There is a good joke that’s been around for a while, but it’s especially pertinent when it comes to evaluating AI: “It must be true, I read it on the Internet.” Everyone knows this means it’s more likely not to be true. But when it comes to AI, almost everything it “knows” comes from the Internet. And because it tends to weigh true and false by frequency of encounter, the more AI encounters the same images, assertions, statements, treatments, opinions, etc., the more statistically weighted it will be. The term, “There’s safety in numbers,” comes to mind in that the idea is, the more frequently something is encountered, the more genuine it probably is. This becomes AI’s “default assumption” about the material it is trained with. It can only utilize, evaluate, and regurgitate the material it is trained with. This turned out to be quite a problem early on because the sheer amount of racist, violent, and derogatory material on the Internet was not fully appreciated until AI started digesting it. It became necessary to employ untold thousands of low-paid (on the order of two dollars a day) “content evaluators,” mostly in third-world countries, to filter out gore, hate speech, child sexual abuse material, and pornographic images. If AI read it on the Internet, it must be true? What is Creator’s perspective?

This question is focusing quite keenly on the limits of language and the consequences you see play out in how language is used, and its inadequacy, and how easily it can be manipulated and even weaponized for destructive ends. The largest problem by far is one of the sheer ignorance of human knowledge individually and collectively. So the Internet, inevitably, as a repository of human knowledge comprised quite heavily of a huge collection of communications, not of a scholarly caliber but casual communications, impromptu, often based on partial or faulty memory but conveyed as truth, will be inherently of questionable reliability. So, in a sense, the whole idea of AI striving to be some kind of human transcendence to result eventually in a kind of superhuman capability to generate new knowledge, Nobel Prize winning science, and advancement of human civilization is a kind of folly. The race to develop a higher and higher level of humanlike performance is an illusion because it is based on the fact it draws from a partial vacuum, meaning incomplete human understanding and awareness. You are not whole. You do not know everything, individually or collectively, through pooling together the knowledge of individuals. Everyone is ignorant about most things and only has partial knowledge in selected areas, with considerable overlap but about the most commonplace accepted descriptions of reality as experienced by humans directly. There, of course, is quite an overlay of impressions stemming from emotional reactions and human yearnings for something better, creating expectations and conjectures and extrapolations about what might be possible. This has led to many new artistic endeavors, as with science fiction, but it is one thing to envision something possible and another to turn it into reality. One of the truisms of physical science is that "nature abhors a vacuum" and, in a sense, wants to fill it. But what can fill it has to come from somewhere. It will be something already in existence that fills that vacuum, eventually, one way or another. When it comes to information, the same thing is true. What is created by AI to fill the knowledge gap of humanity will be a simulation of what is believed to exist currently, most of which is flawed, incomplete, or faulty altogether. Even science, the icon of investigative effort leading to deeper understanding of reality, is always provisional and almost always incomplete and overturned by proving it to be faulty, because it is always a series of conjectures, and the ways of proving scientific hypotheses are heavily flawed and usually inadequate to the task beyond making incremental gains in understanding of what turn out to always be surface details poorly understood. The meager state of human knowledge at present is a bigger limitation than appreciated. What we can see that you cannot, as a human being, is that even things you yearn for are beyond human understanding in their nature, their composition, and realization in terms of bringing them into reality. Those knowledge gaps are not penetrable by pattern matching, text extrusion algorithms like chatbots trained on the contents of the current Internet.