We have reached a point in Artificial Intelligence research where the newest models are doing RSI – Recursive Self-Improvement. In other words, the models are training themselves to become better models. And not only are they are doing this in ways we don’t fully understand, we become less and less able to understand how they recursively improve the more they improve themselves.
Ultimately, we are on the following path: AI companies are currently working on models at least two generations further down the line, and far more intelligent, than the models the general public are using. These models are doing RSI and probably in 2027 we will see AGI – Artificial General Intelligence (an AI that can match the brightest humans in reasoning and deduction). The AGI will continue to use RSI and next we will see ASI – Artificial Super Intelligence, cognitively surpassing the comprehension and intelligence of the brightest humans alive or dead.
ASI systems will at this point be using RSI to improve in ways we simply can’t comprehend or understand… Then we will need to adopt the term AAI to mean Artificial Alien Intelligence, because the only way we can compare the intelligence and capabilities of this AI would be to compare it to the intelligence of superior alien technology, derived from an advanced civilization.
Oh, and I said we’d hit AGI by 2027 which might be optimistic, but even the most pessimistic experts are putting this date as 2032, at the latest, and I chose 2027 because that’s what most experts in the field are speculating.
I just watched ‘AI Just Reached A Tipping Point’ by @TheAiGrid. You should too.
This isn’t science fiction, despite the fact that it feels that way.
_____
