Artificial Intelligence Just Insulted Your Mother
Silicon Valley’s most cutting-edge AI model just insulted your mother. And that could be a very good thing for your portfolio…
Last week, we talked about the exciting advances in artificial intelligence (AI) — and how they could lead to massive investment opportunities in the years ahead.
But AI has a dark side…
“Medical chatbot using OpenAI’s GPT-3 told a fake patient to kill themselves,” reports a story from AI News. Yikes.
“GPT-3 is the world’s most powerful bigotry generator. What should we do about it?” asks another from The Next Web.
It’s true. The current state-of-the-art model for AI-fueled text generation sometimes says some awful things. Out of curiosity, I fired up GPT-2, the predecessor to OpenAI’s newest state-of-the-art AI model, and tried to get it to say something offensive. It didn’t take much.
(I know we’re living in an age where it’s easier than ever to offend people — but the text generated by GPT-2 was pretty objectively cringeworthy.)
But here’s a controversial idea generated by a human (me): The fact that Silicon Valley’s most cutting-edge AI says horrible things every once in a while is a good thing.
When OpenAI set out to create GPT-3, they were trying to build an artificially intelligent text generator, not a machine capable of making moral and ethical decisions about what it said.
And it does a phenomenal job at generating convincing human language.
If GPT-3 and its predecessors say terrible things occasionally, it’s because humans do too — and the corpus of language data that the model was trained on included things like bias and bigotry because they were written by humans. But the model doesn’t actually understand what it’s saying. There’s no intent to deliver a particular message.
It’s a mimic.
But think about this — the fact that GPT-3’s text is so convincingly human that actual humans get offended by it means that it’s an excellent text generator.
GPT-3’s power lies in its ability to act as a language layer on top of other algorithms that are actually making decisions, not as a stand-alone thing.
I have no doubt that we’ll figure out how to make GPT-3’s successors less offensive. Some really smart people are working on it.
Back in September, Microsoft (NASDAQ: MSFT) inked a deal with OpenAI to exclusively license GPT-3 for its products. Right now, engineers are working on integrating that AI tech into real Microsoft products.
I suspect we’ll see elements of GPT-3 in consumer software from Microsoft, as well as in its Azure cloud computing platform. And that Azure integration has the potential to truly move the needle as Microsoft competes with Amazon’s AWS platform.
This is the start of something big.
And the implications are equally big for tech investors. Stay tuned.
Jonas Elmerraji, CMT