Believe – The General Purpose AAI Robot Project – Part 1, Overview

I’ve been playing with computers since I was a kid in the late 1960s (it helps when your father worked in aerospace at the time). Amid all of the databases, applications, web sites, utilities, and everything else I’ve worked on, there has always been one constant project going on in the background: artificial sentience (or, as I also call it, Advanced Artificial Intelligence [AAI]).

The science fiction coined the concept that you railroad only when it’s time to railroad. Basically, before you can do something with technology, the technology has to have advanced to the point where you are able to do that something. That is, and to a degree continues to be, a major stumbling block to AAI. However, I think we are at a stage in our technological evolution that we can start seriously thinking about what we are going to do once the foundation technologies finally catch up.

The Brain

So much of AAI necessarily revolves around that combination of hardware and fundamental software that allows a construct to become more than the sum of its sub-processes. Where the machine is able to realize that it is conscious. That it can form and express desires. That it can have the will to effect those desires. That causes and effects are related. And from there perhaps to learn about those things that other sentient beings, biological sentient beings, consider to be important to know.

One of the drawbacks of AAI development has been the fact that we don’t really understand how our only working examples, organic brains, actually function. We have general ideas but nothing that can be completely modeled.

The one thing we do understand is that biology has developed massively parallel processing as its solution to the intelligence problem. Unfortunately for us, we can only simulate such a structure. A human brain has between 50-100 billion (10¹¹) neurons, with upwards of 100 trillion (10¹⁴) neural connections. The best we can do with hardware at the moment is a few thousand parallel processors.

Of course, a big part of the problem is that we don’t understand the problem. We still haven’t created the appropriate family of simulated neurons and their connective structures that would enable us to create nano-scale circuits or NEMS (nano-electromechanical systems) that would allow the billions of neural cells to be manufactured on a reasonable scale.

So, one of the foundations of our “brain” will be the specification of a family of neural analogs.

Once we have these cells, we need to start understanding the processes that form connections and initiate communication. There is some evidence that while some actions are reflexive cause-and-effect due to neural chemistry, there is other evidence that points us to the “magic” world of chaos. While it’s important to create rules for processing signals based on discrete units, we must also work to create an attractor that allows non-linear values to exert an influence as well. We still don’t fully understand how deeply chaos is entwined with biologic systems, but it seems to be an important element.

Once we have the elements of our brain, and the means for them to talk, we then need to organize the brain so that it can attend to various necessary tasks. The single most critical process that must be very stable is the process: Learn. Without that, the rest is just an academic exercise. The brain has to be able to accept new information, process it, discard what is not necessary, store what might not be needed right away, and be able to use this learned information to act on current needs and build on past learnings.

One of the fundamental design problems will be brain size. Since there is likely some minimum size necessary for even the simplest of pseudo-intelligent processes we will have to experiment to find this threshold. It could be hundreds of neurons, hundreds of millions, or even billions. We simply don’t know. We also need to explore the possibility that critical mass is an important factor. No AAI can become sentient unless a necessary level of complexity is first in place.

Somatogenisis

Something that has long strangled development of artificially intelligent (AI) systems, never mind AAI systems, is the need to interact with the environment. Let’s use this as an example: if you take a human brain, divorce it from all sensory input, somehow input data directly into it, would you achieve anything?

Since we are still constrained by the one example that works—biologic systems—it stands to reason that if AAI is to have any chance of success, then it will need to interact with its environment. Cause and effect. Actors and receptors. Body and senses.

We are only now starting to see robots will multi-integrated senses in various “bleeding edge” demonstrations that have gotten out of the lab and into public demonstrations. Some try to mimic biologic processes. Others have been relying on brute-force computer processing. I think that one of the building blocks of AAI systems is the ability to process sensory input. Too often this is considered to be a peripheral add-on when in reality it is an integral part of the entire neural construct. Look at any organic brain and you will see a structure very much dedicated to the processing of sensory information.

Since our plan is a general purpose robot, we’ll need to give it a useful range of senses even from the beginning. For our Mark I model, this should include: vision, hearing, touch, and binary flow. Being a computer, that last sense is a necessary as it is the fundamental I/O sense of all digital computers. Because of the dangers involved, the sense of binary flow would be completely isolated from “the wild”. If it’s necessary to network the computer, there should be no possible physical path, either direct or indirect, that can effect or be effected by our AAI. Safety first.

In concert with the senses, the robot will need to have means of interacting with the world. I’ll discuss locomotion soon, but in this context I’m thinking much more basic: a manipulator system. I’m hesitant to call it an arm or a hand because I don’t want to limit ideas or mechanical complexity. The goal is to allow the robot to have a useful mechanism that it will learn to control so as to effect its world. This would include things such as grasping/attaching, holding, pushing, pulling, lifting, and so forth. The mechanism needs to give kinesthetic feedback.

My hope is that the initial AAI framework will be sufficient that the only instruction that needs to be given is to physically move an arm to initiate inputs. As the AAI gathers the positional information from the manipulator and receives result feedback from its visual, aural, and tactile senses, it should be able to start learning on its own. Just like a child.

Mobility

The last of the basic pieces to worry about is mobility. This might be the area where the most outside-of-the-box thinking can go on. Since this design is meant to be general purpose, it must have the ability to move smoothly and relatively quickly through and over various terrain and obstacles.

My initial thought is abandon wheeled and bipedal methods. Wheeled chassis limit the terrain choices or create complicated kludges to compensate. Bipedal designs, especially modeled after humans, require an amazingly large amount of processing. I’d also suggest avoiding hopping-only mechanisms as these can be clumsy.

Initially, this leaves us with some sort of multi-podal design. My initial thought is a paired biped: two pairs of legs joined with an axle. If we take this a step further, each leg could have “hands” at the ends and be multi-jointed, so that the body itself can adapt to what’s needed: sitting and four-handed, four-legged stabilized walking, or a combination. It might even be modular at the axle to allow additional leg pairs to be added, if necessary.

While all of this seems pretty straightforward, there is one big hitch that might slow us down while we wait for the time to railroad: artificial muscles.

Most robots now use motors and actuators to effect motion. These require a lot of compromises in terms of strength, control, fine-adjustments, and many other related factors. Some materials exist that will bend or straighten in the presence of heat or an electric field, but these are very slow and often very weak.

An artificial muscle (ArtM) will be some sort of MEMS or NEMS device that can be utilized in large numbers in a combination serial/parallel bundle: i.e. a muscle. The key to good locomotion, as well as manipulation, is the development of an energy frugal ArtM solution. Except for specialized applications, our current technologies are laughably inadequate and will impede the project.

Development

How long will it take to do any of this? It’s pretty much open-ended. I’m not a research lab and don’t happen to have the ridiculous amounts of R&D money necessary to develop this on my own. That doesn’t mean that throwing in the towel is our only recourse. By methodically working the problem, solutions will eventually present themselves. These solutions can then be utilized by those with the facilities to bring hypothesis to reality.

What would the purpose of such a robot be? I’d be lying if I said it wouldn’t be cool to have something like the Terminator franchise’s “Uncle Bob” or Cameron, or Star Wars‘ C3P0 or R2D2, or any number of other sentient useful robots. The idealist in me says that it would be cool to have a smart machine that would be able to help around the house, or perhaps explore space. The realist in me knows that these sorts of robots would be purposed by militaries as weapons of war.

The long-view is that when you consider what the inevitable future is for Earth in a couple of billion years, we biologic life forms won’t stand a chance out in the cold creul cosmos. We will have to invest our knowledge and values in our evolutionary successors: robots. If we want to survive, it’s the only sane solution. Who knows how long it will take to develop our “children”? It could take less than a decade or it could take a million years. We have no way of knowing. Better to start now than later.

Figuring that I have nothing much to lose to start work on this, I figured I’d start working on this a little more earnestly than I have. There is much to do and much to learn…not just by me but by those who might find nuggets of inspiration in what plops out. This is a pure-grass roots why-does-it-have-to-be-a-box adventure. There will be stupid things done that might turn out to be brilliant and probably more often just wastes of imagination. But that’s OK. That’s how the process works.

As an example, one of the best of the new-paradigm roboticists is Mark Tilden. Instead of starting from the build-something-and-have-it-learn school, he went the other way and has been exploring robotic evolution via “automatic intelligence”. Basically, instead of starting with people, he’s starting with insects. It’s very interesting work and could find application in this project in numerous ways.

I encourage anyone interested to join in. If enough join, or I write a ton of articles on this, I’ll create a new web site devoted to the project. Think big. Some things might require waiting for the time for railroading to arrive, but when it does, we’ll be ready.