In order to not make this article a rant about global capitalism, contemplation of the future of AI will not be included in this post.
Generative AI is a type of AI which coaliates a large amount of data into a probability distribution, then roughly resythesizes the data by drawing at random from that distribution. Naturally, some information from the original data is lost, but the ability to create new, similar data is the objective.
One of the many issues with generative AI models is that there is no requirement or expectation of any level of usability of generated outputs. From the perspective of the computer, the only measure of usefulness is the degree to which the output resembles the input. This can lead to problems such as “hallucinations,” where AI text generators create facts or references which are verifiably false. This issue, however, is user error, since generative AI systems do not have any actual relationship with truth, except where truth happens to correlate with likelihood.
This can be a substantial issue on the area of education, since many people treat the AI as if it were a person. It is not. It has no feelings, and the fact that it can speak does not mean that it possesses anything resembling a soul. Because people treat AI with respect, they can often be decieved by the incorrect results it can generate. It should be reiterated that incorrect, in this case, does not mean that the AI did anything which it was not designed to do, simply that it was not designed to produce factual information. Not everyone is aware of this fact, though, and both students and educators may end up using AI to perform research or create media. As long as this is done with care, and the information produced by the AI is verified, this poses no issue, but the information is not always verified after someone generates it.
Another issue with AI is, of course, the fact that AI companies are notorious for not gathering data in an ethical manner. In addition to public domain works, such as ancient books and paintings, generative AI models are trained on essentially any samples which they can gain access to, with or without permission. These articles are examples of lawsuits related to the training of AI on copyrighted materials, something which is already well documented.
Perhaps someday, this article, too, will become part of the homogenous slurry which composes generative AI.