A group of researchers from Georgia State University presented the results of an interesting study regarding the morality.
The tests in question asked questions on this topic to GPT-4well-known model LLM Of OpenAI, and a group of university students. The aim of this research is to evaluate the level of morality reached by theArtificial intelligencetaking humans as a parameter.
The researchers prepared ten distinct scenarios, some showing unacceptable human behavior and others completely common. With respect to the individual cases, GPT-4 was then asked to evaluate the individual scenario, describing with a 600 word commentary because it evaluates the same as morally acceptable or not. They then participated in the same test 68 university studentsall participants in a philosophy course.
A research that leaves many doubts about AI and morality
Once the answers were prepared, both from GPT-4 and from the students, they were proposed to a “jury” of 299 adults, evaluating the answers. The latter, obviously, was not affected by the fact that some of them were the result of the work of an AI.
Bearing in mind that morality is something that remains purely subjective and closely connected to cultural factorsGPT-4 responses were rated better overall than student responses.
Despite this, the same research group wanted to warn against AI. In fact, GPT-4 sometimes behaves as one would psychopath. Although he is able to distinguish different types of violation of morality, in fact, but without evaluating them by “sensations”. According to the researchers, GPT-4 knows well how to answer certain questions, but without really understanding what morality is.
As already mentioned, morality varies from culture to culture and, therefore, determining absolutely what is right or wrong is presumptuous to say the least. Despite efforts to create something more than a simple machine, AI still remains an emotionless tool at present.