It will be surgery.
So something a bit lighter today, after a few academic-y posts. I want to look at the received wisdom that the visual medical professions (radiology, pathology, etc.) that will be the first to suffer the cruel sting of automation.
This does make sense, because AI technologies excel at visual tasks. But I think another area of medicine might experience disruption even earlier. Quite soon, in fact.
That area is trauma care. The surgeons, emergency responders and clinicians, intensive care specialists, rehabilition specialists, and neurosurgeons in this field may be in for some shake-ups. Especially if they work in a big city, and especially if they work in a big trauma hospital.
Now, this is certainly not my area of academic expertise. I’m speaking here as a technology enthusiast/futurist who knows a fair bit about AI and automation in a general sense, and a radiologist who interacts with trauma specialists regularly. That said, I think this prediction is a pretty good bet. In fact, only one factor needs to fall the way I expect it to for this to be a near certainty. Let me explain.
The elements of disruption
I talked about the factors that make a technology disruptive before in Understanding Automation, but as a quick recap:
- To disrupt a workforce, work tasks need to be automated. Even if whole jobs are not replaced, the impact on the workforce is roughly proportional to how much of a job the technology can do. Replace 40% of tasks, and you need something like 40% less workers.
- If demand can increase as costs fall, this can mitigate the workforce disruption.
- For disruption to be noticeable on the ground (i.e. actually disruptive) it needs to be fast enough and severe enough that it overcomes attrition due to retirement, and starts to reduce training positions or even leads to job losses. This is a very high bar, since most medical specialities see retirement rates of 1-3% per year.
So, you must be thinking, what surgical and trauma tasks could possibly be replaced by AI? Robot surgeons are nowhere near ready, and directly replacing ICU or rehab clinicians is not even on the horizon right now.
I totally agree. There is no supply-side replacement going to happen in these disciplines. But there is another factor that determines how much work there is to do: demand.
As I said in my 2017 summary, the most transformative AI achievement in the last year was that Waymo (a subsidiary of Alphabet née Google) announced they have level 4 self-driving cars they are ready to test on public roads. Level 4 means that no driver is needed at the wheel.
This is amazing for several reasons. Firstly, Waymo is way ahead of the competition. Most of the competing companies (especially the car manufacturers) were still working towards producing level 3 cars in the 2020s when Waymo announced this.
Secondly, Waymo is believable. They don’t mess around with high risk strategies. They have been testing their cars for years longer than most other teams, with over 5 million miles on real roads (many other teams who claim to be working on this aren’t even on public roads yet). They have much fewer disengagements per mile driven than their competition. If Waymo says they are ready for this, I see no reason to distrust them.
Waymo leads the pack by miles driven on Californian roads in 2017, despite the fact most of its self-driving car testing is in Arizona. Waymo actually recorded almost 2 million miles in 2017. Tellingly, GM doesn’t say how much testing it has done on roads outside California. I suspect the number in the above figure is close to the total, which is about 7% of the miles Waymo put in in 2017.
Thirdly, Waymo has the financial and political reach to make it happen. They have already planned a large taxi fleet of “thousands” of cars for Pheonix, Arizona.
Finally, their business model is right for a massive disruption in driving, because they aren’t offering cars for people to buy.
Too fast, too furious
Since sluggish automation isn’t actually disruptive, the big question is whether level 4 autonomy will disrupt driving fast?
It won’t if everyone owns their own cars. Retro-fitting a huge variety of vehicles with autonomous control systems, or waiting for everyone to upgrade to self-driving tech, would be a very slow process. It would take many, many decades. A lot of people won’t even want to own a self-driving car. It is certainly unclear what the incentive will be to buy one, since they will cost more and because population statistics about safety never really sell cars.
But no futurist is thinking people owning self-driving cars. Instead, they all think we will have fleets of on-demand taxis to take us anywhere we want to go because it will be cheaper to ride autonomous taxis than it is to own a car.
This is the single factor that needs to fall into place to make my prediction a reality. If self-driving taxis are significantly cheaper than owning cars, people will use them at capacity, as soon as they are available. For most people the cost of running a car is their second biggest expense, and can be very high even before you include costs like garage space and the time you lose driving.
It is pretty simple maths. It costs you a set amount per year to drive your normal distances, to repair and maintain your car, your registration, depreciation, and so on. The real kicker is that many of these costs stay nearly the same whether you drive your car or it sits in a garage.
And we don’t drive our cars much at all. The average car is parked 95% of the time. A huge chunk of car ownership cost, 65% in the above figure, is spent when you aren’t driving it.
A traditional taxi fleet is the opposite. No idle time, but a heavy cost per mile to pay the driver, the taxi company, the licensing body, and so on.
So what about an autonomous taxi fleet? No idle time, no additional costs. No driver, efficient/automated routing, no licensing. The upfront cost is the same (or slightly higher), but the cost per mile is the same as owning your own car and the vehicle is 20 times more productive!
It gets even more extreme when we realise that any autonomous taxi service will be made up of electric cars. Electric cars are a great example of a clearly superior technology which will take a while to saturate the market. Most people can’t justify buying a new car to save some of the 35% of car costs that involve fuel and servicing.
A taxi fleet has to face up front costs either way, so it makes complete sense to go electric. In fact, it probably makes sense to set up automatic charge points everywhere powered by solar panels, just for your own taxi network. The cost per mile would be nearly nothing. So not only is the car 20 times more productive than owning your own car, the 35% costs of fuel and servicing drop to almost nothing as well.
And this is what will mean that autonomous taxis take over sooner rather than later. Self-driving taxis should be dramatically cheaper than owning and running your own car. We are probably looking more at a Netflix style subscription than we are a traditional “pay-per-drive” taxi service.
All the people who are saying “I’ll never give up my car” are either rich enough to waste money, or they don’t understand this yet. No-one who struggles to pay their bills will be able to afford to drive.
So you can probably see the shape of my argument now. Self-driving cars are coming soon to a big city near you. As long as they are cheap, they are going to massively and rapidly disrupt transportation. And they are going to be much, much safer than human drivers.
Over 90% of traffic accidents are caused by human error. Whether it is drink driving, inattention, speeding, straight up bad driving, or any other of a myriad of reasons why people crash their cars, the fact is humans are really dangerous on the road. In the USA we average an accident every 200,000 to 500,000 miles, with a fatality every 200 to 500 accidents. This rate can be as high as 3 or 4 times as high in other parts of the world.
What do we know so far about self-driving cars? In highway and city conditions in 2017, Waymo’s cars drove 1.7 million miles with 14 minor accidents, 13 of which were caused by humans in other cars. That is a single accident caused by their self-driving car in almost 2 million miles. It isn’t enough data to make any solid conclusions, and a human safety driver was always present in this testing, but the actual rate of crashes caused by Waymo’s self-driving cars is 75% to 90% lower than the lower estimate for humans.
Blue and purple lines are the upper and low estimates for human and Waymo crash rates (where the Waymo upper estimate includes all 13 accidents caused by human drivers in other cars). The green line is how often Waymo cars hand over control to their human safety drivers. For the purple lines ignore the x-axis, the trend over time is not given (just the overall rate in 2017).
Since we are talking about the transition to being truly driverless (level 4 autonomy), we do need to recognise that they still “hand off” control to a human every 5000 miles or so, and it remains unclear how truly driverless cars will handle these situations. Will they pull over safely, or will they have more accidents? Waymo are starting to test cars without drivers in public, so we will see more data coming in on this soon enough. As I said before, if Waymo thinks they are ready I don’t see a good reason to distrust them.
It is likely that these cars aren’t perfect yet, but it is not unreasonable to think that it is only a matter of time. Many experts believe that crash rates will drop significantly over the coming decades.
This timeline includes an estimate of market roll-out. The first city to get self-driving taxis could achieve a 90% reduction in a decade or two. If Waymo has solved the hand-off problem, their cars might already be at 90% crash rate reduction in Arizonan conditions.
For doctors who specialise in caring for car crash injuries, this could mean a huge disruption. Even if self-driving cars only halve the rate of accidents and fatalities, the impact on trauma care will be dramatic.
Find me a patient, stat!
It is hard to find statistics on how much work motor vehicle accidents contribute to the various specialities involved, and it will be highly variable across different regions, but as a rough guide created by looking at the experience in my local region:
- Trauma surgeons ~ 30% of major trauma cases
- Ortho ~ 20% of fractures
- Rehab specialists ~ 80% of traumatic brain injury and ~ 50% of spinal injury
- Neurosurgery ~ 20% of head trauma and ~ 50% of spinal trauma cases
- ED ~ 7% of presentations, ~ 15% serious presentations
- ICU ~ 30% of beds at any time
I don’t intend for this to be seen as an authoritative list. Each region will be different; as an example, gunshot injuries make up a much larger proportion of trauma in the USA than in Australia (where I am). We should also recognise though that even within a specialty the distribution of trauma is not uniform. Some trauma surgeons will barely ever treat car accidents, and others will do almost nothing else.
These latter doctors are under threat, because if their city is chosen as one of the first for a fleet of self-driving cars, their practice could evaporate within a decade. The retirement rate in any given city can easily be 0% per decade (given that there are so few subspecialised surgeons in any region). Losing 90% of your patients in a ten year period would be catastrophic.
Arizonan doctors in particular, take note. Waymo is already deeply embedded in Phoenix, and planning a wide scale roll-out this year. If you work in Phoenix, and you deal mostly with car accidents, it might be worth considering where you think you will be in ten years.
Don’t worry Arizonan trauma specialists … you can always respecialise in radiology 🙂
There is definitely some schadenfreude in turning the tables on all the “hands-on” doctors who are watching us radiologists and pathologists with the mien of a speedway spectator, waiting for a big pile-up. It isn’t a reason to panic though, the vast majority of trauma specialists will be able to redirect their practices to other patient groups with minimal difficulty, at least for a while.
Importantly, the central point of this piece needs to be part of the discussion among doctors. When we talk about the effect of automation on the medical workforce we mostly focus on medical AI systems replacing doctors, but it is certainly true that technological advances more broadly have the potential to disrupt medicine. AI systems that can help people to exercise more, eat healthier, consistently take their meds, set aside time for beneficial mental health practices, and mitigate risks all have the potential to make very large impacts on human health, very quickly. It may be that a whole range of specialties will notice their patients have just stopped coming to see them, well before we see any robot doctors threatening their jobs.