Greg Brockman, the CTO of OpenAI, recently gave this talk on the last 6 years of progress in AI, and made a few brief comments in the talk about AGI (artificial general intelligence):
I wrote a response, which jimmy_d commented on in this thread if you’re interested.
AGI is related (in my opinion) to machine translation, since machine translation is a version of the Turing test. Jimmy_d and I discussed that in this thread, if you’re interested.
The analogy I use between AGI and useful software and robotics applications like voice-to-text transcription and self-driving cars is rockets vs. lighter-than-air aircraft (i.e. hot air balloons, blimps, and rigid airships/zeppelins). If the current AI paradigm — deep learning and reinforcement learning, as we know them today — is fundamentally limited from reaching AGI, just as hot air balloons and zeppelins are fundamentally limited from reaching the Moon, that doesn’t diminish the current AI paradigm.
You can imagine an alternate world where balloons and airships are as ubiquitous and technologically advanced as helicopters and airplanes today. If what you care about is moving around the Earth, it doesn’t matter that lighter-than-air aircraft don’t work outside the atmosphere. Likewise, if what you care about is using AI for revolutionary software and robots applications like self-driving cars, then it doesn’t matter if the current AI paradigm can’t reach AGI.
The current AI paradigm might never reach AGI, but it could still precipitate a new industrial revolution, vastly increase global GDP, and make a lasting positive change to the human condition on Earth. That alone is enough for AI to deserve the attention it gets.
By the same token, even if there is rapid exponential progress in deep learning and reinforcement learning (as we know them), that doesn’t necessarily imply rapid exponential progress toward AGI. No matter how fast or impressive the progress in lighter-than-air aircraft, the difference between the physics of air and the physics of vacuums means it’s impossible that they will ever reach the Moon.
In this analogy, the analogue to rocket engines etc. might be cognitive mechanisms that are discovered while mining the brain for insight on how general intelligence works.
I wrote a response, which jimmy_d commented on in this thread if you’re interested.
AGI is related (in my opinion) to machine translation, since machine translation is a version of the Turing test. Jimmy_d and I discussed that in this thread, if you’re interested.
The analogy I use between AGI and useful software and robotics applications like voice-to-text transcription and self-driving cars is rockets vs. lighter-than-air aircraft (i.e. hot air balloons, blimps, and rigid airships/zeppelins). If the current AI paradigm — deep learning and reinforcement learning, as we know them today — is fundamentally limited from reaching AGI, just as hot air balloons and zeppelins are fundamentally limited from reaching the Moon, that doesn’t diminish the current AI paradigm.
You can imagine an alternate world where balloons and airships are as ubiquitous and technologically advanced as helicopters and airplanes today. If what you care about is moving around the Earth, it doesn’t matter that lighter-than-air aircraft don’t work outside the atmosphere. Likewise, if what you care about is using AI for revolutionary software and robots applications like self-driving cars, then it doesn’t matter if the current AI paradigm can’t reach AGI.
The current AI paradigm might never reach AGI, but it could still precipitate a new industrial revolution, vastly increase global GDP, and make a lasting positive change to the human condition on Earth. That alone is enough for AI to deserve the attention it gets.
By the same token, even if there is rapid exponential progress in deep learning and reinforcement learning (as we know them), that doesn’t necessarily imply rapid exponential progress toward AGI. No matter how fast or impressive the progress in lighter-than-air aircraft, the difference between the physics of air and the physics of vacuums means it’s impossible that they will ever reach the Moon.
In this analogy, the analogue to rocket engines etc. might be cognitive mechanisms that are discovered while mining the brain for insight on how general intelligence works.
Last edited:
