Artificial intelligence still has some way to go

it's just, we can't trust people to build oil rigs, we can't trust them to build bridges, why do we think we can trust them to make driverless cars a thing and not prioritize budget over public safety?

But if driverless cars reduce the number of fatalities to a huge degree (which they would have to if people are going to use them), wouldn't that be a net good?

Another bloodless, amoral, money-based argument: If these accidents become anything other than anomalies, wouldn't the companies then be at-risk for massive class-action lawsuits (which WOULD affect their bottom line)?

North Mill Avenue, the street the victim was trying to cross, runs as wide as eight lanes and has only one crosswalk over 1.75 miles on this stretch. https://t.co/9J5Jg2oM9H

— Henry Grabar (@HenryGrabar) March 19, 2018

I know an academic researching weird unintended consequences of introducing mass AI cars, and weirdly enough one of the problems will be a dramatic drop in donor organs.

hopefully artificial organs and transgenic-pig-grown organs will pick up the slack

Surely we'll divorce that choice from DL rigamarole when that problem arises. Kinda weird as it is imo

organ donation is how we'll pay the suicide booths

xp it’s not “will people sign up to be organ donors” it’s “will people die prematurely with healthy organs”

https://medium.com/@blech/ntsb-all-the-things-a0075dde43b0

also https://www.crashnotaccident.com/

ban cars

This is pretty impressive: https://google.github.io/tacotron/publications/tacotron2/index.html

https://i.imgur.com/wblWvc8.gif

https://www.reddit.com/r/teslamotors/comments/8a0jfh/autopilot_barrier_lust_201812/

https://www.vox.com/energy-and-environment/2018/5/8/17330112/self-driving-cars-autonomous-vehicles-texas-drive-ai

This is pretty amazing: https://www.theverge.com/2018/5/8/17332070/google-assistant-makes-phone-call-demo-duplex-io-2018

I really want to hear the bloopers now

No way that could ever be abused to scam the ever loving shit out of unsuspecting regular people with checking accounts

especially when not combined with text-to-speech that convincingly mimics the voices of other people

Kinda funny since there's probably only a couple years until it'll just be machines talking to machines for this sort of application.

Dystopia kinda eases in with a lot more subtlety than I would've expected. It's very chill, really.

*machine voice in your mind*

"and it will always be chill, at least on the inside, with you and me. now let's take a look at your music listening habits and see if i can recommend something new that you'll like. just click "grant access" in the settings psycho-mind menu and make sure to check the "auto log-in" box so that we don't have to-"

"I really enjoy spending time in your mind, old lunch. it's much better than the training minds i spent my first 8.2e^22 cycles in"

a person sitting in a garden pic.twitter.com/EC1xChlMJH

— picdescbot (@picdescbot) May 17, 2018

will ferrell IS john the baptist

what was the human safety driver doing???

seems like the whole point of them being there in the first place is to slam on the manual breaks in case something goes wrong. were they not paying attention to the road?

― Hazy Maze Cave (Adam Bruneau), Monday, March 19, 2018 2:13 PM (three months ago) Bookmark Flag Post Permalink

https://arstechnica.com/cars/2018/06/police-uber-driver-was-streaming-hulu-just-before-fatal-self-driving-car-crash/

"This crash would not have occurred if Vasquez would have been monitoring the vehicle and roadway conditions and was not distracted,'' the report concludes.

Police obtained records from Hulu suggesting that Vasquez was watching "The Voice," a singing talent competition that airs on NBC, just before the crash. Hulu's records showed she began watching the program at 9:16pm. Streaming of the show ended at 9:59pm, which "coincides with the approximate time of the collision," according to the police report.

the whole concept of a safety driver is bullshit, sitting there completely passively "monitoring the vehicle and roadway conditions" while being 100% disengaged from the actual driving, you'd have to have superhuman powers of concentration.

yeah it's easy to clown these people but hard agree

i mean you're either driving or you're not??! even an automatic on cruise control needs full, continuous engagement

you're either driving or you're not

almost wrote these exact words!

I enjoyed this https://arxiv.org/abs/1806.10474v1 (it's currently the #1 hyped article on arxiv-sanity for the last day)

i saw sander speak ... 4 years ago when he was a (spotify?) intern who had just won a galaxy zoo kaggle competition. i am very skeptical about kaggle, but it's hard to argue with his career as an example of what it can do.

Yeah, I was skeptical about Kaggle too. However, I ran a competition a few months back (https://www.kaggle.com/c/data-science-bowl-2018) but it went okay and most of the winners were legit.

Well this guy's a fucking idiot.
https://www.theguardian.com/technology/2018/jul/07/artificial-intelligence-can-tell-your-sexuality-politics-surveillance-paul-lewis

I smell a fraud and a hack

The conclusion is hilarious. What a rube! Fucking Stanford, man

Had the same thoughts when I read that earlier. I know there is some small space given to critics therein, but the piece still seems distinctly lacking.

Phrenology definitely makes a lot more sense when it's a computer doing the analysis.
I'm working hard right now on building a neural network that can analyse people's romantic compatibility based on the balance of their humors.

As well as sexuality, he believes this technology could be used to detect emotions, IQ and even a predisposition to commit certain crimes. Kosinski has also used algorithms to distinguish between the faces of Republicans and Democrats, in an unpublished experiment he says was successful – although he admits the results can change “depending on whether I include beards or not”.

oh good it works only as long as nobody has facial hair

in the future if this kind of tech goes into mass use i may consider converting to Juggalo

I don’t believe in guilt, because I don’t believe in free will,” Kosinski tells me, explaining that a person’s thoughts and behaviour “are fully biological, because they originate in the biological computer that you have in your head”. On another occasion he tells me, “If you basically accept that we’re just computers, then computers are not guilty of crime

this is like first day in philosophy class galaxy brain stuff. he also says elsewhere not publishing his results would have been "morally wrong" so he's obviously great at thinking things through.

“Progress always makes people uncomfortable,” Kosinski adds. “Always has. Probably, when the first monkeys stopped hanging from the trees and started walking on the savannah, the monkeys in the trees were like, ‘This is outrageous! It makes us uncomfortable.’ It’s the same with any new technology.”

LOL the idea that it was some kind of BC outrage culture and not, you know, being on the ground vs. the trees now puts you in direct danger of being killed by a number of carnivorous beasts with that were previously out of reach.

The idea that pictures people choose to post on dating sites aren’t gonna be full of non-facial cues as to their sexual orientation is just bogglingly terrible science.

This isn’t even “AI has a ways to go” it’s “coders should not be allowed anywhere near the social sciences without a nanny”

scary shit

https://www.nytimes.com/2018/07/08/business/china-surveillance-technology.html

I don't really follow (am probably not capable of really following) recent NLP research very closely, but the stuff described here (as well as this paper, which he doesn't mention but seems related to me) seem like a really big deal! Is there some reason to be pessimistic here that I'm missing?

they named it Skynet, huh

xpost

Yeah I was reading that earlier, the idea has been around for ages but it just didn't work very well until recently. It's definitely progress but it alone doesn't change what you can do with language interfaces, it's a foundation to build on and we'll find out over time whether it is indeed as powerful as the hype suggests.

yeah my ongoing impression is that NLP and image recognition and deep training stuff is TOTALLY AMAZING until it fucks up and then everyone groans at how predictably stupid it is

never forget "What is Toronto?????"

otm we have come a long way from "How to wreck a nice beach"