Artificial Super-Intelligence is an Illusion
Whether it’s the James Cameron Skynet doomsday scenario, popularized by the Terminator movies where a super-intelligence becomes self-aware and exterminates humanity, or the Ray Kurzweil singularity where humanity’s rate of progress curves closer and closer to vertical, artificial super-intelligence will be a long, long time coming. Skynet was capable of designing and assembling war machines on it’s own but today’s state-of-the-art artificial agents like Siri are only capable of mimicking human behavior and are nowhere near curiosity or invention. It has to do with relativity and the physical constraints of the universe, and how information is derived from it. You can go on this thought-journey with me and see for yourself.
General Purpose A.I.?
Researchers around the world are working toward a theorized “general-purpose A.I.” algorithm that contains the logic that the human brain needs in order to form memories and ideas. In practical terms it would be a program or a script that runs away from you and returns later as the Borg to assimilate you. That’s what we’re talking about, right? People like Steven Hawking and Elon Musk have weighed in to say we need to be cautious for fear of this scary scenario. Would A.I. have the same survival instincts that we have?
To each of us our thoughts seem to arise spontaneously, out of nowhere, but in fact it takes time, space, energy and a multitude of previous discoveries in order for your mind to compute things. All of the thoughts going through your head are a product of stimuli from the environment around you, electrical signals that start in your eyes, or what you ate for breakfast for example. Thoughts consume energy; you need to eat plants and animals, burning the calories in exchange for thoughts. They reside amongst molecules that occupy the volume of your skull in the form of connections between neurons.
This biological computer that propels humanity forward has had millions of years to evolve. We understand the general structure of it and we can re-create a similar system inside a digital computer using neural net algorithms, but these haven’t really delivered the goods that we’ve been searching for. They do interesting computations but it’s not the general purpose algorithm we imagine. Still no sign of that.
Does it even make sense to refer to human thought as general purpose? We do have a purpose after all which is to thrive and survive.
We Live to Experiment
Life and the universe have this interesting relationship to each other. With the universe we get this thing called Time that seems to go on and on. It is fundamental to everything we do. Organic life is a pattern of activity that emerges out of time and ends up surrounding us everywhere you look.
We don’t know yet how life began out but it spreads around, changing and adapting to everything in the universe that it comes across, testing and probing things trying to understand it. Life doesn’t just understand the universe, it puts it to good use too. Order comes out of chaos in the form of human intelligence but there is also a such thing as plant intelligence. Plants know how to harvest the energy of the sun and combine it with the resources on earth to form shoots and leaves.
What if we found life on Mars? Would it look like it came from Mars? I think so. I think it would know as much about Mars as we know about Earth, and it would be outright impossible for this life form to know, with any bit of certainty, some of the things that we can observe and learn about here on earth rather easily. This is a pattern that you see throughout the universe: the further things are the less they know about each other.
Now, suppose you developed an A.I. with all the capabilities that a human has including an equal amount of brain, and you dropped it on Mars where it reproduced and thrived, becoming a planet-wide civilization. How long do you think it would take before this artificial civilization approached the technological capabilities of Earth? Would this artificial civilization have the capacity to enhance it’s very own genes or code in order to make itself smarter? Eventually it would, but you can expect it to take the same amount of time that humans have taken to get here. Or is there something inherently better about artificial computing that gives it a better understanding of the universe around it?
In the last 10,000 years or so there have been a series of events that changed the rate of progress that we were experiencing as a civilization, such as the advent of spoken language, writing, mathematics, computing and most recently the internet. These events all formed hockey sticks in the graph of human progress, which is what A.I. is expected to do as well (maybe even more). The artificial civilization would go through all of these same iterations, no?
But on the artificial planet what comes after the internet? Is it biological intelligence? In this case there is no physical difference between the A.I. and the computers they use to communicate. They are both made of the same type of switches, it’s the function that differs. The distinction between the internet and the next leap may not be that clear, either. It’s another form of or enhancement of communication, like the internet was.
What the Internet did to us
By laying down fiber-optic cable and connecting all of our computers together, and putting a mobile terminal in everybody’s pocket, we have created the foundation for a lot of new things. Thinking about it as a nervous system for the world makes a lot of sense to me, but it’s a bit more than that. It also has it’s own inherent knowledge to it. Due to a combination of factors, the internet decides what news you hear about, or what’s trending or whether that stock price rises or falls.
Now the little bits and pieces of knowledge that are stored in one person’s brain are accessible to another person on the other side of the world with the aid of a search engine. The telephone system and radio communications existed before the internet which allowed low-bandwidth communications worldwide, but it was still up to the individual to initiate a phone call or transmission, and they had to go and look it up in a book or something, so this did not have the huge effect that the internet had. The biggest difference is the computing layer that exists in between the internet communication layer and your brain. There is actually no need at all for any magical computing algorithm at this point, and chances are it doesn’t exist. Even if it did exist, how would it discover things? The experiments still need to happen. The atomic clocks and space shuttles still need to be invented.
Computers can decide which events you and other people hear about. This all by itself is much more powerful than any algorithm will ever be. Billions of individual computers all chunking away at the world around them, trying to discover something, with a big computational overseer spreading as much information around in the most optimal way it can think of. It would be a very unique machine designed for a specific purpose, to propel humanity forward.
Experimenting with the Physical World
I want to touch on how knowledge is created and the form it takes, and for that I’ll start with a question: how does an A.I. know what honey is made of?
I could either tell it or let it figure it out on it’s own through experimentation. If I were to tell it then I am certifying that I did the experiment and that my conclusion is sound and I’m an expert and all that. So either way there is an experiment involved, using a tool to measure the properties of honey, like the Mars rovers does with a laser and a camera.
The thinking (or lack thereof) behind the general purpose A.I. is that there would be some way to simulate the universe to the degree necessary in order to come up with new discoveries. This gets right at the core of what it means to be intelligent. Why don’t we just simulate CERN and the Higgs particle? Why not simulate ITER while we’re at it? Perhaps an artificial intelligence might be better at catching missed opportunities to simulate something and thus form new ideas quicker, but we can build a physics simulator and start making discoveries with it right now, can’t we? No general purpose A.I. required.
If you can tell the A.I. what’s real then why not do that? We could. We could start writing down all the things we know but instead of English they’d be in some sort of source code. It would be huge, trillions of lines of code the size and shape of the universe, stuffed full of random facts and fictions from all different areas including fictional universes and what it means to be entertaining and interesting and funny.
This still wouldn’t be able to predict the Higgs particle any better than Higgs himself could. It would have to do the experimentation, and for that it would need a factory to build machines.
What you ultimately end up with when you set out to build the future is software. Software is what you use to interface with the computer, and the things being built on the internet or mobile devices every day are optimizing the flow of information around the entire world, getting it to the people who need it.
Where does the knowledge rest? Ultimately knowledge can take any form, but the form most useful to us as humans is going to be some sort of audiovisual experience I would imagine. At the moment the best way I know of to write down ideas are words, drawings, diagrams and things like that, some ideas might take a full-length feature film to cover.
The best direction I think for computing to take would be to focus on an individuals effort to convey an idea. By speaking, waving their hands, pointing, drawing and other interactions with the world, a trained user should be able to create anything they want in the computer buffers in the time it takes to think the idea up.
Couldn’t you just build a bigger brain? Suppose you genetically engineered a bigger brain for yourself and started to use it. What would this give you? It probably wouldn’t give you the super-intelligence you’re looking for, only a bigger selection of existing ideas to relate with in your local memory banks. It’s not likely to be any faster than the brain you used to have, in fact it might even be slower due to the extra distance the ideas have to travel inside your head.
When you break it all down life, no matter how simple or complex, from homo sapiens to house flies, all life is order in a universe of chaos. It’s what we do with the super-intelligence that really matters. What we ultimately intended for it was to create our destiny for us, who we are, which planets we live on. But do we really expect a computer to decide all of this for us? In the alternate future where A.I. already exists, humanity will still be the thing that grows and survives.
Oh boy is the future starting to look awesome…
I heard a commenter on a spacecraft design page the other day saying “Guys, guys, you know those aliens with the cool spaceships when you were a kid? We get to be those guys!”
Imagine the rate of progress when all 7.3 billion people on earth doing what they are best at, just sitting and thinking about exactly the right thing. What more could there possibly be to discover than what’s here already? The universes secrets are there, scheduled to be discovered at the specific time and place, like a distant supernova. And If you want to learn about stars then someone first needs to make a discovery using a telescope.