May 22, 2017 at 9:31 am
I think a lot of people are confusing automation that has clearly defined logical steps as AI or machine learning when they are clearly nothing more than automated scripts with rules. I've talked with many data scientist who have struggle when explaining how things like machine learning and so forth really work compared to what people think they work.
Ideally, I think one of the best ways I've heard it explained is that predefined rules by humans are actually skewing / influencing the results. True AI is going to let the data or variables tell them (actually think and learn) what to do and learn from that doing in a sense of either making the same choice again or making a different choice based on the prior results.
For example, in marketing and it's common to for someone to say if this data shows X, then it's good and you should do Y. The problem is even though you can automate something to look for what is good and have a action for it, they can't wrap their head around if what they think is good is actually bad based on what the data is telling them. This is where the machine learning and so forth really comes in. Not letting something influence it to make the wrong decision as opposed to letting the data tell you what decision to really make without the outside influence from a human (i.e.: the programmer saying turn left if this happens, turn right if this happens).
So yeah, a lot of this is becoming buzzwordy when in reality, it's just automation with logic defined by humans who are basically skewing the decisions based on what they think is right or wrong; not the machine.
May 22, 2017 at 10:56 am
Eric M Russell - Monday, May 22, 2017 8:35 AMAutomated customer support is an example of one application where AI really shines. No?
:rolleyes:
Not really true. It does do well here, depending on how people have implemented. There are some good AI systems beyond If..Then, which can determine some intent and context to help customers solve issues. These aren't the bank systems where you call and get a balance, pay a bill, etc. Those are programmed with defined paths. There are AI components that can follow paths and provide information they weren't programmed to deal with. The systems learn what types of things can be asked and then infer possibilities, like a human would.
This is still fairly basic and works well within specific domains, but it definitely reduces labor needs.
May 22, 2017 at 10:57 am
xsevensinzx - Monday, May 22, 2017 9:31 AMI think a lot of people are confusing automation that has clearly defined logical steps as AI or machine learning when they are clearly nothing more than automated scripts with rules. I've talked with many data scientist who have struggle when explaining how things like machine learning and so forth really work compared to what people think they work.Ideally, I think one of the best ways I've heard it explained is that predefined rules by humans are actually skewing / influencing the results. True AI is going to let the data or variables tell them (actually think and learn) what to do and learn from that doing in a sense of either making the same choice again or making a different choice based on the prior results.
Yes, very much how I see AI. The computer learns and does things outside of its programming. There aren't predefined automation and programming flow paths.
May 22, 2017 at 11:04 am
Jeff Moden - Sunday, May 21, 2017 11:44 PM...
Systems can learn and grow based on their inputs and measured outcomes... if they are programmed to do so.
OK, let's take the first sentence you have. The programming. AI and machine learning systems use models and data to learn how to respond. The model programs itself. The programmer hasn't decided ahead of time how the paths work. I saw a great explanation that was something like:
data ---> program --> output
Machine learning and AI work like this:
program --> data --> output
In your spreadsheet example, you'd input some spreadsheets, some outputs, identify what a column and header is, and the application would then learn to pick up all sorts of spreadsheets and input the data. Have a chart, it learns to ignore charts, or find the data around the chart. A spreadsheet isnt' the greatest example, however, since clever programming can find the data as well.
Image and video (I guess audio) is a way more interesting application, since it's hard to program all the possibilities there. Certainly some of those problems benefit from AI, and I'd see those as the fastest advancing areas.
In terms of indexing, programming a routine to look at indexes and workload and try to work out the best system could be done, but I suspect it wouldn't work well. The code and paths are too complex. It's a difficult problem to solve in the general sense because of the wide range of possibilities and too many variables. However, an ML system would look to learn from things that work well and things that don't, without being programmed to look for 5 number of indexes or a certain number of included columns, or anything else. It could learn to make changes, measure impact, try new changes, but not in a rote method, but in a more intelligent, educated guessing method. Like humans do. Without writing all the code necessary in a plain automated system.
May 22, 2017 at 12:09 pm
Steve Jones - SSC Editor - Monday, May 22, 2017 10:57 AMxsevensinzx - Monday, May 22, 2017 9:31 AMI think a lot of people are confusing automation that has clearly defined logical steps as AI or machine learning when they are clearly nothing more than automated scripts with rules. I've talked with many data scientist who have struggle when explaining how things like machine learning and so forth really work compared to what people think they work.Ideally, I think one of the best ways I've heard it explained is that predefined rules by humans are actually skewing / influencing the results. True AI is going to let the data or variables tell them (actually think and learn) what to do and learn from that doing in a sense of either making the same choice again or making a different choice based on the prior results.
Yes, very much how I see AI. The computer learns and does things outside of its programming. There aren't predefined automation and programming flow paths.
Me too. That's why I don't believe that something that will auto-magically create indexes is actually AI.
--Jeff Moden
Change is inevitable... Change for the better is not.
Viewing 5 posts - 31 through 34 (of 34 total)
You must be logged in to reply to this topic. Login to reply