Hacker News Reader

Comment by unexistance | Hacker News Reader

I can only wonder what DARPA are doing currently behind all the secrecy, assuming whatever disclosed in the book is, let say, 20 years back technology.

p/s: there doesn't seems to be a lot of red-tape in approving whatever the 'researchers' wants to do, at least in the past

View on HN

You can actually see what DARPA wants to do by looking at their business solicitations. The US government makes a giant wishlist available as a website for businesses to browse and submit proposals for. The last few years, the big ticket item for most organizations is some kind of artificial intelligence that can intelligently process mountains of data. DARPA wants one, too. But they're also into some kinkier stuff.

As one related example, DARPA is interested in "explainable artificial intelligence". Essentially machine learning where the decision making process and intermediary steps are less opaque to humans.

http://www.darpa.mil/program/explainable-artificial-intellig...

What's funny is Google's actually doing a fair bit of this already with the TensorFlow playground, flow-based programming, Distill, etc. DoD will shower money on researchers, but those same researchers will do the work for Google for stock, a decent salary, and access to interesting problems that don't involve killing people.

That's not what "explainable ML is".

That project is about explaining how (say) a given classifier came to a decision on a given set of data. This is fairly easy on a tree-based classifier, but on a deep neural network is extremely difficult.

The most impressive piece of work I'm aware of in this area is LIME, from “Why Should I Trust You?” Explaining the Predictions of Any Classifier[1]

The acknowledgement section of that paper read, in part:

This work was supported in part by ONR awards #W911NF-13-1-0246 and #N00014-13-1-0023, and in part by TerraSwarm, one of six centers of STARnet, a Semiconductor Research Corporation program sponsored by MARCO and DARPA.

Today I saw ELI5[2], which encapsulates LIME and a number of other similar tools.

[1] https://arxiv.org/pdf/1602.04938.pdf

[2] http://eli5.readthedocs.io/en/latest/

> but those same researchers will do the work for Google for stock, a decent salary, and access to interesting problems that don't involve killing people.

Regardless of origin you can rest assured the technology will make its way into the hands of a killer. It always does.

But making intelligent killing machines is half the fun of heralding into a Terminator inspired future distopia.

This is a 404 page now.

Hmm, it still loads for me.

I hope you don't think they don't have a wish-list that you can't see because I assure you, they have one.

They're not gonna publish an RFP for something that's actually groundbreaking. That's gonna be classified.

But that's not what I said, is it?

RFP: quantum algorithms, lots of them ;)

They were asking after therapeutic nanoparticles 6 years ago, so by now there's probably some foreign head of state with a full Fantastic Voyage crew sailing around his nervous system, waiting for the signal.

why foreign? you lack imagination, my dear friend.

Por que no los dos?!?

Fairly interesting genetic live-editing program: http://www.darpa.mil/attachments/IA%20ProposersDay.pdf

This program has civilian uses, but can be weaponized easily.

Link seems to be dead. Is there another copy somewhere?

They missed a %20 from the link http://www.darpa.mil/attachments/IA%20Proposers%20Day.pdf

Reply on news.ycombinator.com