5 ms·

Here’s an article about some MIT folks who tried to do this and were shut down by annoying rich people. My greatest frustration is that they complain about bla

Here’s an article about some MIT folks who tried to do this and were shut down by annoying rich people. My greatest frustration is that they complain about black box algorithms when in fact it is a perfectly clear, ambiguity-free optimization function. https://www.google.com/amp/s/www.wired.com/story/joi-ito-ai-and-bus-routes/amp https://www.google.com/amp/s/www.wired.com/story/joi-ito-ai-...

MBCook 7y ago
Real url: https://www.wired.com/story/joi-ito-ai-and-bus-routes/ https://www.wired.com/story/joi-ito-ai-and-bus-routes/
swampthinker 7y ago
The OP article is also about the Boston bussing system changes, before it was cancelled.
blihp 7y ago
It was a good article but my main takeaway was that the city botched the marketing of the planned change. Had they communicated what was being planned and why ahead of time, as well as providing the tool/simulation to make it more tangible to residents, they might have had less objection to the change. If not, they at least would have likely had more support from the majority who stood to benefit to counteract it.
- Spooky23 7y ago
  No, they tried a build perfect, cost optimal system to maximize ROI. They ended up losing money. A solution you cannot implement isn’t good engineering. Big changes that impact tens or hundreds of thousands of people need to be staged out to minimize impact. The Big Bang approach doesn’t work for problems like this — i bet a few thousand people would lose jobs over these schedule changes.
scarejunba 7y ago
It's the same system and it's mentioned in the OP.
- b_tterc_p 7y ago
  I was overeager
joe_the_user 7y ago
Oh I remember this story. Back then when I dug into it, it was a real spin-versus-spin situation. Sure, when you introduce an approach that suddenly creates considerable inconvenience for some population of parents, sure they complain (no one likes having to drop off kids at ~ vaguely recalling ~ 10 am and pick at 5 pm, sheesh, it was something terrible). Some large portion of working people are fairly dependent their children having ordinary school schedules. Maybe these were relatively better off parents in the public school system - where the actually wealthy use private schools (I know Boston well enough to attest to this). So the algorithm failed to "optimize for those they could effectively screw" but hey, would it be too much to ask to just give all schools a sane schedule? Maybe pony up that extra money, give up the need to optimize everything, keep the tail of bus optimization from wagging the dog of school scheduling?
- yardstick 7y ago
  For elementary students apparently “Hundreds of families were facing a 9:30 to 7:15 a.m. shift.” That’s a significant change in wake up time. Yeah, I’m not keen on that. It’s a non-starter.
  - shoo 7y ago
    Most substantive changes leave some fraction of stakeholders strictly worse off than before.
  - Dylan16807 7y ago
    It's a change between years, and students moving to high school are already facing the exact same changes. The disparity can't be a non-starter because it already existed. Those early hours exist. Someone has to take them (without doubling the bus fleet size). And it's better for the students if they have the early hours in elementary and not in high school.
    - elcomet 7y ago
      Well yeah but when you choose a school, you choose it based on those criterias. They cannot just change the schedule like this
superpermutat0r 7y ago
The thing is, all of these algorithms should be able to do well with arbitrary constraints. If it means saving less money then so be it. Many of these systems start optimizing some simple criteria and work fairly well. But with additional constraints they fall apart. Even companies that manage their own fleets for their own needs would like different things to purely optimizing distance/duration. They might want to balance the workload across all of their drivers, separate the drivers so they drive in a particular area, undefined start or end of the driving job. In the case above, something could have definitely be done, if the algorithm was efficient enough.
- mipmap04 7y ago
  I’m not sure I follow. All of those additional optimizations are pretty simple problems in the OR field.
  - superpermutat0r 7y ago
    Undefined start or end time with a limit on driving duration is not trivial. Well, routing school buses is already less than trivial. You have pickup & delivery. So there's a precedence constraint to orders. Child won't be kept for hours in the bus, so the bus has to make several trips to school. Limiting the time between delivery of the child and the pickup is already less than trivial. Just to know if route is feasible (all constraints satisfied) given a list of child pickups, child dropoffs and time limit between pickup and dropoff is nontrivial and can mess up the optimization. Adding school shifts to the equation, minimizing number of vehicles and a bunch of other constraints might make the optimization just too slow or too constrained for an algorithm that was working incredible without all those constraints.
    - JoeAltmaier 7y ago
      In rural Iowa (where I live and grew up) the middle-school bus and high-school bus are full the first couple of days of school, then they all start driving/getting rides to school. At the middle of the school year, maybe 12 people on a bus build for 60. That makes it complicated too. And 'optimization' solutions result in 1hour+ rides to a school just down the road, as the bus winds about the countryside scavenging the few remaining students. The longer the trip gets, the more that find another way, the emptier the bus, the longer the trip to try to fill it again. We drove our kids to school for most of their school careers, even though three busses went by our house every day.
gerbilly 7y ago
Cry me a river for the poor optimization function.

MBCook 7y ago

Real url: https://www.wired.com/story/joi-ito-ai-and-bus-routes/ https://www.wired.com/story/joi-ito-ai-and-bus-routes/

swampthinker 7y ago

The OP article is also about the Boston bussing system changes, before it was cancelled.

blihp 7y ago

It was a good article but my main takeaway was that the city botched the marketing of the planned change. Had they communicated what was being planned and why ahead of time, as well as providing the tool/simulation to make it more tangible to residents, they might have had less objection to the change. If not, they at least would have likely had more support from the majority who stood to benefit to counteract it.

Spooky23 7y ago
No, they tried a build perfect, cost optimal system to maximize ROI. They ended up losing money. A solution you cannot implement isn’t good engineering. Big changes that impact tens or hundreds of thousands of people need to be staged out to minimize impact. The Big Bang approach doesn’t work for problems like this — i bet a few thousand people would lose jobs over these schedule changes.

scarejunba 7y ago

It's the same system and it's mentioned in the OP.

b_tterc_p 7y ago
I was overeager

joe_the_user 7y ago

Oh I remember this story. Back then when I dug into it, it was a real spin-versus-spin situation. Sure, when you introduce an approach that suddenly creates considerable inconvenience for some population of parents, sure they complain (no one likes having to drop off kids at ~ vaguely recalling ~ 10 am and pick at 5 pm, sheesh, it was something terrible). Some large portion of working people are fairly dependent their children having ordinary school schedules. Maybe these were relatively better off parents in the public school system - where the actually wealthy use private schools (I know Boston well enough to attest to this). So the algorithm failed to "optimize for those they could effectively screw" but hey, would it be too much to ask to just give all schools a sane schedule? Maybe pony up that extra money, give up the need to optimize everything, keep the tail of bus optimization from wagging the dog of school scheduling?

yardstick 7y ago
For elementary students apparently “Hundreds of families were facing a 9:30 to 7:15 a.m. shift.” That’s a significant change in wake up time. Yeah, I’m not keen on that. It’s a non-starter.
- shoo 7y ago
  Most substantive changes leave some fraction of stakeholders strictly worse off than before.
- Dylan16807 7y ago
  It's a change between years, and students moving to high school are already facing the exact same changes. The disparity can't be a non-starter because it already existed. Those early hours exist. Someone has to take them (without doubling the bus fleet size). And it's better for the students if they have the early hours in elementary and not in high school.
  - elcomet 7y ago
    Well yeah but when you choose a school, you choose it based on those criterias. They cannot just change the schedule like this

superpermutat0r 7y ago

The thing is, all of these algorithms should be able to do well with arbitrary constraints. If it means saving less money then so be it. Many of these systems start optimizing some simple criteria and work fairly well. But with additional constraints they fall apart. Even companies that manage their own fleets for their own needs would like different things to purely optimizing distance/duration. They might want to balance the workload across all of their drivers, separate the drivers so they drive in a particular area, undefined start or end of the driving job. In the case above, something could have definitely be done, if the algorithm was efficient enough.

mipmap04 7y ago
I’m not sure I follow. All of those additional optimizations are pretty simple problems in the OR field.
- superpermutat0r 7y ago
  Undefined start or end time with a limit on driving duration is not trivial. Well, routing school buses is already less than trivial. You have pickup & delivery. So there's a precedence constraint to orders. Child won't be kept for hours in the bus, so the bus has to make several trips to school. Limiting the time between delivery of the child and the pickup is already less than trivial. Just to know if route is feasible (all constraints satisfied) given a list of child pickups, child dropoffs and time limit between pickup and dropoff is nontrivial and can mess up the optimization. Adding school shifts to the equation, minimizing number of vehicles and a bunch of other constraints might make the optimization just too slow or too constrained for an algorithm that was working incredible without all those constraints.
  - JoeAltmaier 7y ago
    In rural Iowa (where I live and grew up) the middle-school bus and high-school bus are full the first couple of days of school, then they all start driving/getting rides to school. At the middle of the school year, maybe 12 people on a bus build for 60. That makes it complicated too. And 'optimization' solutions result in 1hour+ rides to a school just down the road, as the bus winds about the countryside scavenging the few remaining students. The longer the trip gets, the more that find another way, the emptier the bus, the longer the trip to try to fill it again. We drove our kids to school for most of their school careers, even though three busses went by our house every day.

gerbilly 7y ago

Cry me a river for the poor optimization function.