A new era for ChatGPT? GPT-4 may have passed the famous Turing test

The use of language models such as the famous ChatGPT is becoming more and more common in society and the model is getting better and better. Since the release of ChatGPT OpenAI has already tested new updated versions which are better than the old versions. Recently, the company released GPT-4, which offers even better performance compared to other versions.

Models are so good at talking and answering questions that they often give the impression we are talking to another person. The test of whether an artificial intelligence can imitate a human and deceiving other people means the Turing test. The test was proposed in 1950 by Alan Turing in his famous paper on thinking machines.

A new study by researchers at the University of San Diego concluded that the GPT-4 language model passes the Turing test. It would be able to pretend to be a human in order to deceive other people. The group tested three language models, including GPT-3, and had a group of people interact with each model or with a person for 5 minutes without knowing.

Article from 1950

In 1950 Alan Turing published his famous article Computing Machinery and Intelligence, which became one of the most important articles in computer science. At the beginning of the article he asks whether machines can think, thereby triggering a discussion about possible artificial intelligence. This article is considered one of the starting points in the search for artificial intelligence.

In the 1950 paper, Turing calls these machines, which would do the same things as humans, thinking machines.

One of the key points of the article is the discussion that Turing explores the possibility of a thinking machine. The article has a very deep philosophical approach. It stands out especially for its discussion of how machines can learn through patterns. Learning through patterns is the basis of machine learning today.

Test the Turing

Within the article There is also a proposal for a test that became known as the Turing test. The idea behind the test would be to test whether a machine would be able to pretend to be a human and deceive another. It could make an interrogator believe that he was talking to another human via text message. If the machine successfully passes this test, it could be considered intelligent.

The In the following decades, these tests gained importance and several competitions were launched in which researchers and companies tested their machines. One of the applications of the Turing test is CAPTCHAs (Completely Automated Public Turing test to tell Computers and Humans Apart), which are used on websites to find possible bots.

Language models

A language model is a model that has been trained to understand, create, or interact with text. The best-known example is models that learn to translate from one language to another. The idea is that the language model receives an input, which may or may not be text, and returns an answer in texts or sentences. The specialty is called Natural Language Processing (NLP).

Currently, language models are advanced and can maintain a conversation like Answering questions and performing tasks such as summarizing a textT. Most of these current models are based on neural networks and a technique called Transformers. They are trained with a large amount of text in which they can find patterns, interpret text and perform tasks.

Which models pass the test?

A group of researchers from the University of San Diego examined current language models proposed to subject GPT-3.5 and GPT-4 to the Turing testThe group’s idea was to get people to chat for 5 minutes via text with a person. These people could be real people or a language model to be tested.

In the newspaper, TThey concluded that GPT-4 managed to fool people in half the casesTo confirm the test, the group also used the chatbot Eliza, which has a performance of about 22%. So the study concludes that GPT-4 would have passed the Turing test, but humans are still better at identifying other humans in most cases.

GPT-4

GPT-4 is an updated version of the language models underlying ChatGPT. These models were created by the company OpenAI and are in their fourth generation. Recently, OpenAI announced that ChatGPT would be based on an even more optimized version of GPT-4, called GPT-4o. One of the biggest differences is GPT-4’s ability to produce longer, more coherent texts.

Aside from that, ChatGPT also has an interface that can search the Internet and answer current questions. This is an improvement over GPT-3, which was one of the first to be used in ChatGPT when it was introduced, but for which information was only available until 2022.

Message reference:

Jones & Bergen 2024 Humans cannot distinguish GPT-4 from a human in a Turing test Retrieved 2018-08-18.