BY ROCCO MAGLIO, MS, CISSP
Artificial intelligence has the confidence of a young teen. Its responses indicate that it is incredibly sure of its answers. Humans run the spectrum from very confident to unsure, usually based on the certitude that they have in the answer.
An Instagram/Facebook account called “thesportsmemery” had a story about Tugg Buck, a made-up undrafted NFL player. Google search’s AI presents the story as gospel when asked “Is Tug Buck a real football player?”
The disconnect is that the AI’s response is super confident, but it should not be. There is only one result in the search, which means that it is a questionable result. A single source should always be questioned.
AI’s confidence makes it seem trustworthy, but this is a con. The AI definitive answers mean that people will trust the responses at first, but this trust will be significantly damaged by confident, incorrect responses.
The AI should calculate the probability that something is possible and calculate its response based on how likely it is to be true.
This is what often happens as people get older. Initially, when they are young, they are confident that they know the answers and confidently tell people what they believe is the solution. As they age and gain experience, they see that there are several possible answers and are less confident with responses.
These AI models seem to be following the pattern of a young child. The response is very confident, even when the answer is unclear. The result is that questioners expect that the AI was very sure of its response, even though it is likely incorrect.
There needs to be a score or some other indication on how probable AI determines the result is to be true. In the case of Tugg Buck, the score would be low since only one source was reporting it. A successful athlete would be expected to have more than one source reporting on them. The super confident answer to every question is great for demonstrations, but will quickly lead to mistrust of the AI.
Rocco Maglio is co-founder of Hernando Sun, a software engineer and cybersecurity expert.
