Uber and Out
“Freaks become norms, and norms become extinct. Monster by monster, evolution advanced”
― Siddhartha Mukherjee, The Gene: An Intimate History
Without a car, Los Angeles is as navigable as the moon and yet with one getting around feels only marginally more efficient than wading through wet concrete. However last week I found out that the experience of navigating this particular urban sprawl was transformed with Uber. The upstart ride sharing service just works everywhere. However in spite of the transformative effect of Uber on navigating cities, I believe in future it will be considered a canonical example of Silicon Valley folly. I appreciate that Uber is one of the iconic entrepreneurial companies of our time and a lot of people who know a lot more about investment than me are much more positive about Uber but here is why I disagree.
Additionally, I have a particular interest in this subject as I believe the limitations of the taxi industry 20 years ago are the limitations of healthcare’s primary care system now and just as tectonic shifts in technology have enabled the creation of a distributed network of drivers and riders so will it become possible for patients with long term conditions and clinicians managing large panels of them remotely to interact in a machine optimized distributed network. Just as taxis are now available on demand with minimal latency so will it come to pass with care where it will no longer be necessary to store problems on aggregate till the next visit — the equivalent of only being able to hail taxis from the taxi rank at certain times. At Wellframe we are working on using technology to organise parts of the health system dynamically around the needs of patients as those needs emerge. At the moment a lot of my recent work is in conceiving AI systems to address needs before they occur — the concept of negative latency which I will go into later.
Like all great entrepreneurial ideas, the idea of “ride sharing” seems obvious in retrospect but imagine looking at the market for taxis 20 years ago. At this time the only way of matching supply of taxis to demand for them was if someone wanting a taxi goes to a taxi rank or happens to see a taxi driving around and hails one or “hires” a telephone dispatcher. At that point one would think that if two new pieces of information were available this supply/demand mismatch would be much easier to reconcile. Firstly a driver can exist in three states: inactive, active and looking for a ride and active and on a ride. A passenger has three complementary states and both have a location at a point in time. It was not at that time possible to get this information without significant active human input and as such it was not possible to get it with any frequency. The result was that neither party knew which state the other was in unless they met in person and as such there was a suboptimal equilibrium and a solution that was worse for passengers than it needed to be.
Even if it had been possible to dynamically and passively get this information, that would not have been enough to make the market function optimally as each passenger would need to know that the driver was to be trusted (and vice versa) and would need some method of paying him or her with low transaction cost. Now of course the combined effect of advances in GPS, digital mapping, smartphones and mobile banking changed the technology variables of this market profoundly. Once all the customers and drivers had (smartphone) computers with sensors and could continuously identify their state to a distributed system and those same computers were connected to the internet enabling them to mediate financial transactions alternative solutions became possible. And hence there was an entrepreneurial opportunity. Timing is invariably everything.
Love them or loathe them, Uber are everywhere and it seems that no amount of driver abusing or flagrant executive ass-holery seems to reduce the obvious appeal of the core experience. However, I feel that the engine of growth is much more fragile than it seems and just as tectonic plates shifted to enable Uber to form such are they shifting again to fashion their demise. But firstly I would like to look at their current business in their current market.
Firstly, let’s address the elephant in the room if that elephant is in fact a bull and that bull has just done a number two in the middle of the room that we are now looking at. This is no more ride sharing than having a colonoscopy is endoscope sharing. And similarly, to stretch an already weak analogy beyond breaking point, whilst it is inadvisable to receive a colonoscopy from someone who is not a professional, Uber drivers are professional drivers in everything but name. Pretending otherwise, might stave off legislation in the short term but I really wish for everyone’s sake that the Uber leadership would stop. But the reality is that they can’t and arguably it is difficult to blame them for this. They need to maintain the ride sharing charade to be viable as legislation typically moves slower than innovation.
Secondly, it is concerning that their drivers are infamously not employees but consultants. This means that they don’t have to pay benefits or incur regulation as a transport provider as they are able to make the argument that they are a technology company that enables people who want to share a ride to do so. Without universal basic income (or more extensive social provision) the propensity of this employment-lite model to shield the gains of economic production from the people supplying their labor and forego this shield of protection from costs of unpredictable event’s that limit’s someone’s ability to work is regressive and not in anyone’s long run interests.
My friend Bill Richards recently remarked that he had never met a driver that likes driving for Uber. As far as I am aware this is reflected in the retention of Uber drivers as well. Reports are that turnover is high and it is increasingly difficult to recruit new drivers. In a way this does not matter as they are the biggest game in town (you could exclusively drive for Lyft or any number of other smaller competitors but it seems like you would invariably make less money so most drivers don’t) and for each driver that leaves there are many more who will join. But they know that such turnover has implications for service quality and also for consistent availability of rides — the very convenience that is Uber’s hallmark.
In the US, Uber has developed fleet arrangements with car makers and are able to supply vehicle finance to drivers who realise part of their payment by maintaining a certain number of hours driving for uber per month. Tellingly the net cost of leasing a car in this way from Uber is equivalent to the cost of doing so privately from the car maker which is unsurprising as they both use the same financing companies as intermediaries. It seems that their exposure to bad debt and hence their cost structure is the same as for a typical vehicle financing organisation which is ironic as arguably it would be lower if their drivers were employees rather than contractors…
Uber, though the largest is not the only ride sharing company and there are a plethora of copycats large and small with subtle variations on the central theme. Uber is fighting for a monopoly position in the ride sharing market by executing a predatory pricing strategy where they are using their greater access to capital to run at a loss and in so doing run rivals to bankruptcy such that in the long run they can assume an unassailable monopoly position. Typically this is inadvisable unless something about the product/service itself is truly unique as at some point Uber will run out of capital at which point it will be relatively straightforward for new entrants to displace them. In fact Magellan Financial Group recently referred to Uber’s financing strategy as a “Ponzi scheme.” Strong words.
But these issues collectively affect how Uber will reach what I argue is a local maximum. However the tectonic shifts that enable autonomous cars profoundly challenge the likelihood of Uber reaching the more ambitious and distant general maximum as stated in their mission statement of (making) “transportation as reliable as running water for everyone,everywhere.”
I believe in the next 5 years there will be a natural evolution (freaks become norms!) from the internal combustion engine powered car of today to the autonomous (electric) car of tomorrow. This evolution will consist of increasing automation of driving functions ( collision avoidance, lane keep assist, dynamic cruise control, parking and highway driving then urban driving) and increasing electrification (hybrids to plug in hybrids to full electric powertrains). In fact it may be argued that car will be a misnomer for this vehicle. Such a vehicle will arguably have more in common with a current mobile phone than a current car: high density battery, display, intuitive interaction language, sensors, graphics processors for massive parallel computation, mobile CPUs, connectivity and reciprocal dependence on cloud resources.
The advent of automated cars will make private ownership and usage of cars an anachronism. Either automated cars are owned by a third party (such as a car maker or even a city government) and rented out as and when they are needed or private ownership persists and your car makes money for you whilst you are away by doing Uber “itself.” Either way it is not clear to me why the current Uber model makes sense in either of these scenarios and as such this becomes a discussion of how successful Uber are likely to be in creating autonomous cars themselves or becoming a fleet provider of autonomous cars.
The protagonists in this autonomous car race are technology companies (Waymo/Google, Baidu, and maybe Apple), tech-enabled transport companies (Uber, Nutonomy and Didi) and car companies (Tesla and everyone else(!))
To be successful an autonomous care company has to be able to deliver on technology and business model.
Technology
Ability to make and market a car
Batteries
Motors
Sensors
Onboard processing
Communication with cloud processing
Cloud processing
Learning algorithms on car and in cloud
Data set of different driving scenarios to feed above
Mapping
Business Model
Hybrid of private ownership and on demand/micro rental
Sales network
To cut a long story short, I think different companies will succeed in different areas here and the company that aggregates the hardware, software, data value chain most effectively will most win overall in the long run. In my mind there is only one likely winner here and it is not Uber. I think Tesla will take an early lead in autonomous cars but will resolve into a mass market battery company by way of a niche car manufacturer (this is a very clever strategy) and existing car makers and their suppliers will become very good at manufacturing and selling autonomous vehicles consisting of other people’s software and hardware (like they do now.) Overall I believe that Google/Waymo will win as they will aggregate the technology value chain, be able to scale by running on anyone’s cars (rather than just their own) and leverage synergies in their existing class leading cloud services platform and machine learning architecture. Everyone else is further behind and from the computing pov are going to have to run on Google’s or Amazon’s infrastructure anyway so are unlikely to be able to compete in the long run.
Google recently pulled back from producing cars themselves to focus on what they do best and I believe there is magic in this strategy. Simply if they can focus on being at least as good at the sensors and have the best performing algorithms and they can provide all of the software elements of autonomous (basically the OS for a autonomous car in effect) to people who are good at making cars, in turn relieving them of the R+D burden of tackling a software problem, then it is likely that they will achieve a predominant position. Effectively whilst Tesla will only be able to scale as fast as their own cars sell, Google will be able to scale at the rate that licenses it’s hardware and software scales. Greater scale, means more data and better algorithms. Like all instances of AI, bigger is better and every increment of data generates in turn more value. At the same time Google, with its push into cloud services will be able to generate subscription revenue on this “autonomy as a service” model. This all goes to say that I think Google will win and you will see autonomous cars by a number of different manufacturers dependent on Google technology in the mass market.
However autonomous cars collectively will not be able to overcome the physical constraints of urban environments where it will be reasonable to expect more congestion as people switch from a relatively small number of high density (public transport) vehicles to many more low density (privately owned) vehicles within the same congested urban streets with inevitable consequences. In turn as autonomous cars grow in number without intervention there will be less investment in public transit as city governments in response to reduced rider volume reduce investment in infrastructure, thereby increasing the appeal of private transit until you basically have LA everywhere.
I think an ideal future would involve public transport itself being remodelled in the image of Uber with autonomous high density vehicles taking routes dynamically organised according to peaks and troughs of demand and user preferences regarding pick up and drop off. Of course human logistics is just one flavor of general urban logistics and such a fleet of vehicles could also be used for deliveries of other things. Either way, transport infrastructure, like health infrastructure is being profoundly remodelled as leaps in computation broaden possibilities and at the same time focus attention on the information problem within. Whether Uber succeeds or not is ultimately a sideshow distraction to the main event which is cars essentially becoming computers and in turn transport systems becoming one instance of a computing problem: optimising the delicate interplay of sensor data and vehicle action for individuals and for cities as a whole.
Just as the uber model significantly reduces the latency between the call for a taxi and the taxi arriving in response so are we aspiring to reduce the latency between a patient identifying a need and having the right clinician respond in the right way. This co-ordinated onslaught on latency is, outside of the automation of the cognition involved in sensing and classification, the most interesting and transformative application of existing AI. It reminds me of a (possibly apocryphal) conversation between Larry Page and the early Google engineering team. Each week the latency (the response time between a user entering a search and Google delivering a result) fell and each week Page goaded the engineers to go further and solve more complex problems until someone rebuked: “what are we going to do when we get to zero?”. “Why stop at zero?,” he replied.
