Hacker News Reader

Hacker News

Comment by Geee | Hacker News Reader

Full thread

Geee

I made a joke once, that in the future captchas would be so difficult that only bots are able to get in.

View on HN

It's the saddest joke ever. Distorted text got pretty bad for a while there.

We've crossed that threshold a couple of times. I think that's why we keep getting new captchas.

That joke in video form:

https://www.youtube.com/watch?v=WqnXp6Saa8Y

I've been wondering if the captchas are Google's way to get cheap image classification by humans.

Isn't this why Google Voice offered speech recognition, with a UI to send corrections to them?

There's also the "please select all squares that contain Sarah Connor" meme.

That is exactly what it is.

EDIT: From Google's own reCAPTCHA website - https://www.google.com/recaptcha/intro/v3.html#the-recaptcha...

> Hundreds of millions of CAPTCHAs are solved by people every day. reCAPTCHA makes positive use of this human effort by channeling the time spent solving CAPTCHAs into annotating images and building machine learning datasets. This in turn helps improve maps and solve hard AI problems.

Thnx. When people talk about captchas I usually reply something about how google is using them for machine learning, and suprisingly often people shrug it off as a conspiracy-theory.

Why have I never just done what you did, linked to google stating that this is what it is in plain text....

Of course they are!

That's exactly what it is. If you recall, these image based captchas were originally in 2007 for digitizing books. [1]

In 2012, Google started using captchas to identify house numbers from google street view. [2]

Now, users are identifying cars, bikes, traffic lights, and crosswalks most of the time. While Google/Alphabet has been mum on what specifically they're using the data for, it is speculated by engineers at competing firms that they are using this data to help Alphabet's subsidiary, Waymo, with its self-driving car program. [3] This data is either used as training data or to validate outputs that were already classified by their system.

[1] https://en.wikipedia.org/wiki/ReCAPTCHA

[2] https://techcrunch.com/2012/03/29/google-now-using-recaptcha...

[3] https://www.ceros.com/originals/recaptcha-waymo-future-of-se...

> Now, users are identifying cars, bikes, traffic lights, and crosswalks most of the time.

I got umbrellas the other day.

If you are a self driving car you definitely do not want to misclassify an umbrella as they almost never get about independently.

I have serious, serious doubts about the whole approach of "identify known objects in a camera feed so you can try not to hit them". I don't know what the current approaches are but there needs to be some kind of subsumption setup where if you don't recognize an object, at the very least you assume it's a stationary solid object and you don't hit it. It doesn't matter if your classifier says it's a brick, or a grandma, or a hibiscus, or (unknown). Unless you've positively identified it as something that's safe to hit (say, a plastic bag wafting along in the breeze) then don't hit it.

If you do have an identification then you can layer behaviours on top of this (eg. Is it a person? They usually walk forward or in the direction they're looking, so anticipate this.) But the default behaviour cannot be "dunno what that is so I'll ignore it".

"Don't hit the thing" is the most basic, fundamental behaviour for a self driving vehicle.

I couldn't agree more.

And this is why I like to call them "Self-Crashing Cars".

I'm fully capable of ploughing in to moving and stationary objects at high speed, under acceleration, myself. Thank you very much.

And OTA updates that can change the behaviour of the vehicle between uses? Just no. I have enough trouble switching between my Japanese and European cars (one of each) where the indicator stalks are on opposite sides of the steering column. I'm forever indicating my intention to turn with the wipers!

I'm not replying to myself but to all who posted to confirm my suspicion. I'm amazed that I've been so blind to it. It's so blatant. And now it makes sense how many of the challenges are: traffic lights, school buses, etc.

Reminds me of this joke from Futurama, where a Robot has a captcha tattooed on his arm.

Comic version: https://i.redd.it/dnosraioogm31.jpg

https://www.youtube.com/watch?v=fsF7enQY8uI

I remember some sites had a captcha where you had to count something, maybe it was kittens. Think I managed to solve them 1/3 of the time

RapidShare.

Reply on news.ycombinator.com