I’ll accept something as AGI when it can write a solid set of original jokes good enough to make real people laugh
Given the overall GPT-based AI progress in the last few months, I’ve been thinking a lot about AGI. Specifically, how will we know if it happens?
I think there are lots of ways to answer that based on what you think you’re getting from an AGI. In this case I’m thinking of capabilities, not consciousness.
And for capabilities, I think one ability that will truly impress me, and make me wonder if we have a real AGI, will be stand-up comedy.
Think about how hard stand-up is. Think about everything that goes into it. It’s one part society commentary. Another part reading a room. Another part phrasing and timing. And some other parts I am sure I’m not thinking of.
But once an AI can write an audio track, like you might hear on stand-up comedy radio, or on a podcast, and it makes you laugh out loud—that’s going to be a major event in artificial intelligence.
As far as actually getting on stage and such, that’ll be the true test, but I don’t want to conflate the comedy portion with the robotics and uncanny valley portion of the problem.
Let’s say—the content itself, delivered as audio.
If we can give a topic to an AI, and it can write a routine similar to:
- Chapelle
- Gervais
- Bartzgate
- Louis CK
- Norman
…that would convince me that it can do one of the most creatively difficult tasks that humans are capable of, i.e., making people laugh.
If it can hit that regularly, like 1/10 attempts for a given topic, that would be extraordinary.
I’m already having it write jokes in the voices of those comics above, but it takes a lot of work on the prompting side.
The true test will be just giving a topic and having it do all the work.