Editing CAPTCHA (section)

== Characteristics ==
CAPTCHAs are automated, requiring little human maintenance or intervention to administer, producing benefits in cost and reliability.<ref>{{Cite web |title=How CAPTCHAs work {{!}} What does CAPTCHA mean? |url=https://www.cloudflare.com/learning/bots/how-captchas-work/ |url-status=live |access-date=October 27, 2022 |website=Cloudflare |archive-date=27 October 2022 |archive-url=https://web.archive.org/web/20221027061629/https://www.cloudflare.com/learning/bots/how-captchas-work/ }}</ref>

Modern text-based CAPTCHAs are designed such that they require the simultaneous use of three separate abilities—invariant recognition, [[image segmentation|segmentation]], and parsing to complete the task.<ref>{{cite journal|last1=Chellapilla|first1=Kumar|first2=Kevin|last2=Larson|first3=Patrice|last3=Simard|first4=Mary|last4=Czerwinski|title=Designing Human Friendly Human Interaction Proofs (HIPs)|journal=Microsoft Research|url=https://research.microsoft.com/pubs/101726/HIPSCHI2005.pdf|archive-url=https://web.archive.org/web/20150410195118/http://research.microsoft.com/pubs/101726/HIPSCHI2005.pdf|archive-date=10 April 2015}}</ref>

* Invariant recognition refers to the ability to recognize letters despite a large amount of variation in their shapes.<ref>{{Cite journal |last1=Karimi-Rouzbahani |first1=Hamid |last2=Bagheri |first2=Nasour |last3=Ebrahimpour |first3=Reza |date=2017-10-31 |title=Invariant object recognition is a personalized selection of invariant features in humans, not simply explained by hierarchical feed-forward vision models |journal=Scientific Reports |language=en |volume=7 |issue=1 |pages=14402 |doi=10.1038/s41598-017-13756-8 |pmid=29089520 |pmc=5663844 |bibcode=2017NatSR...714402K |issn=2045-2322}}</ref>
* Segmentation is the ability to separate one letter from another, made difficult in CAPTCHAs.
* Parsing refers to the ability to understand the CAPTCHA holistically, in order to correctly identify each character.<ref>{{Cite web |title=Making CAPTCHAs Expensive Again: If You're Using Text-Based CAPTCHAs, You're Doing It Wrong {{!}} Tripwire |url=https://www.tripwire.com/state-of-security/youre-using-text-based-captchas-youre-wrong-making-captchas-expensive |access-date=2022-10-28 |website=www.tripwire.com |archive-date=28 October 2022 |archive-url=https://web.archive.org/web/20221028040010/https://www.tripwire.com/state-of-security/youre-using-text-based-captchas-youre-wrong-making-captchas-expensive |url-status=live }}</ref>

Each of these problems poses a significant challenge for a computer, even in isolation. Therefore, these three techniques in tandem make CAPTCHAs difficult for computers to solve.<ref name=bursz>{{cite book|last1=Bursztein|first1=Elie|first2=Matthieu|last2=Martin|first3=John C.|last3=Mitchell|chapter=Text-based CAPTCHA Strengths and Weaknesses|title=ACM Computer and Communication Security 2011 (CSS'2011)|year=2011|chapter-url=https://www.elie.net/publication/text-based-captcha-strengths-and-weaknesses|access-date=5 April 2016|archive-date=24 November 2015|archive-url=https://web.archive.org/web/20151124055747/https://www.elie.net/publication/text-based-captcha-strengths-and-weaknesses|url-status=live}}</ref>

Whilst primarily used for security reasons, CAPTCHAs can also serve as a benchmark task for artificial intelligence technologies. According to an article by Ahn, Blum and Langford,<ref name=Ahn2003>{{Cite book| chapter-url=https://link.springer.com/content/pdf/10.1007/3-540-39200-9_18.pdf| doi=10.1007/3-540-39200-9_18| chapter=CAPTCHA: Using Hard AI Problems for Security| title=Advances in Cryptology—EUROCRYPT 2003| volume=2656| pages=294–311| series=Lecture Notes in Computer Science| year=2003| last1=von Ahn| first1=Luis| last2=Blum| first2=Manuel| last3=Hopper| first3=Nicholas J.| last4=Langford| first4=John| isbn=978-3-540-14039-9| s2cid=5658745| access-date=30 August 2019| archive-date=4 May 2019| archive-url=https://web.archive.org/web/20190504115630/https://link.springer.com/content/pdf/10.1007%2F3-540-39200-9_18.pdf| url-status=live}}</ref> "any program that passes the tests generated by a CAPTCHA can be used to solve a hard unsolved AI problem."<ref>{{Cite conference |last=Moy |first=Gabriel |last2=Jones |first2=Nathan |last3=Harkless |first3=Curt |last4=Potter |first4=Randall |date=2004 |title=Distortion estimation techniques in solving visual CAPTCHAs |url=http://www.cs.duke.edu/courses/cps296.3/spring07/breaking_captchas.pdf |publisher=IEEE |volume=2 |pages=23–28 |doi=10.1109/CVPR.2004.1315140 |isbn=978-0-7695-2158-9|archiveurl=https://web.archive.org/web/20200729175253/https://www2.cs.duke.edu/courses/cps296.3/spring07/breaking_captchas.pdf |archivedate=29 July 2020|conference=[[Conference on Computer Vision and Pattern Recognition|Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition]]}}</ref> They argue that the advantages of using [[AI-complete|hard AI]] problems as a means for security are twofold. Either the problem goes unsolved and there remains a reliable method for distinguishing humans from computers, or the problem is solved and a difficult AI problem is resolved along with it.<ref name="Ahn2003" />