Hacker News Reader

Comment by PavlovsCat | Hacker News Reader

It would absolutely possible to train neural nets to use speeches and interviews of politicians and powerful people to suggest likely canditates for sociopathy and other fun things. The trouble is that those people would not submit to actual clinical diagnosis, and even if they did, nothing useful would come of it. Failing to even think in such directions, and instead pondering how to treat small fish even more like lab rats, that can't be the way.

View on HN

I have actually been wanting to do something very similar to what you propose for a long time. Not as an effective measure or anything, but specifically to show people how such technology CAN be used. My specific idea would be to take news stories and other data around each member of Congress, and train a system to predict how likely each member of Congress is to be cheating on their spouse.

I got this idea when everyone responded to the revelations about the NSA spying on everyone in an extremely specific way 100% of the time. Everyone focused on human analysts reading the text or listening to the audio of conversations. Read literally any article you find where the NSA, Obama, others in the government, absolutely anyone talks about the things the NSA has done and how they practices might be altered. In every single case, they talk solely about human analysts reading things directly. They never once talk about automated systems analyzing communications and inferring behavior profiles, using those inferred profiles to make decisions or target investigations, nothing like that. And that's where essentially all of the actual danger lies as far as I'm concerned.

So perhaps if a website popped up rating each Congressperson on how likely it is they are cheating on their spouse determined by a neural net (or hell even a Bayesian filter, it doesn't actually need to be accurate at all), maybe we could actually get someone to talk about the issue. Because as it is, no one ever has or is willing to as far as I have seen. I don't know if it's because the politicians aren't aware of what such systems can do, or if they've been tricked into thinking they can actually be effective (they can't) so they intentionally try to misdirect people to concentrating on "they'll stop listening in on your phone calls" or what. The only reason I haven't built the site is that I'm unsure if there would be legal implications and if I'd end up charged with slander or something.

There's a good reason for that. The NSA don't consider data to be "collected" until it is presented to an employee or processed into a human readable form[1]. This allows them to gloss over all the data they gather in the normal, dictionary definition of the word "collect", as well as ignoring any automated analysis of that data before it's presented to a human.

As far as I can tell, the NSA don't have a term to properly describe gathering and analysis of data which isn't presented to a human for analysis.

[1]: https://fas.org/irp/doddir/dod/d5240_1_r.pdf (Page 15, 2.2.1)

I'm sure it's intentional they don't have a term for that.

If there's no name for it, you can't accidentally name it.

You are correct. I watched a documentary ages ago, years before Snowden, I believe back when the whistleblower had come out and told everyone about the black boxes AT&T put on their networks for the NSA, where it was mentioned that the biggest fear the NSA had was of ever having to appear in court and defend that exact definition of 'collect'. To the best of my knowledge, that definition has still not been challenged in court one way or another.

So that's why the NSA parses their words very carefully, and likely Obama as well... but what about their critics in Congress?

Speeches are written. Interviews are scripted. Personas are cultivated. I'm not sure how it would be able to ascertain whether a politician is a sociopath any more than it would be able to determine whether Brad Pitt is after analyzing one of his films.

Your argument would have been a lot stronger this time last year. Brexit and the Philippines are notable examples of some fairly sociopathic political behaviour that fairly closely followed on-record comments. What happens next year will be the exciting part.

We don't know how neural nets can do stuff like schizophrenia from CCTV footage, merely from how someone walks, but apparently this and other things are already emerging, and with some earnest effort could provide some real results.

Mind you, I was talking about neural suggesting people for clinical diagnosis, not about neural nets diagnosing people. And to be perfectly honest, I'm not even that much interested in the idea (which strikes me as a bit Orwellian), but in the cold feet it would give a lot of people with skeletons in the closet. They don't know that it's not fool proof, they just know they don't want to be found out.

    > to suggest likely canditates
    > for sociopathy

But what use would that information be? I don't care one iota if my president is a sociopath as long as he or she implements the agenda I want them to.

If anything, sociopathy will actually help them do it.

And since you have zilch recourse if they don't, making predictions based on their past behaviour and personality is your only bet. Also, marketing budgets in the millions of all sorts of people and corporations engineered to make them seem less banshee-like speak another language -- you may not care, but most people still have the hair on the back of their back stand up when faced with the alien, barren inner landscape of a sociopath. Faking it is, uhm, kind of important to everything they do and are. Without it, they have a meltdown, at best only destroying themselves. If we're talking about the real deal and not just someone who is sometimes a bit of a jerk, that is.

If people mind stuff like baby killers, Stalin and other things, they do mind sociopaths, whether they know what the word implies or not.

wouldn't it be good to know upfront that they are likely to not do that precisely because of sociopathic tendencies?

Reply on news.ycombinator.com