February 28, 2017 at 8:37 am
This topic interests me a good deal, but sometimes I've no idea at all how it can be applied. At least to some software I'm working on. First example, I'm currently working on a small project where we are automating a system used for recording grievances. (Someone has a problem with some other employee or supervisor, which they've not been able to resolve. They've taken it to management, etc.) I've long considered things like machine learning and AI to involve data. In particular, numeric data. With grievances, with the exception of the dates of meetings, there is no numeric data anyway. Instead its things like (and I'm making this up, to illustrate what's in the data), "He blew a paper straw at me!", "Oh yeah, well she blew a spit wad at me!" And this only shows up once. And a lot of it is that way. Some grievance which isn't repeated again. How in heck do you apply machine learning/AI to that?
Kindest Regards, Rod Connect with me on LinkedIn.
February 28, 2017 at 9:33 am
Rod at work - Tuesday, February 28, 2017 8:37 AMThis topic interests me a good deal, but sometimes I've no idea at all how it can be applied. At least to some software I'm working on. First example, I'm currently working on a small project where we are automating a system used for recording grievances. (Someone has a problem with some other employee or supervisor, which they've not been able to resolve. They've taken it to management, etc.) I've long considered things like machine learning and AI to involve data. In particular, numeric data. With grievances, with the exception of the dates of meetings, there is no numeric data anyway. Instead its things like (and I'm making this up, to illustrate what's in the data), "He blew a paper straw at me!", "Oh yeah, well she blew a spit wad at me!" And this only shows up once. And a lot of it is that way. Some grievance which isn't repeated again. How in heck do you apply machine learning/AI to that?
You can't. Machine learning only works with lots of data, which could be a lot now, or a lot across time with similar conditions. Machine learning is better at recognizing, and remembering past occurrences, predicting when they might reoccur or appear.
If your problem doesn't lend itself to looking at what might be the same or different, ML isn't a good fit.
February 28, 2017 at 9:34 am
wbc2 - Tuesday, February 28, 2017 7:49 AM"Computers make excellent and efficient servants ... but I have no desire to serve under them." Mr. Spock, The Ultimate Computer 🙂
They still are with ML systems. Just as they are with current reporting systems.
February 28, 2017 at 9:36 am
lnoland - Monday, February 27, 2017 5:15 PMIsn't that pretty much what I said?On the other hand, while people make plenty of mistakes, if a person causes multiple accidents due to persistent bad judgment we take his license away and he may face some lawsuits. If a self-driving car is given bad judgment which causes one or more accidents, that bad judgment could be replicated thousands of times before it is stopped. And that presumes that it is stopped -- look at the Audi 5000 fiasco. Audi pretty much began by blaming everything on the drivers (an incompetent group apparently largely unique to Audi 5000 purchasers); then their actions suggested that Audi believed that their physical design layout might be contributing to driver error so they did a recall to make some changes to pedal positions; NHTSA suggested that it went beyond that to an actual failure which was then exacerbated by the driver's response. To my knowledge, Audi never admitted to a problem and it was only the many lawsuits which convinced them to do anything at all.
Perhaps, but that wasn't what I took from it. Certainly computers allow us to make mistakes quicker, but also correct them quicker.
As more complex systems become more prevalent, we certainly need new legal constructs to deal with them and decide what/how we deal with corrections. In my mind, all companies running these systems need ways to fall back quickly until patches can be applied, and deal with the appropriate liabilities.
February 28, 2017 at 9:52 am
Steve Jones - SSC Editor - Tuesday, February 28, 2017 9:34 AMwbc2 - Tuesday, February 28, 2017 7:49 AM"Computers make excellent and efficient servants ... but I have no desire to serve under them." Mr. Spock, The Ultimate Computer 🙂They still are with ML systems. Just as they are with current reporting systems.
Agreed. Like everyone else here, I have worked with computers a long time and will continue to. I just have no desire to have them "over" me or myself subservient to them. 🙂 In the same vein, I do not, and will not have any interest in putting my car into the hands of a computer to drive.
February 28, 2017 at 10:15 am
Rod at work - Tuesday, February 28, 2017 8:37 AMThis topic interests me a good deal, but sometimes I've no idea at all how it can be applied. At least to some software I'm working on. First example, I'm currently working on a small project where we are automating a system used for recording grievances. (Someone has a problem with some other employee or supervisor, which they've not been able to resolve. They've taken it to management, etc.) I've long considered things like machine learning and AI to involve data. In particular, numeric data. With grievances, with the exception of the dates of meetings, there is no numeric data anyway. Instead its things like (and I'm making this up, to illustrate what's in the data), "He blew a paper straw at me!", "Oh yeah, well she blew a spit wad at me!" And this only shows up once. And a lot of it is that way. Some grievance which isn't repeated again. How in heck do you apply machine learning/AI to that?
I doubt that machine learning would be very useful in such a case, however, the area of AI (or knowledge engineering) where we spent most of our time in the class I took was in Expert Systems, which is basically a method of using a computer to reproduce human skills using logic-based rules rather than algorithmic programming. There are things people do which would be very difficult to reproduce with a typical programming language but which are "easily" taught to others. Expert systems attempt to capture the reasoning which the "domain expert" goes through in order to accomplish a task and create rules for a computer to follow to do the same. In the past it would have been too expensive and involved to apply these to systems like yours but Microsoft has been putting some of the tools for doing so into their other products, most notably, Windows Workflow Foundation. You could probably find applications for those tools in your work. Just a simple, for instance, consider if handling of a grievance is a multistep job where the steps are often handled in different orders, it might be a lot easier to create a set of rules to determine what steps have been handled, which still need to be done and which have dependencies on other steps (so they can't be done until the other steps are). While this could be done algorithmically, it is typically a lot clumsier and messier to do. Of course, like most things in this industry, the learning curve is what so often keeps us from using the right tool for the right job.
March 1, 2017 at 8:19 am
Steve Jones - SSC Editor - Tuesday, February 28, 2017 9:33 AMRod at work - Tuesday, February 28, 2017 8:37 AMThis topic interests me a good deal, but sometimes I've no idea at all how it can be applied. At least to some software I'm working on. First example, I'm currently working on a small project where we are automating a system used for recording grievances. (Someone has a problem with some other employee or supervisor, which they've not been able to resolve. They've taken it to management, etc.) I've long considered things like machine learning and AI to involve data. In particular, numeric data. With grievances, with the exception of the dates of meetings, there is no numeric data anyway. Instead its things like (and I'm making this up, to illustrate what's in the data), "He blew a paper straw at me!", "Oh yeah, well she blew a spit wad at me!" And this only shows up once. And a lot of it is that way. Some grievance which isn't repeated again. How in heck do you apply machine learning/AI to that?You can't. Machine learning only works with lots of data, which could be a lot now, or a lot across time with similar conditions. Machine learning is better at recognizing, and remembering past occurrences, predicting when they might reoccur or appear.
If your problem doesn't lend itself to looking at what might be the same or different, ML isn't a good fit.
Ah. OK Steve, thanks.
Kindest Regards, Rod Connect with me on LinkedIn.
March 1, 2017 at 8:36 am
lnoland - Tuesday, February 28, 2017 10:15 AMRod at work - Tuesday, February 28, 2017 8:37 AMThis topic interests me a good deal, but sometimes I've no idea at all how it can be applied. At least to some software I'm working on. First example, I'm currently working on a small project where we are automating a system used for recording grievances. (Someone has a problem with some other employee or supervisor, which they've not been able to resolve. They've taken it to management, etc.) I've long considered things like machine learning and AI to involve data. In particular, numeric data. With grievances, with the exception of the dates of meetings, there is no numeric data anyway. Instead its things like (and I'm making this up, to illustrate what's in the data), "He blew a paper straw at me!", "Oh yeah, well she blew a spit wad at me!" And this only shows up once. And a lot of it is that way. Some grievance which isn't repeated again. How in heck do you apply machine learning/AI to that?I doubt that machine learning would be very useful in such a case, however, the area of AI (or knowledge engineering) where we spent most of our time in the class I took was in Expert Systems, which is basically a method of using a computer to reproduce human skills using logic-based rules rather than algorithmic programming. There are things people do which would be very difficult to reproduce with a typical programming language but which are "easily" taught to others. Expert systems attempt to capture the reasoning which the "domain expert" goes through in order to accomplish a task and create rules for a computer to follow to do the same. In the past it would have been too expensive and involved to apply these to systems like yours but Microsoft has been putting some of the tools for doing so into their other products, most notably, Windows Workflow Foundation. You could probably find applications for those tools in your work. Just a simple, for instance, consider if handling of a grievance is a multistep job where the steps are often handled in different orders, it might be a lot easier to create a set of rules to determine what steps have been handled, which still need to be done and which have dependencies on other steps (so they can't be done until the other steps are). While this could be done algorithmically, it is typically a lot clumsier and messier to do. Of course, like most things in this industry, the learning curve is what so often keeps us from using the right tool for the right job.
Thank you. Very interesting. This problem space I'm referring to may lend itself to an Expert System approach, eventually. I'm still pretty new to this grievance situation here at work. I've only been on the project for a little less than 2 months and we're getting close to releasing it. (Like I said it is a small project.) I understand that the HR department used MS Excel spreadsheets to keep this data and that they have been doing this for quite some time, but God only knows what sort of format those are in. Most likely not in any coherent nor consistent format. Bottom line I doubt that they'll try to import the data from, well who knows how many different spreadsheets, into the new database we've made. But in time we'll probably begin to have a large enough amount of data where we could use an Expert System on this. That way we could help some future HR person by making suggestions, based upon the data we've collected, as to what step could be taken next in a grievance process towards a successful resolution.
Kindest Regards, Rod Connect with me on LinkedIn.
March 2, 2017 at 2:10 pm
I am optimistic about self drive cars. The bit that excites me is that a whole chain of vehicles could be talking to each other so when the front one brakes every vehicle behind it knows that it has to brake as well, it isn't waiting for a visual clue. Neither are they limited to a small range of physical senses. They can have more senses than the human brain could possibly process. They won't get tired,won't have an off-day where they drive like a drain.
I think of machine learning and AI in the same way that I think of automated testing. We have a mechanical something that can consider vastly more scenarios than a human mind could comprehend and apply the rules with absolute consistency. It can also adapt where gaps are discovered.
I could be pithy and suggest that a lot of the excitement about artificial intelligence is down to unmet expectations with the real thing. Some of the machine learning stuff has been around for decades and operating in a production environment albeit in a niche field. We are now seeing the consumerization of it and it is beginning the early passage through the Gartner Hype Cycle. I suspect that some of the experienced practitioners are holding their head in their hands cursing the parents of the marketeers who have got hold of their pride and joy.
Viewing 9 posts - 16 through 23 (of 23 total)
You must be logged in to reply to this topic. Login to reply