In our last conversation with Charles Simon, we spoke about artificial general intelligence (AGI), how far it has progressed in the industry and Brain Simulator II, an open-source software platform to prove how AGI will emerge.
We thank Charles Simon from FutureAI for taking part in this follow-up interview to talk about the progress Brain Simulator II has made and more importantly his new companion book, “Brain Simulator II: The Guide for Creating Artificial General Intelligence.” We explore the book further and try to understand how one can learn to apply AGI to various scenarios.
Brain Simulator II and “Sallie” – Talking about the progress so far
How does Brain Simulator II differentiate itself from other AI software in the market?
First, the Neuron Engine in the Brain Simulator II is a spiking neural simulator. That means that it is a biologically plausible emulation of the brain. Basic experimentation with biologically plausible neuron simulation shows that mainstream Neural Network software is impossible with neurons because there is no way to set or read back the weights of synapses with any degree of precision. So instead of storing any information in the weights of synapses, information must exist in a more binary fashion in the existence of synapses. Then the weight indicates the importance of a particular bit of knowledge, the more you’ve thought about something, the stronger the memory will become.
Second, the Brain Simulator introduces “Modules” which can perform any custom computation on any cluster of neurons. This allows for computational shortcuts for:
- easy experimentation into new algorithms.
- higher performance in areas where the CPU can be significantly more efficient than neurons.
- solving problems we don’t know how to solve in neurons, like how your binocular vision can create depth perception which is straightforward with trigonometry but we don’t know how the brain does it.
Since our last conversation, how has “Sallie” progressed in its journey towards AGI?
One of the most interesting new features is the ability to recognize objects independent of their apparent scale, position, or orientation. In order to do this, visual information is represented by storing the physical relationships between visual components. Consider recognizing a “B” regardless of its size, position, or orientation. The way to do this is to know that there are multiple segments, in various relative positions, and relative orientations, some touching at ends, some straight, some curved. By storing the information only in terms of the relationships between various objects, Sallie is getting a recognition system that is more like a three-year-old’s.
Talking about the new companion book
Tell us about your new companion book “Brain Simulator II: The Guide for Creating Artificial General Intelligence”.
The Brain Simulator II software is coming of age with its V1.0 release and it is now robust enough to need a full explanation which not only documents the program but explains the background and philosophy behind it. The unique functionality in the approach merges a spiking neural simulator with modules for higher-level functionality, but there is also a general approach to experimentation that will lead to AGI.
Using the above book what applications can one hope to build an AGI system for?
AGI differs from AI in that it will emulate “understanding.” This means that AGIs will be able to create explanations for the input it receives and the actions it takes. Everything an AGI knows will be in the context of everything else it knows. One of the first applications we’ll see is truly intelligent speech recognition. Such a system will intuit words from the context of other words and will be able to ask for clarification on things that it misheard, just as a person would. This will have great ramifications for personal assistants and improved web searching, for example.
In your book, you talk about the Universal Knowledge Store (UKS) as the backbone of AGI. At what rate would we have to build up the UKS to achieve near AGI?
The UKS can store and relate any kind of information. Further enlarging it on existing hardware will meet the needs of AGI, at least initially. I see the key breakthrough for AGI is in creating relationships. Any three-year-old can infer that some things are other things in the way red is a color and blue is a color and therefore if you ask about colors, she can respond with the examples red and blue.
The knowledge in your mind consists of these relationships, that some things contain other things (or conversely are contained by), that things are nearer, bigger, louder, heavier, etc. The three-year-old learns these abstract relationships by interaction with the environment. She even learns that a specific word means that thing or relationship and, conversely, that that thing can be described by these words. ”Means” is just another relationship.
An AGI which can build its own relationships will appear intelligent even early on when the content of its brain is much smaller than a human’s.
The next steps for AGI and Brain Simulator II
Do you think quantum computers would accelerate the growth of reaching AGI?
It’s a hot topic, but I don’t see AGI as relying on solving the types of problems quantum computers are good at. Once the AGI ability to learn relationships exists, quantum computers may provide a performance boost.
What are the next steps with Brain Simulator II and FutureAI?
After the dramatic performance and UI improvements for the V1.0 release, we will be working specifically on the key AGI problem of learning relationships. This will involve experimentation with larger and more diverse datasets. Currently, all the input to an experimental AGI is through simulators but this will need to be replaced with robotics which can sense and act in a real environment.
The Brain Simulator development community is expanding, and I encourage anyone to try out the problem and join in with this open-source project. AI novices can learn a lot about neurons and their capabilities and limitations, while AI professionals and developers can load the source code and explore new approaches to AGI.
All the software is available at: http://brainsim.org
I am the Co-Founder & CTO of Xaltius, a Singapore based Data Science and AI startup, and a machine learning enthusiast who loves to interact with people and learn more about how artificial intelligence is shaping the lives of organizations and people and how it is being used to optimize business operations today.
Are you interested in sharing information and knowledge about AI and connecting with some of the brightest minds in the field?
Learn about joining as an Associate Editor.