that are important, because they affect personal transport decisions²¹.

This is because public transport needs to be viewed as a system. If you want to make a transport network that people use, then slack in the system matters because it makes it more convenient. If you can be confident that there will always be a bus to get you where you need it, then it reduces the need for private vehicles.

This is why London buses are so good, because you really can just turn up and go, without any planning ahead. Autonomy can make achieving this level of service cheaper, and therefore more viable, over a wider area.

So I think if autonomous buses can be perfected, and in turn they can increase the capacity, speed and density of bus networks, they could have a big multiplier effect in terms of getting people out of their cars.

And if we can get more people using buses, it also serves some important other policy goals: It will most obviously help reduce carbon emissions, it will reduce traffic congestion, and it will make it more economical in terms of both time and money to live in more places, making the housing market more flexible. It could reduce the cost of living. And it might even improve our productivity too.

So let’s get on board, and bet on an autonomous future.

Congratulations! You’ve made it to the end. I can’t believe you’ve just read almost 3000 words about buses either. So if you have, you must really like my stuff. In which case you should subscribe (for free!), follow me on Twitter, share this post with your friends and most importantly pre-pledge a paid subscription in case I do ever launch a paid tier.

1. Update (23:38): This intro has been updated to reflect the fact that Gove didn’t actually suggest people have had enough of experts - he was actually referring to “three-letter acronyms” like the OBR or whoever.

2. It’s not clear exactly how much public money the trial is getting - but it is expected to cost £6.1m and some of the cash is coming from the government’s £100m Intelligent Mobility Fund.

3. I also regularly enjoy his demolitions of ludicrously implausible transport projects that he calls “gadgetbahns” – things like the Hyperloop or pretty much every monorail ever constructed.

4. As a candidate for the Labour nomination, he was (I think) the first person to seriously pitch pedestrianising Oxford Street, something which should have been done decades ago and yet still hasn’t, presumably because the road is officially managed by Westminster Council instead of TfL.

5. The other amusing aspect of this for fans of low-key Twitter beefs in the mould of Ford-vs-Goodwin, is that though they’re in agreement on driverless buses, they regularly spar with each other over various issues. Christian’s actually a vocal opponent of HS2, so I guess I disagree with him about that too.

6. At this point I want to remind readers once again that I do indeed have a partner and she is a real, human, woman. She also liked the Wuppertal suspension railway and sighed in what I assume was agreement when I prompted her to agree with me that it was an incredibly romantic trip.

7. Something I found very surprising when I went to the Nissan demo in 2017 was how low resolution the images the car was processing were. I can’t remember exactly, but it was of the magnitude of 320x240 or 480x360 – not massive 4K images like I’d assumed. This was because larger images require much more computing power to process, and obviously a car needs to process images quickly so it can make split-second decisions. It wouldn’t surprise me if they’ve upped the resolution since then, but this sort of underscores my point, as the Nissan was able to navigate the test route completely competently, by looking at images smaller than the average Windows 95 desktop wallpaper.

8. The same guy has posted a bunch of similar videos and, tragically, I’ve watched a few of them in full and in others the car does screw up a few times. But none in any particularly significant ways and in almost every case he’s not had to actually touch the wheel to fix it. And remember, the baseline isn’t “perfect”, it’s “better than humans”.

9. I’m not sure Teslas make for a perfect comparison with full, “Level 5” self-driving experiments like Waymo – as its “autopilot” is arguably just a branding exercise for much lower level autonomy features that only work with the expectation there will be a fully attentive driver. But I’m including the link because I know otherwise someone will complain “But what about evil Elon Musk?”.

10. Just as I was writing this a reported Tesla data breach emerged which reportedly contains 4000 complaints about Tesla’s “driver assistance”. But without further context or detail it is hard to know what this means.

11. I did attempt to put together a rough back of the envelope, apples-to-oranges comparison of TfL bus safety data with Waymo’s purported 18 incidents over 1m miles. Of the over 7000 TfL “incidents” recorded between 1st April 2022 and 31st March 2023, 755 were classed as “collisions” (the others were, for example, passengers falling ill and other things that probably aren’t the bus’s fault). And amazingly, six people actually died after being hit by a bus. Once you factor in how far vehicles travelled though, you can spin it as arguably good for humans (one TfL collision for every 378,261 miles travelled), or good for Waymo, as the bar for Waymo basically appears to be a scraped bumper, whereas many TfL collisions resulted in hospitalisation or medical treatment. So to cut a long story short, given that we’re comparing cars in Arizona to buses in London to try and make a point about suburban buses in Scotland… I think we need to wait for more data to draw any strong conclusions.

12. Eventually, cars will probably communicate with each other on the road in order to share information about their position, hazards ahead and behind, and so on. But this is going to take some time to emerge, because of a need for technical standards for sharing data, agreeing what those are going to be internationally, and worrying about privacy and all that sort of thing. But in a sense, even if that’s where autonomous vehicles end up, I think it is even more striking that we can already get autonomous vehicles that are competitive with humans in terms of driving competence by only using local sensors.

13. Something I’m particularly curious about is how the autonomous bus will handle passengers boarding. Will it be smart enough to speed past stops that do not have anyone waiting? How will it discern between people waiting for the bus and teenagers just hanging out at the bus stop? Or perhaps it will just go for the simpler and less time-efficient option of stopping at every stop for a fixed period of time, with some sort of sensor on the doors to close them once the flow of people boarding has stopped (perhaps it could then not move the bus until internal cameras detect that everyone is seated/stationary?).

14. The one slightly difficult bit I can envisage is needing a human to help people with limited mobility. But even in this case, there’s a solution, albeit with a slightly fraught trade-off: Say you have an existing route with 3 buses an hour, each with a driver who can help people in person. If autonomy comes along and can make it economical to run six or even twelve buses an hour, could humans only be present on a handful? It might limit specific buses for people with limited mobility (albeit with the same service level they enjoy with a manually driven service) – but it would otherwise massively increase capacity on the network and make it more convenient for many more people. So would it be the right thing to do? This is probably more of a question for The Moral Maze.

15. When jet planes first emerged, typically they were required to have four engines to travel long distances, because the assumption was that if one engine malfunctions, the other three will give pilots a better chance of reaching an airport to land safely at. More recently, cheaper to operate twin-engine planes have been allowed to cross oceans and fly much further away from airports because of proven reliability and confidence they can glide for much further – with no uptick in accident statistics since.

16. And I’m sure that the government could bleat on about job creation, STEM investments, and the economic boost that science has etc etc.

17. This is a total aside, but one great “what if” question is I wonder what would happen in a universe where Thatcher didn’t close the mines, and the coal industry continued to be a big deal to this day (let’s imagine coal deposits didn’t run out too). What would Labour be saying about Net Zero and climate in that circumstance?

18. Whether that’s retraining opportunities, bigger hand-outs or UBI, is beyond the scope of this already very long article about buses that you’ve been reading. But don’t worry – there’s only a few hundred more words of bus chat to go.

19. One example of this might be construction, as building sites almost by definition are not controlled environments like roads – so driving diggers and vans and the like is still going to be a largely human pursuit. And better still – perhaps we could create loads of new construction jobs by building tonnes of houses?

20. The buses in the Scotland trial will apparently actually be running on diesel, but I’m pretty sure you can use your imagination here.

21. Even if I can do 90% of the journeys I need to on a bus, it’s the last 10% of edge cases that will persuade me to own a car – and then by owning the car, I’ve a big fat incentive to drive everywhere sat in front of my house. So the way to make public transport popular is to make sure it is suitably expansive.