Google's newsmaking blog post, First Drive, about their prototype driverless cars has created much Internet buzz regarding our societal vision of the transportation systems of the future. Pundits have come down on all sides, from those who believe these cars are decades away to others who see these cars amongst our human-driven machines within five years. Google has created a rhetorically powerful case by completely removing driver controls -- this is a version of extremal design that aggressively tests readers' limits and visions of the future.
There are several worthy tropes that revolve around the question of driverless cars, and these are being given ample press space. Accountability is a big one: what is the division of responsibility between manufacturer, lead passenger and the car itself when blame will, doubtless, be apportioned after an accident? Ownership: should driverless cars be privately owned, or do they offer the thin edge of a massive sharing economy, where utilization rates top out precisely because every auto-car serves the needs of many passengers using a hyper-efficient, optimized, system-wide scheduler? Traffic efficiency: can these cars not only provide greater safety but significantly change the dynamics of dense traffic enough to reduce commute times materially?
But the issue that calls most loudly to me, and has been covered the least, involves the question of taxis. We frequently come across vague statements regarding how driverless cars will revolutionize the taxi industry; and in these broad statements the hidden subtext seems to be that we will fire all taxi drivers, replace cabs with automatic cars, and save passengers bundles of money. It is deceptively hard to respond to these generalizations, but then I encountered John Markoff's article in the New York Times about Google's driverless cars. Markoff cites work by the Earth Institute to quantify the cost impact of driverless cars, and the numbers floored me. Part of the Earth Institute's analysis, which you can read here, centers on individual, private owners; and the savings there is clear from reducing the number of cars per family and so forth. But in the case of Manhattan taxis, Markoff reports that Google's cars would take the cost of taxi service in the Borough from four dollars per mile today to fifty cents per mile tomorrow.
There is, of course, the social impact of erasing an entire category of work from the job pool. Surely, this is an ideal case for serious discourse. When is it appropriate to use technology to make society more efficient, at the cost of massive local job loss? How much of a cabbie's role is social? I, for one, use cab drivers during my trips to great advantage -- that's how I discovered places like Essa's for a quick breakfast in town. You can certainly imagine driverless cabs that still have a cabbie on board, combining human supervision and oversight with hyper-efficient scheduling.
But, the economic impact cited by the New York Times deftly cuts away social considerations because it is such a massive savings that it overpowers the normal balance of reason: you can cut the cost of taxi travel by 88 percent simply by removing the cabbie! So how does the Earth Institute arrive at such an extreme result? The way researchers accomplish the economic savings is, unfortunately, by comparing apples to oranges. They calculate the operating cost of a New York cab by looking at passenger prices and subtracting 15 percent margin, arriving at their $4 per mile target cost. As for the driverless option, they start by expecting utilization rates to go sky-high with automated cars, under the assumption that smartphone-based scheduling means that passenger requests will be serviced so much faster that the cars will spend far less time wasted, unhired. Even if you buy this argument, the surprising part is that the total reduction in required cabs is relatively small, from 13,000 taxis to 9,000 Google cars. This is unsurprising since Manhattan taxis are already very heavily utilized: go take a look and you will be convinced. This efficiency gain will certainly not reduce cost to one-eighth the current value. Next, they compute the operating costs of each Google car by estimating an arguably lowball price tag ($25,000), then dividing the cost, with interest, into a 250,000 mile lifetime. Their number to behold, 11 cents per mile, represents the effective ownership cost of the Google car, before insurance, gas and other incremental costs.
But how can they compare this operating cost to the reverse-engineered cost of a cab, starting from today's fares? Where are the equivalents in that? A quick search on the Internet yields some lease costs for real New York cabs: I see numbers well North of $1,200 per week to lease the cab. That adds up to $5,000 or more per month just for a true yellow taxi lease. And at, say, 400 miles of travel per day, the lease alone adds up to 40 cents per mile of effective cost. And that's before we even start talking about the medallion price! So the price of the vehicle alone is already quadruple the Earth Institute's assumptions about driverless cars, and this has nothing to do with utilization or the profit of cabbies. So the Earth Institute result is not really talking about the cost differential of robot cars; rather, their result is saying that if you remove the medallion system, make taxis use scheduling that is 30% more efficient, and cut lease prices by three-quarters, then you end up saving massively on cab fares. Well, yes. And this all holds whether or not there is a cabbie in the left front seat.
Of course, all this fails to even begin to address the societal cost of automation and job loss; however if we are to antiseptically study the ways in which robotics can help us live more cheaply, let's take care to run the right maths.