Continuing the discussion from On enforcing password changes: Don't!:
On a different tack, did you know that captcha is an acronym for ‘Completely Automated Public Turing test to tell Computers and Humans Apart’? Next question: How many people do you know, know the Turing test and the criticisms against it? Did you also know that captcha tech now uses mouse movement to tell that you’re not a bot (because humans don’t move like robots)?
I had no idea that’s what CAPTCHA stood for. The amount of time I have sat arguing with myself as to whether a bike/traffic light is overlapping into a different square and never thought to ask what CAPTCHA meant.
Now imagine that you are blind or partially sighted and you’ll see how difficult it is for people with disabilities to prove they are not robots. It’s good to see that more captchas are dispensing with the confusing images.
Great observation about the Turing Test, Karen. It’s got a long and weird background based partly in Turing’s sexuality and gender norms of the 1950s. The Wikipedia page is a fascinating read: https://en.wikipedia.org/wiki/Turing_test
And that page has expanded a lot recently since the introduction to the general public of LLMs with chat interfaces. Despite my huge respect for Turing and all his amazing and diverse work… the Turing Test has looked rather shabby for many years now, and I think fall flat when it comes into contact with LLMs. (Or fell flat when ELIZA was shown to computer niaves in 1963 - depends on how you look at it).
K
A follow up question - have you looked into the role of CAPTCHA for training artificial vision models Karen? I haven’t done any reading on this, but it seems obvious to me that they’re also using humans to train AIs in the backend. Am I onto something?
K
Yes you are onto something Keith. I’ve just done a quick n dirty search on Google Scholar using “artificial vision models captcha” as the search words. It appears that the research is (logically) about how these models are designed not to be broken.
Here’s a 2022 article that evaluates a range of CAPTCHA models.
Kumar M, Jindal MK, Kumar M. A systematic survey on CAPTCHA recognition: types, creation and breaking techniques. Archives of Computational Methods in Engineering. 2022 Mar;29(2):1107-36.
You may find this article interesting because of its focus on a vision AI model. Yes, humans create these tools and models but when we’re dealing with generative AI it becomes a different ballgame.
George D, Lehrach W, Kansky K, Lázaro-Gredilla M, Laan C, Marthi B, Lou X, Meng Z, Liu Y, Wang H, Lavin A. A generative vision model that trains with high data efficiency and breaks text-based CAPTCHAs. Science. 2017 Dec 8;358(6368):eaag2612.
There’s heaps more in Google Scholar (and I’m sure if you were to check out ACM and IEEE journals you’ll find more interesting/useful stuff that’s of a high calibre).