Why did John Carmack, the developer of 'DOOM' and former chief technology officer of Oculus, advance into AI research?
by
Meta's VR division, which has changed its company name with Metaverse as its main business, was originally a VR company called 'Oculus.' John Carmack, the long-serving chief technology officer (CTO) of Oculus, was the co-founder of id Software , known for developing games such as ``DOOM'' and ``Quake.'' Mr. Carmack is involved in research and development of artificial general intelligence (AGI) at Keen Technologies, a startup he founded after retiring from Meta, and talked about the process in an interview with IT-related news media Dallas Innovates.
Exclusive Q&A: John Carmack's 'Different Path' to Artificial General Intelligence » Dallas Innovates
https://dallasinnovates.com/exclusive-qa-john-carmacks-different-path-to-artificial-general-intelligence/
Mr. Carmack announced in December 2022 that he resigned as CTO due to dissatisfaction with the inefficiency of Meta's business and the insufficient development of Metaverse and software. Masu.
Former Oculus Chief Technology Officer John Carmack resigns from Meta, saying ``I'm tired of fighting'' - GIGAZINE
by Official GDC
Q:
What kind of work are you doing to realize AGI?
Mr. Carmack:
I'm always sitting in front of a computer, coming up with concepts, verbalizing them, building theories, and testing them. That's my job now. No one knows where we are heading. But I think I have a better chance than most for a variety of reasons.
Some people have raised billions of dollars (hundreds of billions of yen) to pursue AGI. The results are interesting in a sense, and machine learning in a narrow sense may be able to present some pretty amazing results right now, but it is not clear whether this is a necessary step to achieve AGI. There are many side routes that can get you valuable results without having to spend hundreds of millions of yen. “Narrow AI” that specializes in one purpose and functions highly still has the power to change the world.
When I get results for the first time, I just say, ``Yes, this is what I spent over 100 billion yen to get,'' and I realize that what we have understood and obtained has the potential to revolutionize various industries. I am concerned about this. When that happens, you won't be able to think further ahead or focus on bigger things that lie deeper. So I try to be really upfront about what I'm doing. There are no short-term business opportunities.
Q:
What made you interested in the topic of AGI in the first place?
Mr. Carmack:
We are currently in the midst of a scientific revolution, but 10 years ago there was no sense that AI was playing an active role. There have been several AI 'winters' like this in recent decades. It's interesting that VR (virtual reality) is following a similar trend. VR was such a huge failure in the 1990s that no one wants to talk about it.
AI has also gone through many cycles of hype, fundraising, industry decline, and failure, and no one wants to talk about AI. But the amazing things that have happened in the machine learning industry over the past decade are truly profound. People who don't realize that the current excitement is different from previous ones are not paying attention enough.
This made me think, ``Okay, let's try working with AI seriously.'' I understood machine learning and AI from a technical standpoint, and even read books related to it when I was a teenager. AI was interesting to me because I knew about symbolic AI, which models the real world using mathematical logic . However, since I was busy working on games, aerospace, and VR, I didn't know what was actually going on in the world of AI.
Picked up a 40 year old textbook on AI at Half Price Books. Pure GOFAI with LISP. Neural networks aren't even in the index. pic.twitter.com/bLWxvk15hq
— John Carmack (@ID_AA_Carmack) June 24, 2017
So, the first thing I thought to myself was, ``I need to sort out the difference between what is being hyped and what is actually happening.'' What I am truly capable of has always come from understanding things fundamentally. At the deepest level, we can only gain insight when we understand how things happen from the very beginning.
I took a week off around 2019. Armed with a computer and reference materials, you spend a week immersing yourself in the basics of the AI industry. Then, I got to the point where I thought, ``With this, I can talk with researchers on an equal footing.'' I was so excited to be able to understand it to that level.
After that, OpenAI CEO Sam Altman invited me to a conference called 'YC120' hosted by venture capital firm Y Combinator . I don't usually participate in events like this because I tend to be a recluse, but I was invited by OpenAI's CTO Greg Brockman and Principal Researcher Ilya Sutskever, so I decided to participate this time.
I'm known as an engineer, but at this time I only had basic knowledge about AI. In addition to the fact that the people who invited me are also driving forces in the AI industry, I also felt that I deserved to reach the same line, which led me to think about the importance of what is happening now and the role I can play. It became.
So we asked Mr. Sutskever to make a list of books we should read. Mr. Sutskever gave me a list of about 40 research papers and said, ``If you really study all of this, you'll understand 90% of what's important in the AI industry today.'' So I read all these papers and organized them in my head.
Q:
At that time, you were still working on VR related work at Meta, right?
Mr. Carmack:
I agree. At the time, I was running into some difficult questions about Meta's larger strategic direction. I'm sure you've seen articles about how much money Meta uses, but I thought most of Meta's projects were completely useless. I was facing some difficulties with Meta when my 5 year contract ended when Oculus was acquired by Meta. At that time, I decided, ``OK, now is the time to get more serious about AGI research.''
Everything we've done so far, whether it's games, aerospace, or VR, has a clear vision, even though we're aiming for things that haven't yet been realized. But AGI is different. This is not a simple engineering problem, because no one knows how to achieve AGI. But given what's happened in the last decade, there are plenty of hints. These are not extreme things like black magic or mathematical magic, but many of them are relatively simple techniques, and many of them are now convinced that they make sense. It will probably take a little more insight to create something comparable to a biological drug .
Around 2020, I made a prediction that there was a 50-50 chance that AI would be born in 2030. This ``birth of AI'' means that ``intelligent and conscious beings on the same level as humans will operate on computers.'' After working hard over the last three years, my prediction remains unchanged. In fact, the possibility of AI being born in 2030 may have risen to around 60%. By 2050, we think the probability is about 95%.
Many people predict that AI will do earth-shaking wonders, right? I'm a down-to-earth person, so I try to refrain from using big words. When I was involved in aerospace, we weren't talking about colonizing Mars, we were talking about which bolts to use to secure parts. So I don't want to go on and on at TED about what can be done with cost-effective AI.
However, the coronavirus outbreak in particular has shown that people can also do much of their communication strictly on computers. Most of the world's value today is generated on computers. It is also clear that narrow AI such as deepfakes, chatbots, and speech synthesis can simulate human modalities . Humanity has not yet realized that AI is a learnable entity working in the same workplace, but it has accumulated an incredible amount of knowledge that will allow it to progress even further.
Some people make a big fuss about the Singularity and AGI changing everything, but in 10 years, we will be able to get work done just by supporting AGI running on the cloud, like computing resources that can be accessed in the cloud. You will be able to. You can also say, ``I want to make a movie or a comic or something, so give me the team I need,'' and the AI running on the cloud will perform the work. Isn't accessing AI from the cloud the most mundane and common use of AI? That's kind of my vision.
Please note that the above is just a portion of the interview, and Mr. Carmack talks a lot about his expectations and prospects for AI, so if you are interested, please read the main text of the interview .
Related Posts: