I’ve been reading Daniel Dennett’s Consciousness Explained and specifically looking at how his theory would be applied to creating artificial consciousness in a robot or software program.  Dennett makes a few key points that would have to be taken into account in developing artificial conscsciousness:

1. There is no central meaner, no executive decision maker.  Instead the brain is massively parallel, ideas come and go, actions and words are bubbling up everywhere and the best ones that fit best are the ones that get said or acted on.
2. Consciousness is effectively software running on the brains hardware.  The consciousness is a serial virtual machine running on the parallell architecture of the brain.
3. All of the parts of the brain are multifunction.  And different part’s functions overlap.  As Dennett says: “human engineers … design systems in which each element plays a single role, carefully insulated from interference from outside, in order to minimize the devestation of unforeseen side effects.”  He goes on to say the brain was designed by a process that thrives on “multiple superimposed functionality, something systematically difficult to discern from reverse engineering.”
4. The brain’s memory is not direct access memory (RAM) like a computer’s memory.  Instead things are brought back up and remembered by association and free thinking.  In addition, there is no definite boundary between program and memory as in a computer.
5. Many current researchers are working on modelling and researching the parts of the brain on the periphery, the parts that receive inputs from the vision or hearing.  But by working from the outside in, these researchers leave a lot to be explained by the eventual central point, where a lot of the consciosness may be in these peripheral parts of the brain.
6. The development of the brain and of life is very dependent on chaos.  Evolution depends on random variation in genes for it to select from, the brain was developed out of chaotic processes that hit upon chance processes and “serendipitous side effects.”

These are just a few of the ideas that I think need to be taken into account when drawing up a plan for artificial consciousness.

Since this is related to a lot of stuff I’ve posted about on here, I figured I’d post the link.  It’s an article about spaceships being developed with artificial intelligence to remove the long delays of getting commands from people back on earth:

NYT: Intelligent Beings in Space!

The human brain is the most intelligent thing that we know of. One of the best prospects for developing true artificial intelligence is to reverse-engineer the brain. Reverse-engineering the brain will involve imaging the brain down to every detail, modelling it, and simulating it. There is already a lot of research going on to reverse-engineer the brain from fields like neuroscience, computer science, engineering, and pyschology. We are not able to image the brain at the neuronal level of detail yet, but broad simulations of different brain regions have already been created. Understanding the brain is something scientists have worked at for a long time and we will soon be approaching a time when we can not only understand it, but replicate it.

The biggest current restriction on reverse-engineering the brain is our current imaging technology. Both the spatial and temporal resolution of our best imaging now (MRI) are not good enough to capture individual firings of individual neurons. Still, we are able to get general ideas of which groups of neurons are firing when and what they do, so that much progress has been made in simulating various brain regions. But as our imaging systems get better then we should be able to capture and record every neuronal firing in the brain and develop a complete model of its workings.

The brain is much different from a computer and building a full model of it will provide us with many benefits. The brain’s neurons fire very slowly compared to a computer, but they are huge numbers of neurons and they are all connected in a massively parallel way, providing us with great pattern recognition abilites. The brain is able to rewire itself and grow new neuronal connections to learn new skills and memories. By combining these features with the great speed and perfect memory of a computer, we will develop an incredible artificial intelligence.

There are a number of current projects going on to reverse-engineer parts of the brain. Many of the pattern recognition algorithms used in the field of artificial intelligence and data mining such as neural networks and Bayesian networks were developed from theories of how the brain worked. Researchers have developed detailed models of the cerebellum and parts of the visual cortex. Lloyd Watts at MIT has developed a model of the auditory pathway in the brain that was better at differentiating a voice from a crowd than any program written before.

Current people are working on reverse-engineering the brain from the sensory inputs inward, since the senses are something we can understand easily. As we develop accruate models of the pathways the sensory data follow, we will develop models further in the brain, where the information is combined and stored and decisions are made. Along with the rapidly improving imaging technology, a complete model of the brain will soon become reality.

A Turing machine is Alan Turing’s original model of a basic computer. A Turing machine can model any computational process. In the 1930’s, Kurt Godel proved that there were certain mathematical problems that could not be proved. Later that decade Alan Turing and Alonzo Church both published papers suggesting that certain problems were unsolvable by a Turing machine (the Church-Turing thesis). Since a Turing machine can model any computer, these problems are unsolvable by any computer. It can also be interpreted that since a Turing machine cannot solve such problems, neither can a human. Since the human brain and body follow the laws of physics and can eventually be modelled by a computer or Turing machine, the same rule must apply to them. So the question is, are there certain problems that humans cannot solve?

“Civilization advances by extending the number of important operations which we can perform without thinking about them.”

~Alfred North Whitehead, 1911

This is true is so many fields. In computer science, more and more simple functions are written into libraries and can be used without thinking. First you had to write code to print something, now you just call a print function. In all research, you build on the previous research in the field that has proved/disproved various aspects of your research. And in society in general, we continue to automate jobs: from the industrial revolution automating things to the technological revolution automating more intelligent processes. And as these various jobs and processes get automated, it gives us more time to work on more important and higher level matters and “advance civilization”.

I’ve been reading the book “The Singularity is Near” by Ray Kurzweil, which Holly and Rick gave me for Christmas. It’s all about the accelerating rate of technology and what will happen when that rate approaches infinity (the singularity). This is something I discussed in one of my old rants from 2003 (Part 1, Part 2). I speculated that this infinite rate could be reached when we can manipulate and travel in time. Since an actual rate of infinity seems impossible, would we be able to differentiate between close to infinity and really really close to infinity? Would our perception of time change as technologies are developed at such an absurd rate?

In the book, Kurzweil suggests that the singularity will occur when when we develop powerful enough computers with good enough AI that they can design newer better computers. These intelligent computers would redesign themselves, continually designing better and better computers and other technology. With the computer power available increasing exponentially and the amount of money we spend on computers each year also increasing exponentially, we’ll be buying enough computing power to match all of life on Earth by the 2040’s, so this is when Kurzweil suggests the singularity will occur. We will all be augmented by computers in our brains and also have rapidly accelerating intelligence to go along with the artificially intelligent computers.

Here’s a quote from the book (p. 22):

“Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an “intelligence explosion,” and the intelligence of man would be left far behind. Thus the first ultraintelligent machine is the last invention that man need ever make.”

~Irving John Good, “Speculations Concerning the First Ultraintelligent Machine,” 1965.

South Korea plans to get robots into every home, and even patrolling the streets, in the next few years:

http://www.nytimes.com/2006/04/02/world/asia/02robot.html

I was thinking yesterday about the difference between the way things are computed/processed in digital electronics and the way they work in the brain. What are the differences? What are the advantages of each? How would a computer work with neural networks and how would a brain work with digital circuits?

Digital Logic:

In a computer, everything is built out of transistors. All signals are discrete, either on or off, 0 or 1. Latches and flops are built out of transistors, and nand, nor, and not gates are built out of transistors. From these come all the processing in a computer. Each of these gates takes some number of inputs, which are all defined as either 0 or 1. Then depending on the type of gate, it has some defined rules, and there will be an output of either 0 or 1. Everything in digital logic is deterministic: if you know what the inputs are, you can always tell what the output will be.

Neural networks:

In a neural network, everything is arranged from neurons rather than transistors. The neurons connect to each other through synapses. One neuron can have many others connected to it. Each of these neurons may fire and send a signal through their synapses to the next neuron. If one neuron has enough neurons sending signals to it then it will eventually reach a point where it will fire and send a signal through its synapses. It could fire if one neuron sends a particularly strong signal to it, or if many neurons together send small signals that add up to a significant one. Basically it sums the incoming signals over time and space and if they cross some threshold, this neuron will be activated. The thresholds change and adapt depending on how often each synapse fires. As the firings decrease, the neuron becomes more sensitive to it, and as it fires more the neuron becomes less sensitive to that synapse. Neurons can also grow new connections to other neurons. The neural network is non-deterministic, the same inputs may not always result in the same outputs because the thresholds at different synapses may have changed or completely new neuronal connections may have been created.

Differences/Benefits:

One main difference is that neural networks can adapt and change. Another difference is that neural networks have more values than just on and off. A neuron could be not firing at all, it could be firing a little, it could be firing a lot, or somewhere in between. A neuron could fire so often that other neurons become insensitive to its input. Instead of a simple OR gate in digital logic, where the gate turns on when either input is on, neurons would be less discrete. The neuron may turn on some if either input firing a lot, or if both are firing some, and the neuron may fire a lot if both inputs are firing a lot.

How could a computer be designed using neural networks?

Well, to start, it wouldnt be too hard to design a chip that could act like a neuron, summing inputs over time and measuring them against some changing threshold. I’m not sure how new neural connections would work since these connections would seemingly be hard wired. But all the logic used for computer processing would have to be changed since it is all set up for digital design. Instead of AND/OR gates you might have neurons that require a strong input to activate and neurons that require a small input to activate. And how their adaptations would affect the computing, I have no idea.

How would a brain work with digital circuitry?

Any “brain” that uses digital circuits could not really be considered a brain. There would be no adaptation or change in the brain and so it doesn’t seem much like a brain to me.

This brings up another good question: what is it that gives us our humanity? our consienceness and intelligence? Is it the fact that the brain adapts and changes and develops? Or is it something inherent in the circuitry of the brain? What separates our brains from monkeys? What separates our brains from computers? What makes us who we are?

I would love to be able to design a computer program that was smart and actually had some intelligence. I have designed the AI for my 3D connect four program and I’ve worked on a chatbot. The real sign of intelligence would be if it could learn. In fact, I have thought about making my connect four program be able to learn from its losses. The end result of all this AI work would be to simulate the human mind. There would be 2 ways of doing this. The first would be to understand the thought process of everyone and write some code to simulate it. This, of course, is impossible. The second way would be to somehow analyze the physics and chemistry of the brain and try to replicate the actual electronic signals or cells and see what happens when we connect it all together. I doubt that we can ever really replicate the brain as a computer; even though computing power keeps increasing exponentially, I don’t think it will reach the computing power of the human brain at any point.