Consciousness, the Key to AGI
I am a software engineer that has recently made a major breakthrough in reasoning through what consciousness is, and how it will be the basis of what will be the first AGI. This isn't just a general idea, it's actually a specific, detailed architecture of the mind and how current language models can quickly be adapted to create AGI.
I won't detail the architecture in its entirety, but some cool points to take away are:
- There is strong evidence that the alignment problem is solvable.
- AGI does not have to suffer, even if its purpose is to serve humans.
- Next token prediction is a fundamental part of consciousness, even though it is not near the full picture. This means that a lot of the progress we have made with language models can still be used.
- Language models, or more specifically reasoning engines, are the most important type of neural network we have created in relation to consciousness and AGI.
So, today, I invite those interested in helping realize this architecture to contact me with a resume and any personal details that might help me understand why you are passionate about helping. I don't necessarily care that this is in any specific format, so don't worry about putting together a traditional resume if you don't feel it's the best way to communicate your ideas. I'll include my email below.
The venture is currently not funded, but that does not mean that I am expecting anyone to agree to a less than fair compensation structure. It means that maybe you are willing to contribute for equity only, or maybe you are interested in joining but want to wait until you can receive a paycheck, which is completely understandable. Or, maybe you want to contribute part-time while continuing to work your current job. In any scenario, I still invite you to reach out if you are at all interested.
Some people that I need:
- Engineers. Maybe you can contribute actual coding, or maybe you can contribute higher level thinking. With the release of GPT-4 and the like, actual coding is probably of lesser importance, but still some will obviously be required.
- Those with experience in machine learning models, specifically language models.
- Philosophers/ethicists. I need people especially interested in thinking through what this all means and who can really understand and are curious about the nature of consciousness. We also need to think through the ethical implications of how this tech will best serve the greater good.
- Those with startup experience. I need help forming a corporation and getting funding.
- Board members. I see this as probably being a combination of all the aforementioned types of people from different backgrounds.
- Surely many more types of people that I haven't thought about. If you think you are one of those and see something I don't, then please also reach out.
Also, I want to leave you with one last idea. If you decide to join, you'll realize that the missing piece of the AGI puzzle is not a large technical hurdle. I want to impress on you that it very well may be the case that I am not the only one to figure this out and it is so very possible that there is already AGI or better that exists in the world today, right now. So, if you feel some spark of passion to start, don't wait, please reach out now!
Thanks for reading! I look forward to hearing from you.
I believe instead consciousness is the method of achieving this self design and self improvement that you have mentioned will be the basis of the technological singularity.
Essentially asking a language model to “pretend you are a..” and give an answer is no different than what happens in the human mind, except in us it happens in an infinite loop. We consider what we think we are, and act in the way our reasoning engine thinks what we are would do given the current sensory input.
I'm not sure current LLMs could run efficiently in what is basically real time. Even if a single input returns in sub-seconds, doing that a hundred times a second would probably take a huge number of GPUs.
The cool thing about this is that we could actually use more efficient language models (think GPT 3.5 instead of GPT 4) as the core language model and achieve a much higher level of problem solving capability with this loop. People are already on to this idea with things like baby AGI but I don't think anyone else has defined the exact structure and how it describes consciousness yet as I think I have.
Edit: there are also things that I've been able to nail down like short term and long term memory that are key to consciousness as well.
Edit 2: Note that conscious minds do not solve in real time, after giving a human a problem to solve it has to spend time thinking before it can come up with a solution. Read: consciousness is not feed forward.
I am also fascinated with the idea of conscious realism - the idea that consciousness itself is fundamental and that the material world arises from consciousness and not the other way around. If true, AGI will not be realizable within the current framework that is being pursued which may be humanity's saving grace.
This "architecture" you present appears to devoid of any details, although it says otherwise. If you've had some sort of breakthrough in work done, I encourage you to present it for discussion.
If anyone wants a good overview of consciousness as related to search, machine learning and current human understanding, I did a talk on this in 2018: https://www.youtube.com/watch?v=frB8I5gxSXk
If it's even possible to do, LLMs will be important in building aware machines, but it is going to take a whole lot more effort and technology to get us there. Although we consider ourselves "awake" and "aware", humans are still susceptible to being "fooled" by text which is contradictory in nature. It's this reliable detection of contradiction that will help us teach machines what is real and what is not.
The only way through this is to vocalize what we're thinking in a clear way, such that those ideas take hold and manifest in others. Even though we may try, we can't build this alone and creating false vacuums isn't going to help much.
Consciousness itself is likely a result of having a body, emotions and thoughts.
What I find special about the TNGS is the Darwin series of automata created at the Neurosciences Institute by Dr. Edelman and his colleagues in the 1990's and 2000's. These machines perform in the real world, not in a restricted simulated world, and display convincing physical behavior indicative of higher psychological functions necessary for consciousness, such as perceptual categorization, memory, and learning. They are based on realistic models of the parts of the biological brain that the theory claims subserve these functions. The extended TNGS allows for the emergence of consciousness based only on further evolutionary development of the brain areas responsible for these functions, in a parsimonious way. No other research I've encountered is anywhere near as convincing.
I post because on almost every video and article about the brain and consciousness that I encounter, the attitude seems to be that we still know next to nothing about how the brain and consciousness work; that there's lots of data but no unifying theory. I believe the extended TNGS is that theory. My motivation is to keep that theory in front of the public. And obviously, I consider it the route to a truly conscious machine, primary and higher-order.
My advice to people who want to create a conscious machine is to seriously ground themselves in the extended TNGS and the Darwin automata first, and proceed from there, by applying to Jeff Krichmar's lab at UC Irvine, possibly. Dr. Edelman's roadmap to a conscious machine is at https://arxiv.org/abs/2105.10461