Known as the "Father of Doom"John CarmackAfter resigning as Chief Technology Development Advisor for Meta Virtual Reality at the end of last year, he focused on the artificial intelligence technology company he founded in 2019.Keen Technologies, and aims to create a general artificial intelligence system, hoping to create an artificial intelligence system that can interact with people more naturally.AnnounceWorking with Richard Sutton, known as the "father of reinforcement learning," he plans to demonstrate the feasibility of developing general artificial intelligence technology by 2030.
John Carmack and Richard Sutton announced the news at an event held by Amii, the Machine Intelligence Institute at the University of Alberta where Richard Sutton works. Richard Sutton will join Keen Technologies, an artificial intelligence technology company founded by John Carmack, while maintaining his faculty position at the University of Alberta.
At the same time, both parties stated that although Keen Technologies currently has only four members, it is much smaller than large artificial intelligence technology companies with hundreds or thousands of employees, and even the current financing amount is only US$4 million, which is incomparable to large artificial intelligence technology companies with more than billions of funds.
But in John Carmack's view, artificial intelligence is not only composed of thousands of lines of code, but can even continue to grow through self-learning, and transform from the current single artificial intelligence that can only correspond to calculations in a specific field to a development model that can correspond to interactions in different fields.
If the current Chrome browser created by Google consists of about 2000 to 3000 million lines of code, John Carmack said that an ideal general-purpose artificial intelligence system would only need tens of thousands of lines of code at most to operate.
John Carmack said that the human brain's capacity is actually not very high, but it can remember and conceive a lot of things. The reason is that the brain has about 860 billion neurons, and each can form more than 100 trillion groups of connections. These neurons can be regarded as parameters in artificial intelligence systems, so in fact, not too much capacity is needed to produce "thinking."
According to John Carmack, compared to the past artificial intelligence model that ultimately came up with the most appropriate answer through large-scale data comparison and analysis, future artificial intelligence will be able to use less code and automatically generate content, thereby simulating the human brain's ability to "think," "remember," and even "draw inferences from one instance."
