Stephen Cass: Hello, I’m Stephen Cass for IEEE Spectrum’s “Techwise Conversations.” Congestion on city streets or mass-transit systems would be much less of a problem if more commuters were willing to shift their travel to off-peak times. Some cities, such as London, have tried to address this problem by charging drivers a congestion fee during busy times.
But Balaji Prabhakar, a professor in the department of electrical engineering and computer science at Stanford University, believes that rewarding drivers for good behavior gets better results than punishing them for bad behavior.
To this end, he has created a so-called nudge engine. This system uses sensor data to qualify participants for entry into an online lottery system: The more they drive off-peak, the more entries they get in the lottery. Professor Prabhakar has demonstrated his nudge engine with car drivers in Banglore [PDF], India, and on the Stanford campus, and also with train riders in Singapore. He’s also used a version of the system to encourage more exercise among employees of Accenture.
Professor Prabhakar joins us today by phone from Stanford. Balaji, welcome to “Techwise Conversations.”
Balaji Prabhakar: Hi, Stephen. Nice to join you on this call.
Stephen Cass: So local and state governments have used things like fines to enforce good behavior for a very long time. What’s wrong with punishing people for bad behavior?
Balaji Prabhakar: It’s not bad. Sometimes it may be necessary when you have to guarantee that roads aren’t loaded by more than a certain number of vehicles at a given peak time. It is simply that it is also valuable, and perhaps even more expedient in terms of getting commuter buy-in to use incentives. And the goal of our work has been to show that incentives on their own, without an accompanying charging mechanism, could shift the behavior of commuters, and not just from peak time to off-peak time, but also from single-occupancy vehicles to carpools, and mode shifting away from cars into bikes, and onto walking. So that’s what we believe, and we’ve been validating it in a number of experiments.
Stephen Cass: In recent years there’s been a lot of work on so called digital gamification technologies that reward good behavior, but they typically do so in very well-defined and consistent ways. So if I drop my weight by a kilogram, I get a badge, and maybe some points toward a prize. But your system uses a more random approach toward it. Why is that more effective?
Balaji Prabhakar: Well, if you look at how rewards need to be structured so as to be cost-effective, that is against a goal of achieving some amount of shift in a measure of performance among a population. How many dollars do I need to spend? When you look at that, often you end up with asking for a lot of behavior and being only able to reward with a small amount of money. And if you say, “Walk 100 miles” or “Walk 500 miles,” and then let’s say I’ll give you [US] $50 or $100, the amount of time someone has to wait to earn a reward is disproportionately large relative to the size of the reward itself.
If, on the other hand, we could find a way to reward them earlier, so you walk the first 5 miles out of a 100-mile goal and you stand to win some money, and that could be anywhere between $1 and $100. And sometimes you don’t win anything, and so the chance of winning something soon, and the chance that it is large, could motivate you to just make this mini milestone, meet as many milestones in a long journey, and that’s one of the things that underlies our approach.
Stephen Cass: So how effective has it been in nudging places like Bangalore and Stanford? And we also talked a little bit about the work improving health at Accenture.
Balaji Prabhakar: Yeah, so let me first give you some quick context about how many participants there were and the goals of each of these things. So in Bangalore, we ran a six-month pilot for the employees of Infosys Technologies, which has about 20,000 employees in [the] Bangalore office. And they are, relative to the general Bangalore population, quite well-paid. And our goal was to see if modest amounts of money could be used to incentivize them to shift from peak time travelers to off-peak travel. And about 14,000 employees of Infosys were participating in our program, and we shifted the commuting base of peak time travel to off-peak by about 17 percent. And the amounts of money we had to pay amounted to about 20 [U.S.] cents per employee per week.
And so it is difficult to imagine anybody shifting behavior for just 20 cents a week, but when given the form of lotteries or raffles, the prize moneys were at a minimum of $10 to the winner and a maximum of $250, so they were given prizes in that range. And the Stanford project is for our drivers, and we would like them to shift in time away from peak to off-peak, in mode from cars to biking or walking, and what we have seen here, this is an experiment that started in April 2012 and is still ongoing. But as of March 31 this year, so two years into the program, we have seen 15 percent shift from the peak to the off-peak time in our commuting population in Stanford.
And with Accenture employees, about 3,000 people signed up for a program that encouraged them to walk more. So this was called Steptacular, and it ran for about five months, and they each had a pedometer that measured how many steps they took each day, and what we saw was that over the course of the five months, the participants walked an average of 10 percent more steps. And the thing we found—the Accenture project, this was in 2011—that was also interesting, in addition to the monetary rewards, a social influence had a great effect. So Accenture employees who had invited their friends to “friend” them, so to speak, in the online program, had walked a further 10 percent, or 10 percent more, than those who didn’t have friends. So relative to each group, there was a 10 percent shift. And so this was consistent and significant in the way that the participants reacted to both monetary nudging as well as social influence.
Stephen Cass: So beyond transportation and wellness, are there any other domains where you think this technology will be useful?
Balaji Prabhakar: Yeah, we ran a one-week recycling-encouragement program as part of a freshman seminar class that I taught. This was in the spring quarter of 2010. And many states in the U.S., and certainly California, plastic and bottle recyclables have a monetary value attached to them. So when you buy, let’s say, a bottle, a six-pack of juice or water, or something like this, then each unit has…you’ve paid what’s called the redemption value for that unit. It’s 5 cents in the state of California for small units, and 10 cents for large units. And when we finish consuming whatever it is that these containers had, we tend to recycle because it’s been made rather easy for us. We don’t tend to recycle, typically, to go for the monetary value, because that’s so small: It’s 5 cents. And our goal in this recycling project was to see if through a mechanism of raffles and lotteries we could actually induce people to consider recycling.
So when you walk by a soda can or a bottle of water that’s empty, you don’t just see it and think of 5 cents, because that's its value should you recycle it. It may potentially be valuable to the tune of, say, $20 or $100. You just don’t know it. And so this alone might induce you to go and recycle—that was the thing we tested. And we found that when people are given a choice of what dollar value they would like to recycle for, ranging from $1 to $100 in sort of $1, $5, $10, $20, $50, and $100 steps, we just went off the notes that are available in public circulation, and we saw that monotonically, people, few people chose the $1 option. And we also gave them the 5 cent option by the way, okay?
Few people chose the $1 option. Very few people chose the 5 cent option. A few chose $1, and so forth. And then $100 was chosen by the most. So this is something that is another domain, where the rewards for exhibiting good behavior, when they’re small, could raffles amplify the effect of the reward? So it’s pretty widely applicable where we have to make choices, and these choices matter, either to us personally and/or to society at large, to help us understand and make judicious choices. Nudges are a very powerful tool.
Stephen Cass: So what challenges do you face in scaling the approach up to bigger projects and populations?
Balaji Prabhakar: So if you consider what it means to run, let’s say, a citywide transport nudging program, the most important thing is to make sure that you have technology that senses behavior accurately and cheaply. So if someone has taken the pains to carpool, then you have to find the mechanism that detects that carpooling is being undertaken, okay? Or if someone has taken the pains to park at a more distant parking lot because the more proximal parking lots there at the workplace or wherever they’re going are more crowded, then you have to detect that cleanly and accurately. So sensing is very important.
Now, sensing at scale implies that some form of sensing must be ubiquitous, so smartphones are a very good sensor platform because they have a very large array of sensors in them, and secondly, once the data is sensed, it has to be sent back to some back end, and so it implies the existence of a network. And again, smartphones have built-in networking capability, so sensing at scale is very important, accurately and cheaply, and secondly, scaling the back end.
So when millions of records come in and you’re trying to process them, and sometimes a lot of these things have a real-time quality, for example, you’re waiting at a bus stop or a train station, and the bus that’s approaching you currently is looking full, but right behind it, let’s say 2 or 3 minutes behind it, is another bus which isn’t that full. Now the system has to, essentially in real time, issue you a nudge that says, “Don’t get into the bus that’s in front of you, but rather get into the bus behind it, and then you will earn some 50 points, for example.” Okay?
Now to do that in real time, and we’ve demonstrated this, it’s quite a nontrivial amount of sensing and accurately making recommendations that’s required. If you make poor recommendations, people will simply lose faith in the underlying technology, or even, therefore, the mechanism itself. So scaling it up implies sensing technology, high-speed back-end processing, real-time nudges, and so forth.
Stephen Cass: Well, that’s fascinating, Balaji. Thanks so much for speaking with us today.
Balaji Prabhakar: Thank you. It was wonderful talking with you too.
Stephen Cass: We were speaking with Balaji Prabhakar about using nudge engines to promote better behavior. For IEEE Spectrum’s “Techwise Conversations,” I’m Stephen Cass.
This interview was recorded 3 June 2014.
Audio engineer: Francesco Ferorelli
Segment producer: Barbara Finkelstein
Photo: Getty Images
NOTE: Transcripts are created for the convenience of our readers and listeners and may not perfectly match their associated interviews and narratives. The authoritative record of IEEE Spectrum’s audio programming is the audio version.