Scientists discover AI has already passed the notorious Turing Test

One of the most prolific tests a machine can undertake is called the Turing Test, created by Alan Turing in 1950 that's purpose was to test if a machine is capable of displaying a level of intelligence that can convince, or fool a human they are engaging with another human.

2

Researchers have put OpenAI's GPT-4 up against the Turing Test, and according to a new study the impressive large language model passed the test. The not-yet-peer-reviewed-study asked 500 people to engage with four different respondents. One of those respondents was a human, another was a 1960s-era AI called ELIZA, another was OpenAI's less sophisticated GPT-3.5, and finally GPT-4.

The paper states each conversation lasted approximately five minutes, and according to the study's results, humans found GPT-4 to be a human 54% of the time. These results lead researchers to claim GPT-4 has already passed the Turing Test. However, the study also indicated study participants believed the human respondent was a human 67% of the time, while ELIZA scored just 22%, and GPT-3.5 scored 50%.

The Turing Test is certainly an interesting approach to measuring the intelligence of machines, and the results from this study are nonetheless extremely intriguing, but there is the possibility the Turing Test is outdated and doesn't include the nuances that modern technology has introduced. For example, convincing humans they are speaking to another human requires socio-emotional factors that occur during an interaction, which doesn't necessarily coincide with incredible intelligence.

While the testing method for this study can be debated upon, its results, at the very least, showcase the drastic improvement AI has made in such a short period of time, particularly the GPT language models.