Slashdot Mirror


The AI Boss That Deploys Hong Kong's Subway Engineers

Taco Cowboy writes The subway system in Hong Kong has one of the best uptimes: 99.9%, which beats London's tube or NYC's sub hands down. In an average week as many as 10,000 people would be carrying out 2,600 engineering works across the system — from grinding down rough rails to replacing tracks to checking for damages. While human workers might be the ones carrying out the work, the one deciding which task is to be worked on, however, isn't a human being at all. Each and every engineering task to be worked on and the scheduling of all those tasks is being handled by an algorithm. Andy Chun of Hong Kong's City University, who designed the AI system, says, "Before AI, they would have a planning session with experts from five or six different areas. It was pretty chaotic. Now they just reveal the plan on a huge screen." Chun's AI program works with a simulated model of the entire system to find the best schedule for necessary engineering works. From its omniscient view it can see chances to combine work and share resources that no human could. However, in order to provide an added layer of security, the schedule generated by the AI is still subject to human approval — Urgent, unexpected repairs can be added manually, and the system would reschedule less important tasks. It also checks the maintenance it plans for compliance with local regulations. Chun's team encoded into machine readable language 200 rules that the engineers must follow when working at night, such as keeping noise below a certain level in residential areas. The main difference between normal software and Hong Kong's AI is that it contains human knowledge that takes years to acquire through experience, says Chun. "We asked the experts what they consider when making a decision, then formulated that into rules – we basically extracted expertise from different areas about engineering works," he says.

25 of 162 comments (clear)

  1. So... by benjfowler · · Score: 3, Insightful

    In other words, this is basically Drools, plus a ton of billable consulting hours?

    1. Re:So... by peragrin · · Score: 2

      Even better. We are starting to replace PHB's who know next to nothing bit are still in charge.

      I can't wait until we can replace lawyers with an abacus and a cukoo-clock.

      --
      i thought once I was found, but it was only a dream.
    2. Re:So... by Shoten · · Score: 5, Insightful

      In other words, this is basically Drools, plus a ton of billable consulting hours?

      Actually, it's probably something more like TIBCO BusinessEvents with an orchestration engine added. But what's really cool is that they did the hard part: codifying the actual rules under which the overall system operates. That's where these kinds of systems either fly or fall. There are tons of rules that organizations use to make decisions, but a lot of those rules are quite informal and don't operate at a central point of authority. It takes a lot of digging to find them all, so that the undocumented process (for example) used by the foreman of the team that does rail maintenance to manage overtime among his crew gets incorporated into the overall chaining logic. Otherwise, the new system will either fail to reflect reality as teams rearrange their own schedules out of sync with their directives, or will wreak havoc among the employees.

      --

      For your security, this post has been encrypted with ROT-13, twice.
    3. Re:So... by K.+S.+Kyosuke · · Score: 5, Insightful

      Actually, it's probably something more like TIBCO BusinessEvents with an orchestration engine added

      Back in my time, we called what they have done now "an expert system". I fail to see why that designation should be suddenly inadequate.

      --
      Ezekiel 23:20
    4. Re:So... by Shoten · · Score: 2

      Actually, it's probably something more like TIBCO BusinessEvents with an orchestration engine added

      Back in my time, we called what they have done now "an expert system". I fail to see why that designation should be suddenly inadequate.

      Because back then, that was a conceptual description that (if it became real) described an entirely custom system that was built from the ground up. These days, there are multiple types of such systems, most of which are built along specific architectural lines using COTS. Just like once upon a time, "car" was a pretty good descriptor because the next level of detail went WAY into the weeds. Now, there are sports cars, SUVs, minivans, coupes, etc.

      --

      For your security, this post has been encrypted with ROT-13, twice.
    5. Re:So... by Charliemopps · · Score: 2

      Does that look at traffic as well?

      Some times it can be better to let the tech see the full job list and plan there own day based on local info.

      It still needs to get fixed either way, but yes the tech can skip a job. There are stats involved so if you skip lots of them you're going to have to explain why. The rulesets are somewhat complicated, it doesn't just pick the closest one... How old is the ticket? Is it a major piece of equipment? Etc... They can even merge the tickets on the fly... are these 10 sites out of service because of that other ticket down the line? etc...

      This isn't a small system. It manages hundreds of techs all over the country and there's a backend of engineers, ticket jockies, records keepers, mapping engineers, facility techs, etc... I'm pretty sure all companies that maintain large networks of equipment operate this way now. I can guarantee you UPS and Fedex have similar systems, as well as the power companies, railroads, etc... I even saw a heavy equipment manufacturer that used such a system to dispatch their service techs to do repairs on customers bulldozers and such.

      This is how dispatching works now, that's why this stories funny. It would be more of a story if they were doing something different.

    6. Re:So... by Charliemopps · · Score: 3, Interesting

      But still you can have 2 tickets and due to traffic you can get to one site right away and the other will take 1 hour to get to but after going to one site traffic frees up and you can get to that 1 hour site in to 10-20 min.

      that's of minor concern. You need to understand how this stuff was done prior to these systems.

      15yrs ago, the tech would drive into work, pick up a stack of work orders off a printer, get in his van and head out in whichever route he thought best. You'd not see the guy for 4hrs, then he'd head back in, grab another stack and so on.

      Now he doesn't even come in. He just looks at his phone and heads to the specified location. The phone even links the location to google maps. Now he can't get lost. The app tells the office where he's at, not that they care... but if he's been on the same ticket for 6hrs they can see he's at the house or if he stopped by a local lake to do some fishing which is something that actually happened quite frequently prior to dispatching software. I used to have to schedule dispatches with other companies we did buisness with and it was common back in the 90s for the tech not to show up on time so, if they were on main street or near the center of town, I'd have the customer stick their head out the front door and look for the techs van. They'd often be at a local don-nut shop or bar waiting for the appointment and get too busy chatting. The customer would have to run down and get them. It was insanely frustrating. Now the dispatcher can update the ticket and make the techs phone chirp.

      Traffic delays aren't that big of a deal in comparison.

  2. Good by kruach+aum · · Score: 4, Insightful

    Everything currently run by committee should ideally be run by an AI with limited human oversight in the future. Groups of humans suck at the two things AIs are great at: remembering things and making decisions.

    1. Re: Good by kruach+aum · · Score: 4, Funny

      That sounds suspiciously like a decision to me, comrade. Do you have documentation signifying that you are 100% pure AI?

  3. I can't imagine something like that in the U.S.A. by ArcadeMan · · Score: 3, Insightful

    Laws, paperwork, unions, paperwork, regulations and paperwork wouldn't allow this to happen.

  4. Re:I can't imagine something like that in the U.S. by oodaloop · · Score: 4, Insightful

    I dunno. If a corporation smells a profit in it, then I think they'll find a way.

    --
    Tic-Tac-Toe, Global Thermonuclear War, and relationships all have the same winning move.
  5. Expert System by DontBlameCanada · · Score: 5, Insightful

    This is a perfect example of an Expert System.

    Expert Systems have been one of the most successful and longest used AI models in industry. FPGA routing and layout programs have relied on this form of AI since the early/mid 90's.

    1. Re:Expert System by CastrTroy · · Score: 3, Insightful

      I don't even know if I'd classify something like this as "AI". It's just running an algorithm using lots of information and doing complex calculations. Way more complex than any person could do, but they are not the kind of actions I would generally consider "intelligent". Efficient allocation of resources works great for computers, because they aren't biased. They don't give their friends extra shifts or wait until later to call in a repair crew because the didn't like the attitude of the person who reported the problem.

      --

      Anthropic principle: We see the universe the way it is because if it were different we would not be here to see it.
    2. Re:Expert System by K.+S.+Kyosuke · · Score: 3, Interesting

      I don't even know if I'd classify something like this as "AI". It's just running an algorithm using lots of information and doing complex calculations. Way more complex than any person could do, but they are not the kind of actions I would generally consider "intelligent".

      It is a curse for the field of AI that it's generally defined quite vaguely as "trying to do things in computers that humans do better than computers". Why, so many things people consider "not AI" today are commonplace precisely because AI researchers dedicated a lot of their time to solving them! Once you solve an AI problem, in minds of many people, it ceases to be not only a problem, but also a matter for AI. The area in question is known as "automated planning and scheduling", and the reason why you don't think it should be classified as AI is to a large extent because it was the AI researchers who largely solved it back in the 1970s or so.

      --
      Ezekiel 23:20
    3. Re:Expert System by blackiner · · Score: 3, Insightful

      Reminds me of how my AI professor described AI. You have two types of AI, strong and weak, strong being something akin to a conscious thinking mind (and not even guaranteed to be possible at the moment), and weak being stuff like data mining, translation, speech-to-text, puzzle solvers, etc. She also let us know that things are only considered AI until they are solved, then they are just 'algorithms', which I think mirrors people's perceptions of AI quite nicely.

  6. Management is becoming obsolete by Anonymous Coward · · Score: 4, Interesting

    What we see now are the first steps towards making a big chunk of management obsolete. Expert systems are well on their way to out-compete managers who in many situations cannot make decisions of the same quality as an AI. Or to put it differently: An AI can make better decisions than a human in many areas. And in these areas humans (managers) will not be able to compete.

  7. The Computer is Your Friend by jovius · · Score: 2

    "People get scared when you talk to them about AI,"

    Team Leader, please report to the debriefing room ASAP.

  8. Re:I can't imagine something like that in the U.S. by i+kan+reed · · Score: 5, Insightful

    Riiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiight. It's those unions. Those ones whose membership has been steadily and measurably been decreasing for 30 years(almost exactly at the same rate as wage stagnation occurs, as a complete coincidence).

    How small does Snowball's organization have to get before you stop believing he's behind everything?

  9. It's here already? by Ken_g6 · · Score: 3, Insightful

    Is it called Manna?

    --
    (T>t && O(n)--) == sqrt(666)
    1. Re:It's here already? by LordLimecat · · Score: 2

      Its from the story he linked. It posits a future where AI runs the American economy, leading to decidedly mediocre standards for everyone. The alternative posited is "the Australia project' which is basically continent-wide communism due to robot labor, renewable energy, and no pollution.

      Its all great sounding stuff, but its also complete nonsense.

  10. Interesting...but not 'new' by dtjohnson · · Score: 3, Informative

    The article was posted by someone who does not appear to have been around computers in industrial applications. Computers have been used for at least 4 decades for maintenance planning in large facilities as well as other areas such as transportation routing, product blending, production scheduling, etc. The maintenance activities for the London tube or the NYC subway are likely also being planned and scheduled using some sort of similar system even if the uptime result is not as good as Hong Kong.

  11. Now is the time fire the experts. by 140Mandak262Jamuna · · Score: 4, Insightful
    We have talked to the experts. Extracted their wisdom. Encoded it into machine readable rules. Proved that all the expertise has been extracted by the 99.9% up time.

    So, naturally, the next step is to fire all those people who would no longer have something to contribute. As a purely added bonus all these people fresh out of things to contribute happen to be with years and years of experience, which means seniority and high pay.

    The mid level bean counter would think, "well, I should be able to fire at least 20 of them. Savings of 2 million on pay, another million in benefits, almost 10 mill over three years. Even if I have to let the SOBs CEO and CFO grab a mill each, I should be able to get at least 250 K for myself. Time to fire up power point, 'Work Force Optimization due to the increased Efficiency achieved by the AI system. By Gottah Avemyb Onus, Sr Vice President, Hatchet Division'"

    --
    sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
  12. Re:I can't imagine something like that in the U.S. by CRCulver · · Score: 2

    Not to mention that high-performing metro systems worldwide are highly unionized.

  13. Re:I can't imagine something like that in the U.S. by PPH · · Score: 2

    Of course not for somthing as critical as maintenance planning and scheduling. But its OK for air traffic control functions. Where the consequences of a bad rule set are not nearly as serious.

    --
    Have gnu, will travel.
  14. Maybe not a conventional expert system? by kurisuto · · Score: 2

    The article says that they're using a genetic algorithm. I'm no expert at AI, but my understanding is that an ordinary expert system doesn't use a genetic algorithm; an expert system just involves percolating propositions through a bunch of human-specified if/then statements.

    I'd hazard a guess that the system described here is using the human-specified rules as part of the fitness function for the genetic algorithm. That's one way a system could use human-specified rules, but I think it's different from how an ordinary expert system uses them.

    If you can call this an "expert system", then at a minimum, it looks like it's pushing the boundaries of the definition of "expert system".