Building AGI - A Society Approach
From Marvin Minsky's "The Society of Mind"
Right, with all the hype around LLMs these days, I would like to visit a topic “Artificial General Intelligence” (AGI). Discussing about what defines it will take a few issues, thus I have actually move my macro research topic to be on “Intelligence” instead.
But I do want to share my initial thoughts after having gone through Prof Marvin Minsky’s “The Society of Mind”. Prof Marvin Minsky is a renowned computer and cognition scientist whom I have great respect for. (And another note: I prefer Prof Marvin Minsky over Prof Geoffrey Hinton). On to our discussion!
Discussion
At the time of writing, I am about 5 chapters into the above-mentioned book and was quite marveled by it. How so? If you look through just the content page alone, you know that Prof Minsky has put in a lot of thoughts on the breakdown of tasks, and agencies, followed by what are the kind of tasks that the brain performs such as learning, reasoning, framing, senses, etc.
The first few chapters he discuss his theories through observations, a very scientific approach which I like. Followed by there are hierarchies in tasks, and we should look at the tasks breakdown (I termed it elemental tasks), alluding that there are agents in our minds that come together and perform the different tasks that the individual wants to carry out.
In order for the individual to survive and thrive, these agents live in a society and have to come together to perform the macro-tasks, like eating cereals, or going to the toilet all the way to giving a speech or conducting a lesson.
Prof Minsky calls it a “Society” and can bring with it a lot more perspectives on how we can make agentic workflow work for humanity and individuals. This is because it is like humanity, where in order for humanity to survive the different challenges and thrive, relevant individuals (agents) are called upon based on the skills and knowledge that they have received to come together. The “Society” approach will make it easier to understand how to make agents come together, work in harmony and precise coordination. Society of agents in the brain!
I strongly recommend Prof Minsky’s book, at least based on the first chapters I have read. I think it will give a lot of insights into the overhyped agentic workflow that vendors are selling. Just to be clear, I still believe in building agentic workflow, but do not buy the overhyped value that it can bring right now.
For my initial thoughts on how to deploy agentic workflow successfully, do have a look at my last October issue of “Successful Agentic Workflow”.
Your feedback and comments are much appreciated! Consider liking and sharing this issue with your community! Donations? Buy me a coffee!
