News for hackers. Stuff that matters.
Re:AI Playbook by mifeng on Monday May 15, 2017 @05:59PM (#14345515)
While this site has some great info for non-technical folk who are serious about learning AI, I think the site will fly over the heads of most "line-of-business leaders" at whom it's aimed.

In my experience, those folks think of AI like...

My company is collecting tons of data in [insert data warehouses here]. How can I bolt on some "AI" and find out what's "interesting"?

Re:AI Playbook by _delirium on Monday May 15, 2017 @06:16PM (#14345643)
> My company is collecting tons of data in [insert data warehouses here]. How can I bolt on some "AI" and find out what's "interesting"?

For quite a few business types, this can be further simplified to just:

How can I bolt on some "AI"?

It's a nice bonus if you have some data and can find interesting things from it. But the goal for a surprising number of managers is just to be able to say they have some advanced AI technology doing, well, who knows what, just so long as there is definitely some advanced AI technology doing it.

Re:AI Playbook by riphay on Monday May 15, 2017 @06:09PM (#14345587)
This is a great point, and I think I fall into that bucket you describe. I have just enough information to be dangerous.

There does seem to be a lack of available tools even for someone with a well-defined problem with available labelled data to validate implementing machine learning to a problem. I think that there's a need out there for a software with a GUI that can accept data and let you try some ML approaches with it and take the coding out of it - and it doesn't seem like it would be insanely difficult to build.

Re:AI Playbook by mifeng on Tuesday May 16, 2017 @09:07PM (#14354791)
Check out outlier.ai. Also I think there was a company in the most recent YC batch that tackled this problem.
Re:AI Playbook by fudged71 on Monday May 15, 2017 @07:41PM (#14346197)
This is spot-on. There's an assumption that if you have large/big data, then AI will be your next big step. That's absolutely not a given. I've been told that AI is just "Statistics at scale", but you wouldn't call simple decision making like A/B testing "AI".
Re:AI Playbook by mattkrause on Monday May 15, 2017 @09:16PM (#14346673)
There's a lot of room for sophistication in an A/B test. You might want to do something like Thompson sampling to minimize the number of samples taken from the 'bad' alternatives, or a contextual bandit to take some demographic information into account.
Re:AI Playbook by quocble on Monday May 15, 2017 @10:30PM (#14346956)
Agreed. It wasn't very insightful in terms of novel methods.