In 1950, Alan Turing posed the question in a scientific paper whether machines can think. For this purpose, he developed a mental game in which test subjects had to hold a text-based conversation with both a human and a machine for five minutes and guess which of the two was in fact human. According to Turing’s definition, if “more than 30%” were wrong, the test was passed.
While in the eyes of many experts the “Turing Test” has not been passed to this day, there is eager debate about whether it is at all suitable for examining human thinking. However, we are all familiar with a descendant of this idea: CPATCHAs (“Completely Automated Turing Test to Tell Computers and Humans Apart”), which are supposed to prevent bots on the Internet from using services that are restricted to human individuals only.
For some time now, however, the question has been raised whether CAPTCHAs are still a sufficient barrier for machines. Artificial intelligence (AI) has made it possible to locate fire hydrants in photos and transcribe scrawly writing with simple means – at the latest since the turbulent year 2022. But where does that leave us? Which tasks are today still unsolvable for machines, but not a challenge for humans? We get to the bottom of this question, design CAPTAIHAs (“Completely Automated Public Turing Test to Tell AI Systems and Humans Apart”) and evaluate them by back-hacking.