“Mindless natural language AI may be good enough. In a story titled “OpenAI’s new language generator GPT-3 is shockingly good—and completely mindless” Will Douglas Heaven spends a lot of time reporting on the impressions of a number of early testers of OpenAI’s GPT-3 “language generator”. Heaven talks about comments these people have made about the “intelligence” of OpenAI’s system. What’s missing? There isn’t much in the story about another important rating: how relevant to human conversation are GPT-3’s responses?
Relevance is more important than “intelligence”. Software developers have struggled for years to produce computer generated language people will accept as conversation. Why? The reason is people expect the computer to be smarter. Why not drop back and punt? Let’s stop wasting effort on what must be called, for now, an obstacle even the best aren’t going to get over. We aren’t going to be able to produce computer generated language (“Natural Language Processing” or NLP) with “smarts” anytime soon. But mindless natural language AI is just enough
Markets still need automation to respond to questions, requests for information and more. There is a big market for OpenAI’s GPT-3 mindless natural language. Software businesses building NLP solutions can use OpenAI’s GPT-3 to produce minimal viable products (MVPs) for them.
Tech writers are stuck on “cognition” because people tend to project “humanness” onto machines. They assume markets are looking for “cognition”. But up to now markets have put up with a lot less than “cognition”. Early attempts to automate conversations (IVR systens) and written communication (NLP) have been irritating, but businesses keep using them. The mindless natural language AI built into OpenAI’s GPT-3 should get rid of a lot of the irritation.
Lessons for tech entrepreneurs? Don’t chase mirages. Ask people what they want before you waste precious resources.
Like this story? Here is a link to another you may find helpful: