A few weeks ago, I woke up in San Francisco to drive down to an 8:30am meeting in Matrix's Palo Alto office. It was 7:15, and Uber showed an ETA of just after 8- plenty of buffer for unexpected traffic, or so I thought.
As I slid into the backseat, the driver (rated 4.9*) gave me a worried look when I told him what time my meeting was. "You're in trouble" he said, "the app is wrong." As we made the drive, it became clear he was right. The ETA started slipping as we sat in a parking lot-esque traffic jam outside SFO and before long Uber predicted an arrival time of 8:45. I was looking every bit like a naive Boston VC, unaccustomed to the hellfire of Bay Area traffic and about to be disastrously late to a meeting.
As we neared Palo Alto, the ETA stayed steady at 8:45 and my anxiety level rose. Google Maps showed a suggested route that took us off the highway and onto side-streets well before Palo Alto, circumventing a massive jam on the Bayshore Freeway as it met the Dumbarton Bridge. I mentioned it to the driver, as he wasn't changing lanes to take the exit Google Maps recommended, and he shook his head. "Trust me." I didn't really have a choice but I'll admit I didn't trust him: instead I kept my eyes glued to the screen waiting to see the ETA balloon past 9am once we passed the recommended exit.
It did, briefly, but then my Uber driver's insight bore fruit. Sure enough, there was an empty exit-only lane on the right side of the highway and we blew through the dark red section of traffic on Google Maps at 70mph. I watched with relief as the ETA plummeted down, and I arrived just in time for the meeting. Uber's tipping option had never felt more appropriate.
On the surface, this is a pretty boring traffic story, and I usually try not to be a traffic story guy (I'm a big fan of Chris Rock's skit on married people in which he trashes such tales as the epitome of marital boringness).
Digging deeper, there is a lesson in my nerve-wracking drive about the future of work. No one doubts that Google Maps is more accurate at delivering directions than all but the most experienced humans. But at the same time, my ride was powerful anecdotal evidence that adding a human into the mix can add value. In this case, it added a ton of value: not only did my driver know that I was at risk of being late despite a way-too-optimistic Uber-provided ETA, he also saved the day by disobeying an algorithm he knew was wrong to save me from disaster.
The insight that this experience underlined for me is that there are three phases of the automation of a given task. Initially, humans go it alone with very little help from machines. Think humans + paper maps. As machines increase in capability, they help more and more until eventually they are superior to what a human can do on its own. I call this phase two automation, where machines are better than humans but still can benefit from human assistance in subtasks. This is the phase navigation applications are in today, arguably a decade or so after they entered phase two. In phase three, humans are simply too outclassed to be of any practical use to machines at accomplishing the task: sweet obsolescence has set in. For some tasks like long division, this has been the case for decades already (not that anyone is complaining!).
Happily for humans, phase two can last for a surprisingly long time. In Tyler Cowen's excellent Average is Over, he notes that despite the top human chess players routinely losing to machines since I was a toddler in the late 1990s, for the past few decades machine + human combos have been the best chess players in the world in a variant known as freestyle chess. We may now be on the cusp of phase three for chess, but it was a pretty good run for humanity- we had some useful skill at the game long after Deep Blue beat Kasparov. Newsweek ran a cover story around that match labeling it "The Brain's Last Stand," but the reality was quite different. The brain's true last stand in chess is happening now-ish, in the much less well publicized but still vibrant world of freestyle chess.
The takeaway here is that there's plenty of room for optimism about the employment prospects of people, even if moving from phase one to phase two brings disruption (and it will). We don't have to be better than machines at any given job to be gainfully employed helping them with it, and the mid-term future of work will see more and more of us making the gradual transition from computer-aided workers to workers aiding computers. Our inevitable total obsolesce is a long way off yet.