Jul 17

Tim O'Reilly

Tim O'Reilly

Another CAPTCHA -- But I failed (partly)

On my last entry, synonymous commented: "I like the idea, but I think this captcha mashup may actually be more effective." I cracked up when I saw this. It uses "the hotornot API" (Web 2.0 is getting out of hand!) to offer up pictures of nine women (or men) and asks you to prove you're human by selecting the three "hot" ones. Politically incorrect, funny and thought-provoking, and potentially flawed by sexual bias to boot. I sent on a link to the radar team with the comment: "I have no problem with picking out the 'hot' women, but I struggle with the men. Some of the obviously 'hot' ones (presumably the right answers) look like doofuses to me, while with women, the mentula ("little mind" -- Latin slang for the male anatomy) easily gets it right. I wonder if women find the men as easy, and if so, how much tastes vary."

Nat replied: "I had the same problem as you. I blazed through the women, five correct answers in a row. Switched to men for a challenge and immediately stumbled. I tried ten times and got one (1) set right. The only rule I found is that men with their shirts off, exposing a ripply chest, are probably hot. This is much the same as the rule that says that women with their shirts off exposing a ripply chest are also hot. I suspect there are two problems: you and I are not hormonally equipped to find the hot men, and that hotornot lacks female users to contribute and vote up the hot men. There were sets of nine pictures where I'm prepared to swear money on there not being a single hot man. Unless dumpy with bad hair is hot, in which case I'm off to the bar to make me some new friends ..."

Clearly, uses "collective intelligence" (or at least collective taste) to rate the images, so it is a good consensus indicator of human judgment, but i found the sexual bias to be the most fascinating part. It's rather amazing to have a program say to you " Wrong! Die, bot, die." I tend to think I'm human, but the fact that a program can't tell that, based on a sexual bias, is very thought provoking. I'd be interested for any of our female readers to see if the program works for them, or if Nat's supposition that hotornot has an inherent male bias is correct.

tags:   | comments: 18   | Sphere It

Previous  |  Next

0 TrackBacks

TrackBack URL for this entry:

Comments: 18

  adamsj [07.17.06 08:08 PM]

Did you happen to read David Leavitt's recent short biography of Alan Turing, The Man Who Knew Too Much? It touched heavily (though not entirely convincingly) on how Turing's sexuality related to his ideas about machine intelligence.

  Ben Hirsch [07.17.06 08:10 PM]

How are there no comments here by now? I too failed miserably with the men. Would love to hear how any women out there did with this.

  Anonymous [07.17.06 08:12 PM]

I'm a gay man, and I did much worse on the men than the women. I think there's more of a problem with hotornot's collective intelligence than there are with Tim's hormones. :)

  Tim O'Reilly [07.17.06 08:14 PM]

I asked my wife and one of my daughters to try it. My daughter got only two out of four of the men, and my wife got zero for two. Watching over their shoulders, I also had to say that it looks to me like the selection for men is really poor, giving credence to Nat's supposition that the hotornot database itself suffers from a gender imbalance, with too few men contributing pictures and too few women contributing ratings.

  Ja [07.17.06 10:11 PM]

My brother and I were just musing the other day that the hard part was finding 3 hot girls on there. It was usually 2 and then one that was fairly homely yet not as obviously grotesque as the others, heh.

But, more seriously, these folks (hotcaptcha) just won 2nd place at MashupCamp2 (and walked away with a really nice new intel duo imac). What does this say about how seriously any of us can take these "mashups" and the state of the web industry currently?

It's a shame because there are some truly useful mashups that add a tremendous amount of value and make web-tasks more efficient. Unfortunately, all the attention is still depdendent on the "neat" factor. In this case I think it was based on the handing out crack at the door factor. It also sends out the wrong message to developers.

  hfb [07.17.06 10:52 PM]

Wow, if those are the hottest guys the hotornot db can provide, it's no wonder nobody can find the hot guys. Maybe the theory needs to be that the kinds of guys to hang out on line and post their picture on hotornot are incredibly likely to be decidedly unhot and dorktastic looking. Jesus, those photos are utterly awful. that pic of Uri as I think that would be a step up in quality :)

  Sencer [07.17.06 11:39 PM]

The problem is, that the question is posed or interpreted in the wrong way. I shouldn't be pick the 3 _you_ find hot. It should be read as: Pick the three which you think "the public" finds hot.

Anyway, there are only 84 way to choose 3 from 9 (removing the permutations and without putting the picks back), which yields a large enough random percentage of correct random picks for bots to simply hammer away on it anyhow...

  ardief [07.18.06 02:18 AM]

Er...I am a woman, and I really struggled with the men - in fact, I failed! But I agree with Sencer: it should be phrased as 'what the public thinks', otherwise it's too subjective!

  Michael Moncur [07.18.06 03:55 AM]

Okay, I'll admit it. I'm a straight male and failed the female test 2 out of 4 times, and found myself guessing the third like "Ja" said above. Perhaps I've been happily married too long.

As a more practical and equally politically incorrect solution, a combined test that asks you to "pick the three women" or "the three men" would be almost unambiguous. We're wired pretty well to detect gender characteristics.

  Nina [07.18.06 06:28 AM]

I'm a woman who was able to get the women right five out of five, and the men, never. It was easy to follow a simple rule for the women: choose the thinnest, least clothed ones. There was no corresponding filter for the men. Ironically, this may mean it's easier for a robot to simulate the male preference (for hot women) with simple rules than the female preference...

  Tim O'Reilly [07.18.06 09:21 AM]

Nina -- I think you've hit on it. Men have a very limited set of filters. In fact, one might conclude from many observations that they aren't completely human :-) Men have very simplistic filters, while women have more sophisticated ones.

But overall, this wasn't meant to be taken seriously (I think). I posted it because of what it reveals, not because of what it does.

  adamsj [07.18.06 09:49 AM]

I'm really enjoying these days, and here's a post from there that snuggles right up against this one.

  Peter [07.18.06 10:51 AM]

someone has thought about an automated bot for solving the hotcaptcha via "barbie-matching". ;)

  Kevin Marks [07.18.06 11:30 AM]

Tim, the evolutionary psychology angle is (crudely summarized) that men pick mates by looks, and women pick mates by status, due to the asymmetry of investment required for each to have children.
So, both men and women are better tuned to 'hotness' in women, men for mate choosing, women for checking the competition. This likely underlies both the ability to discriminate in the captcha test, and the underlying paucity of rated men on hotornot.

  james [07.19.06 01:17 AM]

Hi Tim,

I wrote a response, but doesn't look like it made it through for some reason.. so I'll repost.

From the data we have, it would appear that Kevin Marks (above) is right on target. I looked into this a few years ago, it looked like the variance on men's scores was higher than on that of the women. meaning people don't agree as much when judging men than when judging women. Current hypothesis is that our society tends to judge women on looks more often, and that the media "educates" the public more frequently on what hot women are supposed to look at.. So we (both men and women) are more comfortable with, more experienced with, and are trained to vote similarly on rating women.

Correspondingly, we also noted through some data analysis done by a professor at MIT and at CMU that we gave data to, that it takes both men and women longer to rate a man than to rate a woman. That's not proof of anything, but suggests to me that we do, in fact, all have a harder time judging men.

so basically, with regard to the hotcaptcha-rating-men-mode.. while on average we will all agree which men are the hottest, because variances are higher, as you increase the number of men we all have to agree on, the probability that we will all agree on the same n people being the hottest drops rapidly.. for women, we're all so good at it , and so uniformally aligned in taste (or at least, in knowing what other people think), that the odds at n=3 seems to be pretty good (I got it right like 10/10 times with woman mode.. I actually got it right 4/5 times with men, but I figure i am probably better at rating people than most! :) )

you can qualitatively see this effect just by rating men with friends on hotornot versus rating women with friends on hotornot (everyone huddle around the computer).. you will find that you disagree on the men a lot more.


  Tim O'Reilly [07.19.06 08:45 AM]

James --

Thanks so much for the back story. It's great to get some facts into the discussion.

One question: hotcaptcha seems to reuse many of the same men (and to a lesser extent women). Is this some artifact of the way they are caching the data, or is it an artifact of the difficulty of getting really low-rated people? (I would imagine that most of the really low-rated pictures are posted as a joke.)

Also, can you tell us more about the usage of the hotornot api? I didn't even know there was one. (You should get John Musser over at to list it.)

  bridget [07.22.06 02:08 PM]

Just to round off the gender/orientation mix: I'm a lesbian woman; I got neigher the women (3 tries) nor the men (2 tries) right. I stopped after 5 tries, the "you're not human" thing started to get to me.

  A. Jorgensen [04.02.08 03:10 PM]

Well I flunked, too. But I did notice after a number of tries that the "Hot" women tended to be smiling. I think I could write an algorithm for that.

Post A Comment:

 (please be patient, comments may take awhile to post)

Type the characters you see in the picture above.