h a l f b a k e r yTempus fudge-it.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Please log in.
Before you can vote, you need to register.
Please log in or create an account.
|
Present an agent with a CAPTCHA consisting of a 3D
generated scene. In this scene, something is weird. The
easiest way to accomplish this is to fill the scene with
many objects. One of those objects is composed of
correct parts, but the parts are interconnected in a weird
way. For example
a cat might have it's tail
sticking out of it's head. Or a table might have a 5th
leg.
If the agent clicked on the region containing the
mistake,
they would pass as a human. If the agent clicked on
some
other portion of the scene, they would be considered a
robot.
This would have several benefits:
- No more typing
- It's more enjoyable spotting things that are wrong, than
trying to pass a test imposed on you by some stupid web
site who thinks it's better than you just because it can
scramble some letters.
- Would be quicker to pass through. Humans are wired
to
spot weird stuff quickly.
- Is cross cultural
For better accuracy several of these could be chained
together. You would need to pass 3 in a row to prove
you
are a human. After all a robot could click randomly at
the
first image and it would be right 1% of the time
assuming
the image is consists of 100 areas. Passing all three by
luck would be possible only 1 in 1,000,000 tries.
Highlights for Children - What's Wrong
https://www.highlig...ids.com/whats-wrong
A few too many things, perhaps [smendler, Apr 20 2016]
[link]
|
| |
// a cat might have it's tail sticking out of it's head. // |
|
| |
They sometimes look a bit like that after we've been working on them for a while. More usually, the identifiable parts are fairly widely separated. |
|
| |
Clever idea. You'd need to generate a lot of these little puzzles, otherwise a human could just program the answers for images that the computer would be able to recognize by color properties, size, etc. |
|
| |
//You'd need to generate a lot of these little puzzles |
|
| |
They can be computer generated. You can have infinite
number of them. All the computer has to do is have 3D
models that are made out of components. The computer
will pick one of the models and algorithmically move the
component to the wrong place. For example, the cat would
be a 3D model made out of 4 legs, head, body and tail.
Computer would roll the dice and randomly move a body
part and connect it to the wrong place. |
|
| |
Unfortunately you've hit on the bit of human reasoning (object recognition) that is almost solved this very moment. |
|
| |
Object recognition is not sufficient for this task. A bot
would have to recognize the object AND figure out what's
wrong with it. |
|
| |
If you show a bot cat body parts it will probably be able
to
correctly categorize them as a "cat". What it won't be
able
to do is figure out whether the body parts are connected
together the right way. |
|
| |
Also, it won't be just cats. |
|
| |
I wonder how people with keyboards in other alphabets do those captchas. I assume there are keyboards with other alphabets. |
|
| |
How about the capcha being a link to a random idea
on the Halfbakery? The reader will then have to click
on the first word of the idea which makes it clearly
unfeasible. |
|
| |
No, I'd inadvertently used a which for that which that
should have been a that. So, that which that was
which should have been that. But point taken. |
|
| |
Actually, a better plan would be to simply enforce
the death penalty on people who write or use the
bot-software that has made the use of capchas
necessary. |
|
| |
If every human being has to go through an average
of 1 capcha per week, and if it takes 5 seconds to
do so, that amounts to 57,712 man-years of time
wasted, per year. That is roughly equivalent to
700 human lives. So the death penalty seems
entirely reasonable. |
|
| |
//A bot would have to recognize the object AND figure out what's wrong with it. // |
|
| |
//If the agent clicked on the region containing the mistake, they would pass as a human// |
|
| |
Same thing. What I'm saying is that picking out a mistake
involves 2 steps. Recognition and figuring out what's
wrong |
|
| |
// What I'm saying is that picking out a mistake involves 2 steps. Recognition and figuring out what's wrong // |
|
| |
So, just like the Primaries and the succeeding election ? |
|
| |
//If everybody could just trust each other implicitly
and totally, there’d be no requirement for all this
nonsense// That was precisely the point I thought I
was making. |
|
| |
[+]Clever. But the generating algorithm might have
difficulty making things really "wrong". |
|
| |
For instance, if they swapped the cat's left-hind-leg with the
cat's right-hind-leg, that'd be far too subtle for a human to
see it as "wrong" in a scene with 20 objects. |
|
| |
It would be rather frustrating to be told, "Sorry, you're a
robot" because some robot made swaps that it thought were
different, but really were equivalent. |
|
| |
Exactly. If a computer can generate the images then another computer should be able to decode them. The trivial case would be stealing or reverse engineering the code. |
|
| |
Next up, the "Where's Wally ?" Captcha... |
|
| |
//should be able to decode // |
|
| |
{shakes head wisely, muttering something about entropy} |
|
| |