News for hackers. Stuff that matters.
Re:The Cathedral of Computation by ForHackernews on Thursday January 15, 2015 @04:14PM (#8895393)
I think you're understating the degree to which modern machine learning algorithms are completely opaque, yes, to the everyday person, but also to the developers and researchers that use them.

Check out this section from Caltech's machine learning lectures about neural networks: https://www.youtube.com/watch?v=Ih5Mr93E-2c&hd=1#t=63m25s

"...sorry, we denied credit because lambda is less than 0.5. The fact that you are not able to interpret what happens in machine learning is very, very common."

Re:The Cathedral of Computation by tjradcliffe on Thursday January 15, 2015 @04:27PM (#8895481)
This opacity is a huge deficit in machine learning. A seriously vast amount of work has been done attempting to address it, and machine learning algorithms that are persistently uninterpretable (neural nets being the poster-child for this problem) will always have very limited use.

We understand in principle what any given neural net is doing, but our lack of understanding of the specifics makes it impossible to trust them for anything interesting. We can't tell when they might fail.

For this reason a lot of people are busy solving problems that "machine learning" might one day handle, but we are doing it with relatively conventional algorithms that fall outside the "machine learning" envelope because the areas we work in (computer assisted surgery, say) aren't amenable to the application of things we don't understand.

To go back to his comparison with god: the incomprehensibility of gods is generally considered a feature by religious people. The incomprehensibility of machine learning algorithms is unquestionably a bug to those of us who have hard problems that need to be solved reliably.

Most of the algorithms you use in everyday life are not machine learning algorithms (and certainly not neural nets) for precisely this reason.

Re:The Cathedral of Computation by ForHackernews on Thursday January 15, 2015 @05:38PM (#8895881)
It's good to point out that researchers are working to reduce this opacity, but this comment contains several unsupported assertions.

> the incomprehensibility of gods is generally considered a feature by religious people.

I'm not sure how you reach that conclusion. It's hard to generalize about "religious people" or even "religions" because they are so varied. I might point to the book of Job in the Christian Bible as evidence that the inscrutable nature of God (or the universe, if you prefer) is considered frustrating and tragic enough that scripture takes pains to attempt to explain it. In fact, there's an entire branch of theology [0] that tries to address this shortcoming.

> Most of the algorithms you use in everyday life are not machine learning algorithms

Is that true? My understanding is that almost all modern web services with vast data (Facebook, Google, Netflix) make at least some use of machine learning algorithms.

[0] https://en.wikipedia.org/wiki/Theodicy

Re:The Cathedral of Computation by pixl97 on Thursday January 15, 2015 @04:37PM (#8895544)
> but our lack of understanding of the specifics makes it impossible to trust them for anything interesting. We can't tell when they might fail.

Sounds like we're making 'people 2.0'.

Re:The Cathedral of Computation by qu4z-2 on Thursday January 15, 2015 @08:12PM (#8896668)
"Predictable, or agrees with a human implementer/trainer. Pick one." ?