Google engineer thinks he's discovered sentient AI
20 June 2022
One of the types of AI that is progressing at speed at the moment, in places like Google's AI labs and at a company called OpenAI, is Natural Language Processing algorithms. These deep learning algorithms are pieces of software that are “trained” by getting them to process (read) lots and lots of human written text, and try to infer a set of rules for how to create new text like it's been reading. This source text is usually documents from the internet (using software that crawls through websites, linking to new sites and saving all the text it finds). Once the AI has been trained, when given a new piece of text, like a sentence, it will try to guess which word is most likely to come next. When it does this repeatedly, it can form entire sentences and paragraphs, guessing as it goes - and because its learning algorithm has figured out not only the rules of grammar but also the ways in which humans usually communicate (which words are relevant to a topic, etc), the most recent NLP algorithms do a really good job of coming across as human. They can also use this same technique to draw pictures - Dall-E 2 is amazing at creating unique images given a prompt like “a cat dressed as Napoleon holding cheese” - with the phrase “a propaganda poster depicting a cat dressed as french emperor napoleon holding a piece of cheese” giving this result: