At the corner of humanity and automation
We’ve all heard that an era of ubiquitous automation is coming. And we see it happening. Bit by bit, job by job, new automation technology is migrating across the territory of work. It seems inevitable that machines will be able to do just about any job that a person does, right? Yet for many potential roles, humans stubbornly run circles about any would-be machine replacement. Figuring out what work humans are good at (or perhaps what work is good for humans) remains a perplexing and open challenge.
In the midst of all this change, history suggests a few constants. Automation is winning: in aggregate and at scale. People, simply put, are exceptional. We are diverse and we are immensely adaptable, and we are extraordinary in mundane and grand ways.
I’m a recent transplant to the greater Boston area, and I am constantly struck by how bad traffic can be at rush hour. On the road leading away from our research campus, wait times at the traffic light can exceed 20 minutes on a late weekday afternoon. I will not repeat here the profanity I have heaved at that traffic light, to, I might add, no effect whatsoever.
But that light has gotten me thinking. In the Midwest where I grew up, driving was a necessity and a rite of passage for teenagers. Most of the routes I drove were punctuated by four-way-stop intersections. When you saw a traffic light, it meant you were in town. I now realize that town’s traffic light and all of the other ones on our roads have been one of the most familiar harbingers of our automated future.
I take some pride in the fact that traffic lights were a Midwest invention, a simple piece of automation formed in response to the chaos created with the rise of the automobile. Cars were bigger and faster than horses or bicycles or pedestrians, and in the early days of the automotive transformation it was hard and hazardous for all parties of this mix to navigate through the limited real estate of an intersection.
The introduction of the first automatic traffic signal in 1914 imposed order on this chaos, improved safety, and spread from concept to scaled presence in every major American city in under 15 years. The burgeoning automotive industry drove a startup craze, with dozens of manufacturers competing and innovating in the space. A century of research and development has made traffic lights better, of course. Intersections now have sensors that monitor traffic, control systems and algorithms that adjust timing with demand, intervention systems that give priority to emergency vehicles, and even networked means to coordinate with other traffic signals.
So it might come as a surprise that on the really busy afternoons, the Cambridge police department sends a traffic officer to the Alewife intersection. It turns out that humans beat the performance of automatic signals when traffic is irregular or heavy. You see, intersection performance is all about moving cars safely, quickly, and efficiently across those common choke points where one more roads cross over each other. Compared to today’s most sophisticated, sensor-riddled, algorithm-driven traffic light, a good traffic officer sometimes still can deliver superior results because he or she has a better sense of context, a human touch for motivating drivers, and a more nuanced commonsense of overall traffic flows in the city.
And yet, scaling up the deployment of human traffic officers to take over the automated duty of traffic lights at every rush hour everywhere is impossible. Furthermore, when traffic is regular and evenly distributed, or when the human traffic director isn’t highly skilled, even a simple four-way stop wins out. If you are skeptical about that, I recommend watching a humorous but convincing demonstration of this performance principle in a 2014 episode of MythBusters. At scale, and in aggregate, automation wins.
Traffic lights seem like natural parts of our landscape, but it wasn’t like that in the early days. It took time for human drivers to adapt to the traffic light. When these lights were first introduced, they didn’t get respect and motorists ignored them in selfish attempts to cut down their individual travel time. Indeed, a prescient Cleveland teacher introduced the game “Red Light, Green Light” to teach the youngest generation how to cooperate with this automation and grow towards a safer, more collectivist driving environment.
What seems so simple and hardly worth thinking about — a three color switched light to get city traffic moving smoothly — turns out to be monstrously complex. The needs and impatience of any given driver must be balanced with the good of the city as a whole. With perfect knowledge of car positions and destinations, it is possible to compute optimal traffic-signal timing for a single intersection. However, as soon as you are dealing with multiple intersections — as with any city — finding the traffic light timing that best balances waiting for any given car while maximizing citywide throughput becomes computationally intractable — it is what a computer scientist would call a deterministic, exponential time-complete problem. Even so, researchers have managed to come up with clever signal-timing algorithms to approximate an ideal, balanced solution and they theorize that a truly smart urban system of traffic lights could improve throughput by 20% by combining information from the traffic of one intersection locally with citywide.
As a technologist, it’s tempting to focus on advancing automation and algorithms as powerful tools to tackle tough problems like traffic. It takes some creativity and reflection to realize that the most powerful means to improve intersection performance approach the same throughput problem from entirely different angles and can blow away even the smartest algorithm. Carpooling, bicycling, and walking provide an immediate integer-factor throughput increase. Roundabouts are superior in many scenarios, especially in coupled networks. Perhaps the best solution of all is to eliminate the contested terrain of the intersection entirely, with the magnificent if expensive cloverleaf interchange. These ideas are all driven by insight — by viewing the problem from another direction, and human ingenuity.
In the decades that come, our relationship with automation will evolve, and humanity will adapt rapidly, both within and across generations. I think this is will be true for not only for traffic management, but also for banking and retail and defense and other industries. We will see new and powerful algorithms emerge and a rapid adoption of new forms of automation. At scale and in aggregate, automation will win out. However, much like the engineer — James Hoge — behind the original traffic light, exceptional and clever entrepreneurs will zoom out, and reframe problems find entirely novel solutions to nagging problems. But my bet is that in 20 years there still will be a handful of talented traffic officers in Boston and other urban centers who will step into the middle of intersections where they still will have a sense for exceptional traffic like no machine can.
Dan Patt is the Chief Executive Officer at Vecna Robotics and a Nonresident Senior Fellow at the Center for Strategic and Budgetary Assessments, and advises two academic institutions.
At the corner of humanity and automation
Research & References of At the corner of humanity and automation|A&C Accounting And Tax Services
Source
0 Comments