That’s an interesting point, Dave. I also imagine that understanding something about statistics and probability will help people grasp AI/ML concepts more easily. Notably however, the people who first worked on AI, who had very strong math literacy, engaged in anthropomorphism from the start — hence the name artificial intelligence, and the founding question “Can machines think?”. And this trend continued as the field developed (e.g. adding terms like machine learning, neural networks). So understanding statistics and probability clearly doesn’t prevent people from anthropomorphising computing technologies.

You might be interested in Conrad Wolfram’s work on developing a mathematics curriculum “for the AI age” — he spoke at our seminar series last October.

You’ve illustrated your points with some great examples of how bias predictions can be a consequence of the bias data used to train ML models.

With the recent emergence of publicly accessible applications that use large language models, such as ChatGPT, it’s not hard to imagine that, without education, many people won’t question their accuracy and will trust their output.

The anthropic principle applied to physics and cosmology is an embarrassment to modern scientific thought. The same applies to the technology of AI. Artificial Intelligence, as it exists today, comes nowhere near human thought. It currently resembles those few people who are able to do extraordinary feats of memory, computation, musical performance, visual arts, and other activities without the benefit of education or training at its best. This “Rain Man” phenomenon that was known by the unfortunate and unkind epithet of “idiot savant” occurs spontaneously in humans. In AI it requires massive amounts of data fed into a network as “training”.

The AIs in actual use today are totally influenced by the choice of data, which brings us back to an early concept of computing, GIGO, Garbage In Garbage Out. So when, for instance, an AI is trained with existing historical records of people who are eligible for bank loans, it will continue to discriminate against certain disinfranchised people who have been historically discriminated against. Of course the same negative results contaminate law enforcement, employment, rent taking, and other activities previously “legally” and “socially” designed to keep “those people” in “their place”.

Another harmful effect of AI is the confabulation of text that results in conspiracy theories being accepted by some people. This obviously results in political extremism and violence. In these situations, AI could be said to stand for Artificial Ignorance.