Member-only story
What AI Does (and Doesn’t) Change About Intellectual Property
Some Thoughts on the IP Landscape in the midst of increasingly potent AI capabilities
I Fought the Law and the Law Slowly Caught Up
I recently heard a podcast episode where one of the speakers referenced this Edward O. Wilson quote: “…We have Paleolithic emotions, medieval institutions and godlike technology…”
Intellectual property law is old, but laws are being thrown around to try and stem the tide of things these new AI tools are capable of doing, which are able to learn pretty darn fast to write text and produce images (even with weird hands).
But violating copyright and other forms of IP law are nothing new; you can apparently steal Marvin Gaye’s “vibe”, all without the power of AI, or even more ridiculous is how YouTube enforces copyright via copyright strikes.
My goal is not to debate the merits of copyright, but rather, see what sort of spice Machine Learning will add to an already messy dish. Let’s start with the basics —
“I Know Kung Fu”
In The Matrix, Neo needs to be able to learn how to fight. A (now dated) computed animation pops up on the screen as his brain is pumped full of all sorts of different fighting styles. The line delivery he gives after this cramming session is legendary — “I know kung fu.”
These machine learning models, whether trained on an internet’s worth of content, or a handful of pictures of something in slightly different poses, learn slower than Neo does in this clip, but MUCH faster than you’d consider a human or any other animal would learn.
My daughter, as an example, can draw a few sets of lines going the same direction (she’s two) and speak a handful of words, but she is no ChatGPT or DALL-E. She’s going to take years to get to even drawing stick figures and scrawling out her name in barely readable kindergarten hand writing.
Many many hours of GPUs churning through datasets in parallel gives a training time on the order of hours or days, not the years that humans take to learn this information. And then within a very short window (seconds, minutes), these tools can “infer” from prompts to generate text and images…
