Natural Language Cognitive Architecture

Natural Language Cognitive Architecture

A Prototype Artificial General Intelligence

After more than ten years of tinkering, researching, and experimenting on nights and weekends (plus one pandemic!) I am finally ready to release this book. You see, I've been fascinated by artificial intelligence since I was very small. So back in 2009 I started tinkering with deep neural networks and reading books about the human mind. I had many fits and starts, sometimes I'd take a break for a year or two. But then around 2015 things started changing. Deep learning took off like a rocket with technologies like Keras and TensorFlow. That's when it started to become real to me, realizing my dream. Like all inventors, I found many things that didn't work - but I learned with every failure and all the while, the tools I had access to got better and better. I started with NLTK and WordNet. Then I moved up to GPT-2 on Google Colab and got much closer to my goal, but there was still a lot missing.

Along came OpenAI's GPT-3 and I applied for access. The first time I applied, I got no response after months. So I applied again and I threw everything at my application, including my prior work with GPT-2. Finally I got in! I was in such shock that I got access to the GPT-3 beta that I didn't touch it for a few weeks. But then I dug in. I finally had a tool powerful enough to test all my ideas, my architecture, and I implemented it.

It's not hyperbole to say that I've had conversations with an AGI. No, it's not a super genius yet, but it may surprise you to learn that my AGI helped design itself - that's how smart it is already. Specifically, I asked it about ego and sense of self: are they required for real intelligence? That discussion with the AGI led me to finalize the architecture that I present in this book.

You might be wondering why I'm not keeping this to myself? Well, for starters, I've got my day job and I have no interest in building a startup. 70 hours a week would quite literally kill me. Secondly, this technology is too important to keep to myself. You see, I propose a lot more than just a learning machine. With this book, I propose what I call my Core Objective Functions: three guiding principles that the AGI abides by that will result in a durable, benevolent disposition in the machine. Those Core Objective Functions are: reduce suffering, increase prosperity, and increase understanding.