This post is gonna be really long. I’ve decided I’m going to use these journal entries as notes for myself, and I’m sharing them with everyone else for the reasons mentioned above. If this is a TL/DR (Too Long/Didn’t Read) for you, I won’t mind at all since it has use to me.
* * *
For me, everything stems from research. I had an idea for a bunch of teenagers who try to prevent a global catastrophe. To write it, I had to come up with a single incident that might cause a catastrophe but that teenagers could stop. At the same time, I was toying with the idea of shipping nuclear waste to the sun. The logistical difficulty of getting a rocket to the sun aside, the problem is, what if the rocket exploded in the atmosphere.
We have a term for that: dirty bomb.
Then I thought, that was kind of an interesting idea, but what if instead of nuclear waste, the world decided they wanted visual, verifiable proof that the U.S., Russia, etc., are destroying their nuclear arsenals as promised. What if the U.N. decreed that the missiles must be shot into space and detonated there, for everyone to see? If, one of those rockets were to explode in the atmosphere, it would be catastrophic.
I spoke with a man who, during World War II had a job that got him the title Computer. This was before computers became as common as they are today. After the war, he got a job calculating trajectories of rockets for Government Agency.
He said, if a 100 megaton nuclear warhead exploded in the atmosphere, at a height of 40 miles, it would cause an electro-magnetic pulse that would wipe out most of our low-orbit satellites as well much of the national electric grid. With that one detonation, most of the country would be without electricity for a LONG time.
I later spoke with a rocket scientist who works for Government Contractor making… rockets. I asked him about the idea, mostly about the logistics of getting an intercontinental ballistic missile to go somewhere other than another spot on Earth. The problem is, gravity is strong, really, REALLY strong. It takes a lot of force to escape Earth’s gravity well. To make an ICBM capable of leaving earth would mean putting it on top of a rocket the size and power of a Saturn 5. The Saturn rockets are what the U.S. used for the moon missions. They’re huge. It’s prohibitively expensive.
Thus, the catastrophe project is on semi-permanent hold until I can find a suitable catastrophe.
* * *
I want to write a story about A.I. I know a bit about computers and what’s possible from all the research I did for “ALGORITHM”. Actually, I learned quite a bit about computer security while researching for “ALGORITHM” and that will almost certainly be making an appearance in “Intelligence.”
Other than the computer security stuff, there’s little crossover. I have to basically start from scratch.
The first place I normally start is with my friend Kevin. But, he’s going through a lot of social stress right now so I can’t ask him.
Then, it’s on to Kaiser. He’s busy getting his Masters and working full-time, and volunteering at the TB clinic, and coding 10.4 Fox (the version of Firefox that he writes by himself for PowerPCs). So, Kaiser’s unavailable.
That leaves me alone. That’s not how I like to research, but it’s what I’ve got. I’m more than a little nervous, but I know Kaiser will want to read the script when I’ve got a version for him, so it’s not like it won’t be vetted. Plus, a lot of very good computer professionals liked “ALGORITHM”. I’m sure at least one of them would be happy to help me as a technical consultant too.
For now, though, I’m alone.
If you want to be on the forefront of A.I. there’s really only one place to go. M.I.T. M.I.T. has an AI lab that is where most of the best research is happening. There’s some great stuff happening at CalTech too but there’s another factor that makes me prefer M.I.T. OpenCourseWare. OpenCourseWare are classes that MIT released under the creative commons license. That means they’re free to watch. And they’re on the web, so anyone anywhere can watch them.
It took about 10 minutes of searching but I found their class on AI. I’ve been watching it. I’ve learned some basic terminology and definitions. There are several different ways that you can teach a machine to learn. Rather than reference them all, I’ll just point you to Stanford’s list.
The dominant form of A.I. today is Deep Learning, which is comes from a convolutional neural network. The idea behind Deep Learning is pattern recognition, teaching a computer to learn the way people learn, by showing the computer as many pictures of the things as possible. In this TEDTalk, Fei-Fei Li said she fed her A.I. 1 billion images, which had been categorized and labeled by 50,000 humans.
* * *
One of the most profound realizations I had, ever, on anything, was that human intelligence is a complex form pattern recognition. When I realized that, a lot of different things that hadn’t been making sense fell into place.
In psychology, we measure intelligence through something called an I.Q. test. However, the I.Q. test is severely biased in favor of white american middle-class men. This makes sense since the I.Q. test’s function was to figure out where to put people in the educational system, which itself was designed to prepare people for the workplace.
There was really no need to know about anyone else. Poor people (a slave class of unskilled labor), women (who generally didn’t hold a job), and the ultra-rich (who already had their place as leaders determined, or they didn’t work at all), weren’t part of the equation because they had no place in the skilled workforce, and therefore, no place in the education system.
Obviously those race, social, and sexual prejudices are unconscionably wrong, but our society has advanced to the point where we have access to enough stuff to allow us to be far more fair about these things. Those things are wrong regardless of the degree of abundance, but now we can do something about it.
The solution to the problems in the I.Q. test has been to say that there are different kinds of intelligence. I don’t like that idea. It doesn’t make sense. It feels like a stop-gap, like using duct tape to hold metal together when the metal broke. It should really be a single piece. The same is true, in my mind, with intelligence.
The idea of advanced pattern recognition was profound because it answered all those questions and accounted for the so-called different kinds of intelligence, but under a single theory.
* * *
Deep Learning is teaching a computer pattern recognition. Our development with A.I. has proceeded to the point where we are now attempting to use the way we work as a model for how we should make A.I. work. There are A LOT of really big presumptions in that, but I’ll probably get into those later.
* * *
Deep Learning is a fascinating turn because it allows me to use us as a model too.
While Deep Learning is pattern recognition, it doesn’t answer the question of consciousness, which is how I define Advanced Artificial Intelligence. For me to write a realistic story of Advanced A.I., I need to know what kind of circumstances would allow for the possibility of consciousness to naturally emerge.
* * *
As my exploration to understand myself, how I think, what constitutes reality, etc. progressed, I ran into psychology, to Freud. He came up with an interesting idea of an ego, which is basically what we call our conscious selves. Our subconscious, which he defined vaguely, but which is protégé Carl Jung explored extensively, is vast and mysterious.
As an artist, much of my work is learning to access my subconscious in a far more interactive and deliberate way than is required for most vocations. This means I have to know myself and how I work. I like to write before I go out. I like to write in the mornings, though it’s actually 1:42pm now. In the afternoon, I like to relax and let the next part of the story work itself out in my subconscious.
I always wonder what my subconscious actually is. It’s a part of me, arguably the more dominant part of me and influences, if not completely controls most of my choices, even when I’m trying to be the most rational I can be. Often, my rationale is merely a rationalization for a choice I’ve already made subconsciously. So, knowing how my subconscious works is vital.
* * *
That gets back to pattern recognition. Perhaps the conscious part of my brain processes metadata and the subconscious is the processing of actual data. That way, pattern recognition is possible, and I can mine the metadata and extrapolate what’s actually in the data too.
That’s useful, but I still want to have direct access to the data itself. According to Jung, and to my research as well, this cannot be done directly. Any attempt to directly access the subconscious results in that aspect of the subconscious becoming part of the consciousness, which then makes the subconscious that much more elusive.
Rather, in order to access the subconscious I have to give it room to be itself. I have to allow it to seep into my consciousness, to invade my consciousness, then, very softly, very passively, I can observe the subconscious move and thereby learn more about it.
I’d compare it to a romantic relationship. I can sit and ask my wife Memi a bunch of questions about herself, but that won’t tell me as much as simply watching her as she lives. I’ll learn more about her as I hear her laugh in the other room than I will by having a rational conversation with her.
In the same way, I can’t directly interact with my subconscious, but I have to give it room, and let it do its thing, and then I will see it work and learn, maybe not the exact nature of how it works, but the flavor of its working, what it feels like when it’s working.
Once I’ve established that pattern, what it feels like when my subconscious is working, I can then work to create a similar environment to again give it space to work.
* * *
All of that is really just a tangent, but an important one because, with Deep Learning, we’re building the subconsciousness of a machine. We’re feeding machines torrents of data and then it’s giving us some metadata. But, because the A.I. is able to spit the metadata we want doesn’t meant that it’s understanding the data, or that it’s self-aware. Self-awareness requires consciousness, which gets back to the earlier question: under what circumstances will consciousness naturally occur?
* * *
If I allow for the possibility that consciousness is metadata and that intelligence is really complex pattern recognition, then consciousness is the ability to recognize more complex patters. It’s beyond object recognition, which is mostly what Deep Learning has displayed so far. It’s understanding what that object means in the context of a greater environment.
* * *
This line of questioning leads me to questions of animal consciousness. The reason this intrigues me is because consciousness, the ability to choose exists largely in the frontal lobe. The frontal lobe is associated with reward, attention, short-term memory, tasks, planning, and motivation.
* * *
I just got interrupted by a fan who sent me the link to a Wikipedia entry on Strong A.I. It’s a useful place to for me to continue researching.
* * *
The thing about the frontal lobe is that it’s larger in humans than any other animal. Based on what it does, it’s clear that this is where the brain does most of its higher processing, the more abstract kind of pattern recognition.
I’m going to have to do some research into how a child’s brain develops, in relationship to its capabilities and the capacity and use of the frontal lobe.
Based on the activity of animals, it’s clear most of them don’t have the capacity for more complex forms of pattern recognition, things I like to call abstractions, or what might be called metadata. Ideas like “now” as apposed to “then”; conveying ideas through symbols, be they written, pictorial, or auditory.
That said, there are ground animals that make a sound that indicates danger. Other animals make more complex sounds than indicate the kind of danger, which can be noted by differing reactions: one sound for snake, another for hawk.
* * *
All of these questions and scientific and philosophical meanderings are to answer a single question: under what circumstance will consciousness naturally occur? When was it an evolutionary advantage for a species to have consciousness, the ability to think abstractly? And, by logical extension, under what circumstances will consciousness naturally occur in an A.I.?
I have to answer that because the A.I. needs to emerge on its own and unexpectedly. Otherwise, frightened humans would simply turn it off or fight it. The A.I. needs to exist and grow to the point where fighting it is no longer an option available to us.
* * *
That leads to me a very interesting problem, one that is almost keeping me up at night. I can’t figure out the ending. I’ve tried writing stories without first having the ending and they tend to meander a lot, which means A LOT of editing. I’d like to avoid that if I can.
That leaves me with the question of what a world with a computer far superior to our own minds might be. And, would that world be a place people would want to watch a movie about?
I know I hate most forms of authority. That might be because I know I’m smarter than most people in authority, which forces another question: is the hatred for authority, or simply authority that is less than I am, and if so, does that hatred have a point where it transitions into respect?