The future of student personal data
For colleges and universities, this open approach to managing student data is a real relief in terms of responsibility for data security.
Publish at December 04 2024 Updated December 04 2024
"The Turing Test is a proposed test of artificial intelligence based on a machine's ability to imitate human conversation. Described by Alan Turing in 1950, the test involves putting a human in a blind confrontation with a computer and another human.
If the person engaging in the conversation is unable to tell which of the two is a computer, the computer software can be considered to have passed the test."
Human or not proposes to measure the extent to which we can detect whether we are in conversation with a machine or not. The quality of the interlocutors, when there are any, can be quite random.
The game consists in trying to pass yourself off as a machine and guessing whether you're actually dealing with an artificial intelligence or not.
When the conversation becomes too "machine-like", it may be a human trying to pass himself off as a machine, as A.I. don't spontaneously caricature themselves. Rather, they try to give themselves a personality. To get them to reveal themselves, you can offer them elements of the same context in an incongruous relationship... they then tend to hallucinate. It's also wise not to make too many spelling mistakes to avoid being identified, as machines tend to write very well. You can also write in a language other than English; this is particularly discriminating, as A.I.s are very good in several languages, which is rarer in humans.
The retro look of the site refers to the original AI21 Labs site, which was a research project that nevertheless concluded that, at the time of the test (2022-23), 68% of people were able to correctly identify a human or a robot, with greater ease in detecting a human. The Turing test is apparently becoming less and less discriminating, with now a 52% success rate, barely more than chance...
The site currently uses Jurassic-2, GPT-4, Claude and Cohere, which makes the challenge all the more interesting.