ChatGPT 'defeats' sixties computer program

 

ChatGPT 'defeats' sixties computer program


Viral chatbot 'beats' ChatGPT, a computer program created in the 1960s, in a test comparing artificial intelligence and human capabilities.


In the mid-sixties, this early-stage chatbot known as 'Eliza' was created by MIT scientist Joseph Joseph Wizenbaum. Recently researchers at the University of California San Diego tested its capabilities against ChatGPT.


In those tests, Eliezer's performance is better than ChatGPT's free version 'GPT 3.5'.


The criterion used to test a machine's ability to imitate human conversation is called the 'Turing Test'. And its inventor is Alan Turing, a British computer scientist of the fifties.


A recent test involved 652 people, who had to verify whether they were talking to another human or an AI chatbot on the Internet.


However, OpenAI's financial fee-based chatbot 'GPT4' showed 41 percent more success than Eliza in tests.


Eliza was believed to be human in 27 percent of the tests. On the other hand, the success rate of GPT 3.5 is only 14 percent.


AI expert Gary Marcus described Eliza's success as "embarrassing" for modern tech companies working on AI chatbots. However, a section of academics claim that ChatGPT is not designed to show good results in the turning test.


"Reading the test papers, you won't find Eliezer beating GPT 3.5 at all surprising," Ethan Mallick, a professor of AI at the Wharton School in the US, posted on X.


“OpenAI takes the risk of impersonation seriously. ChatGPT also does not try to exceed human capabilities in ensuring RLHF (Reinforcement Learning from Human Feedback). Eliza, on the other hand, is designed in a way where it is able to exploit our psychology and pass the test.”


Another possible reason why test takers perceived Eliza as human is that it has a higher tendency to 'get wrong answers' than conventional AI models. As a result, many assumed that it was human and could not answer the question.


Although not part of the study, Princeton University computer science professor Arvind Narayanan said, “It is not possible to verify the chatbot's capabilities from the test results. ChatGPT has its own format, where there is no opportunity to express opinions. So it seems less humane than that.”


The study was published under the title 'Does GPT 4 Pass the Turning Test'.

Post a Comment

Previous Post Next Post

POST ADS1

POST ADS 2