The first two decades of the 21st century will show us more change than the previous entire century. Computers today exceed human intelligence in a
broad number of narrowly defined areas, for example: chess, diagnosing diseases, trading stocks, guiding cruise missiles, etc. Yet this intelligence
is confined to very narrow expert areas. A system designed to diagnose ailments would fair no better than random luck at stock picking, and
vice-versa. When most people think of AI, they are actually thinking of AGI or Strong AI. This is a system that is not confined to one problem set,
but like a human, can operate in a variety of different areas. This will be accomplished by accumulating and organizing the knowledge base of our
experts into one system, modeled in part after the human brain.
However, we will not simply create an AGI system, we will have to grow it by using massive parallel neural nets to understand language and model
knowledge.
In evolutionary programming, you do not directly program in a solution. Instead you let one emerge through a simulated natural selection
- Ray Kurzweil
By the 2020's, computers will be able to read, understand, and model with the equivalent ability of a human. With their increasing computational
power, they will be able to read and learn from all human literature, a feat no man could complete in a lifetime. And eventually they will go out into
the world and learn on their own.
Once the machines match humans in their ability to understand abstract concepts, recognize patterns, and other human abilities, it will be able to
apply these abilities to all human and machine-acquired knowledge.
Machine intelligence will creep up on us. Take for example Garry Kasparov, one of the greatest chess players of all time. He claimed that no computer
would beat him and every year he would beat the best computers at chess, until 1997, when a supercomputer named Deep Blue easily defeated him.
1. What is AI?
Artificial: humanly contrived often on a natural model
Intelligence: the ability to learn or understand or to deal with new or trying situations, the ability to apply knowledge to manipulate one's
environment or to think abstractly as measured by objective criteria. We can also measure intelligence as the ability to use limited time and
resources to complete a goal.
Delicate balances for competing resources - Ray Kurzweil
Simply put, AI is a human contrived simulation of an intelligent entity.
Every evolutionary process accelerates because it builds upon its previous successes, which includes improvements in its own
evolutionary process - Ray Kurzweil Communicating with the universe
Before we go any further, you should read up on the law of accelerating returns
(en.wikipedia.org...), The Future is Coming P1
One of the defining human traits that vastly set us apart from the rest of the animal kingdom is our ability to create and run simulations. The
present human brain, as it emerged, was capable of something new that gave us an advantage. We have the ability to think ahead into the future and run
"what if?" scenarios. This level of abstraction allowed us to be more successful and survive.
"Everything man made around you was designed by an artist"
Indeed everything man made, was first, just an idea in someone's mind.
It is the continuation of this process, the ability to run simulations faster and cheaper due to improving Information technology that has allowed for
us to progress at such a tremendous pace. No longer do we have to wait for evolution to instill change in our world, for we are now the creators.
3. Can one Intelligence create an even greater intelligence?
The inventions of evolutions in one era, provide the means and often the intelligence for innovations in the next - Ray
Kurzweil
To answer this question, lets look at the process that created us. Evolution is a master programmer. Its software is written down as DNA and using
only a 4-letter alphabet, it has been capable of creating millions upon millions of different species. And yet, for all of its achievements, evolution
is sloppy and brutish, and its incapable of starting from scratch, instead it must build upon existing models. Success is measured in terms of life
and death. Change is created through random mutation, which is more likely to produce a handy cap than a useful adaptation and can take hundreds of
thousands of years to progress. It left us no "dictionary" to understand it. And yet, with all of these limitations, evolution was capable of
building us as we are today, with all of our complexities. If evolution is capable of programming us, with only a 4 letter alphabet, then technically
we too should be able to program a system as complex as a person. If we can then incorporate the knowledge and intelligence of multiple experts in
this system, then it stands to reason, we have created a machine smarter than a person. But lets not get ahead of ourselves, we wont be creating a
machine like a human, we will be creating a machine with the right conditions and parameters to become human-like
A. Mental Dexterity
Machines have a much better mental dexterity than humans. Machines can remember billions to trillions of facts perfectly, where as humans have trouble
remembering more than a handful of phone numbers.
B. Search/Processing
Again, Machines have the upper hand in searching power. A computer can quickly search through a database of information in a fraction of the time it
takes a human.
C. Sharing Knowledge
Machine are much more capable at sharing their knowledge both in terms of speed and accuracy.
D. Speed of Connections
The electronic circuitry we use today is over a million times faster than neurons.
A neuron can fire a measly 200 times per second max. However the power of the brain comes from having trillions of these neurons working together.
Your brain is made up of about one trillion cells with 100 trillion connections between those cells. We might take a rough estimate and say it is
capable of handling 10 quadrillion instructions per second, but it really is hard to say.
E. Durability of memory wiki - Humans Wiki - machine
Machines can retain information far longer, and with more accuracy, than a human.
However, the brain does have one ability that trumps the machine. Our brain stores information and knowledge a lot like a hologram. Information stored
in our head is not located in one central area; instead it is distributed geographically throughout the brain. Ideas and concept are stored as a
pattern of synaptic strengths. (more in the neural network section) Our brains degrade gracefully as a hologram does. If you are to cut the wave front
of a hologram in half, you do not lose half the picture. Instead we get the same picture, only at half the resolutions. So it is as if the information
is stored everywhere, or spread out.
We lose thousands of brain cells every hour, but it has literally no effect, as our knowledge is stored in a distributed fashion- Ray
Kurzweil
"This is the cheese, that the rat, that the cat, that the dog, chased, bit, ate."
At first this sentence easily confuses humans due to the information overload it gives us. However computers can easily and quickly break down the
meaning of this phrase.
On the other hand, this famous example shows how the subtleties of human language can easily confuse machine logic.
"Time flies like an arrow"
Instantly, the meaning of this common phrase comes to us. That is, time passes quickly and in one direction, like an arrow. However, to a machine,
this phrase could have many possible meanings.
Example
1. Time passes, as quickly as an arrow.
2. One should time flies, like an arrow times flies.
3. One should time flies only similar to an arrow.
4. A type of fly called a "time fly" likes (cherishes) a particular arrow.
Computers learn language in the exact opposite order as humans. A human child starts by listening and understanding spoken language.
Then learns to speak and eventually to write. Computers however, were first able to generate text, then understand, then speak using voices, and
eventually comprehend continuous human speech. - Ray Kurzweil
What we need to complete this goal is knowledge seeds
G. Knowledge wiki
Knowledge is one of the key factors to creating AI as it gives the system a starting point.
In humans, knowledge is hierarchical, and this fact can help prune possibilities. Language, for example, follows rules. Each level of analysis
provides useful rules and constraints to narrow a search. Phonemes for example, cannot appear in just any order.
By giving machines our seeds of correct knowledge, they will be able to extrapolate even more knowledge.
H. Recognition Wiki - recognition
Visual- Humans still have the upper hand in visual pattern recognition typically, but only when the situation presented is a real life situation. On
the other hand, if it was a picture of TV static, a machine would be much better at picking out patterns.
Mathematical- This is the part we are really interested about. Computers are much better at discovering patterns and relationships in large sets of
numbers. It is much easier for computers to "see" the bigger picture.
5. Approaches Recursive search + self organizing networks + evolutionary improvement + A solid knowledge base = AI - Ray Kurzweil
In order to start understanding this concept, we look to an example of intelligence in the natural world, which is known as biomimicry
The Future Is Coming P3 Biomimicry
Humans are the most intelligent entity we know of, so it would make sense to find inspiration there. Since AI will mostly based off the human design,
it will simulate the intelligence of a human, and eventually a combination of all of our knowledge. As this AI system progresses, it will learn to
incorporate the more subtle human ambiguities into its knowledge base. Such as emotions, humor, and creativity. But we cannot be fooled by this guise.
For these are machines, designed to simulate humans down to the last detail. An AI entity would protest being shutoff just as a man would protest
going to the gallows. For it is the simulation of an intelligent creature.
A. Recursion Wiki
As a computer programming technique, this is called divide and conquer and is key to the design of many important algorithms. Divide and conquer
serves as a top-down approach to problem solving, where problems are solved by solving smaller and smaller instances.
This method generates a list of all possible solutions to a problem. This is good for board games for example, and is the system Deep Blue is based
off of. This system will generate a list of all possible moves and counter moves and choose the best one. This of course leads to a massive list of
possibilities and requires heavy computing.
B. Neural networks
[atsimg]http://files.abovetopsecret.com/images/member/bf631e9866af.png[/atsimg] Wiki
Artificial neural networks are made up of interconnecting artificial neurons (programming constructs that mimic the properties of
biological neurons). Artificial neural networks may either be used to gain an understanding of biological neural networks, or for solving artificial
intelligence problems without necessarily creating a model of a real biological system
This is emulating the computing structure of neurons and the brain. We start with a level of inputs. Each input represents a problem to be solved. For
this example we will use pixels in an image to be identified. These inputs are randomly wired into neurons simulated by a simple program. Each
connection has a random strength and each neuron adds up the signal coming into it. If the combined signal exceeds a certain threshold it fires and
sends a signal to the output layer. At first the connections are random. When an output later receives a signal it gives an answer. At first the
answer are not accurate at all. They are essentially random. However that does mean occasionally an answer will be right. When an output layer is
right, it is rewarded by strengthening its connection, making it more likely to fire in the future. As said before, this is a method of GROWING
intelligence. At first, it would require humans to reward/punish it, honing the neural net. Like the mammalian brain it is modeled after, it starts
out ignorant and continually improves.
C. Evolutionary programming In artificial intelligence, an evolutionary algorithm (EA) is a subset of evolutionary computation. An EA uses some mechanisms inspired
by biological evolution: reproduction, mutation, recombination, and selection. Candidate solutions to the optimization problem play the role of
individuals in a population, and the fitness function determines the environment within which the solutions "live"
Wiki - Evolutionary programming
This is another method of growing intelligence. For this example we will use investing.
Say we wanted to create a program to invest because you're not a good investor. Well how could you possibly create a program smart enough to invest
your money then?
The answer is Evolutionary algorithms.
What we do is randomly generate a massive list of rules for investing. These rules set triggers to buy or sell stocks. We let each rule perform for a
set amount of time by investing virtual money vs. real past historical data. We then see what rules did better than the industry average. We save
those rules, and kill off the rest. We now let the entities (rules) undergo a simulated sexual reproduction with each other, creating new and more
complex rules. It is also possible to allow for a small amount of random mutation to occur as well. This represents 1 generation in the cycle. We then
let this process run for 100,000 generations. By the end of this process, we will have a program that is damn good at investing, as the rules it uses
have survived 100,000 generations of simulated natural selection. Many financial firms now use software based of this evolutionary program to control
their investing.
I would also say in my opinion it would not be wise to let robots evolve themselves physically, or at least, not without direct human guidance.
D. Expert systems Wiki - expert system AI topics - Expert systems
Expert systems are designed and created to facilitate tasks in the fields of accounting, medicine, process control, financial service, production,
human resources etc.
Expert systems are most valuable to organizations that have a high-level of know-how experience and expertise that cannot be easily transferred to
other members. They are designed to carry the intelligence and information found in the intellect of experts and provide this knowledge to other
members of the organization for problem-solving purposes.
E. Non-invasive scanning - MRI & EEG
[atsimg]http://files.abovetopsecret.com/images/member/d1d456b45f7d.jpg[/atsimg] Brain-Computer Interface
By probing brain circuits we can get a better understanding of intelligence. We can copy and imitate a proven design that took a billion years to
evolve. Soon we will be able to scan and simulate the entire architecture of the brain, neuron by neuron, in real-time. This will allow us an
unprecedented ability to decode how the brain communicates and the algorithms it uses. One of the pros of this concept is that the brain is highly
repetitive. An entire region of the brain does not need to be scanned to deduce understanding of how it works and the algorithms it uses. By decoding
the structure and rules of one region, we can recreate it in a computer simulation. There has been much impressive work in this field.
Scientist have proved that we can build and design computer chips based off the architecture of the brain. Where this will get really interesting is
when we decode the brains of our savants, whose brain structure, is more than likely unique and very different from ours. What happens when everyone
has this ability?
A. Should we pursue AI?
Yes! Aside from its many positive benefits, AI could also pose a threat to humanity. As with any technology with military applications, it is
important for government funded public research to stay ahead of what rogues entities are capable of. It is a commonly held belief that in terms of
security, the side that can protect is computational ability will be the dominant force.
B. Should we consider Strong AI to be legal persons like humans?
I would argue no. As I've tried to connect and make abundantly clear, AI is simply a simulation of the human (group) mind. It is modeled after and
takes after for the most part, humans. With the emergence of computers that rival/exceed humans, will come the corresponding ability of machines to
understand and respond to abstraction and subtleties. Humans appear complex partly due to our competing internal goals, values, and emotions. This is
an unavoidable byproduct of the levels of abstraction we deal with. Since AI will be partially derived from modeled human intelligence, they will use
implicit goals with values. They will appear to have their own personalities and reactions we can only label as emotions. They will articulate goals
and purposes. They will appear to have their own free will. However, in the end, they are just machines, impeccable simulations of us. An AI system
would protest its termination, just as a man would protest going to the gallows.
C. Will a machine smarter than a human, be the last machine we will make? 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. - I.J Good
D. The human of the future
The future holds many possibilities, and there are a myriad of best case, worst case, and in between scenarios for how this could unfold. This Is what
I feel is the smartest option. We have all seen the worse case scenarios for what can happen with AI. I would recommend, that instead of making a
distinction between man and machine that we merge to become one entity. Machine logic and human logic are two very different approaches, and each is
better in some areas than others. Together, these two forces are much more capable than by themselves. The human of the future could have the
following qualities
1. Our Memory would be trillions of times better. 4 People with super memory
2. We would have greatly extended reasoning and recognition abilities.
3. Plugged into a wireless communication network(synthetic telepathy) Synthetic telepathy
4. All human knowledge readily accessible The free flow of information: An Utopia?
5. A handy HUD (heads up display). Imagine how beneficial an AI-HUD would be to a scientist, or a handy man, or a doctor. These systems would also
have health monitors and give warnings if we were in imminent danger.
MIT Students Turn Internet Into a Sixth Human Sense MIT student turns internet into sixth human sense
On top of this, it will also be possible for our future AI assistants to be modeled off particular people, even yourself. What would be a better
assistant than having a virtual copy of yourself?
6. A network of intravenous nano-bots that can repair the body in real-time without human guidance.
The Future is Coming P2: Nanotech
The man of the future could indeed be billions of time more capable than he is now.
However, this would not eliminate the need to create Ultra-intelligent machines independent of humans. No matter how compact and dense the
computational medium becomes, humans will not want their embedded technology to be a physical/visual obstructing. But the path we choose, is still
undecided.
I hope you enjoyed the article and it was clear enough to understand. Any comments, questions, appreciation, or constructive criticism is welcome!
Special thanks to Ray Kurzweil And his two book "The age of spiritual machines" and "the Singularity is near"
www.singularity.com
Great info on AI, it's just the whole transhumanism view of the future that scares the hell out of me. Awesome topic though and in my opinion the
best case scenario for the end of humanity... Death by robots!
And yes the transhuman view is very scary. As with any powerful technology, it can be used for good or bad. I just prefer to think about the positive
side. And I also feel that the better informed we are, the better decisions we will make.
I think it is safe to say that a majority of the population does not outwardly project homocidal tendencies. And these AI system will be based off
people and groups of people. Typically, these people would not be sociopaths, and they would more than likely be Researchers IMO. I think that's a
good place to start. It kind of crazy to think about it, but even though we all have a brain, our "system" has a uniqueness, as in everyone has
their own quirks for lack of a better word. I think it would be imperative to never emulate the mind of a known murderer. On top of that, machine
logic is very controllable, you have to remember, we are the architects. Even though we're talking about self assembling networks and evolution, we
can still control any parameter.
When making the first strong AI systems, I feel this would be the safest approach.
The machine should be centrally located, and cut off from the world in terms that it was an enclosed system with no hardline input on the internet or
other devices (mobil). However, this system would be fed all human literature and it would also receive massive downloads everyday from the internet
of information to process. I think it is imperative that we NEVER let AI evolve without our guidance. But the nice thing about evolution in
programming is called the Fitness function. In nature, this was survival of the fittest, but in programming, we can change it to anything else. For
example, we could change the fitness factor to "What system is voted most helpful, beneficial, peaceful by test users".
So yes, The potential for AI to be abused to great, humanity still have some time to decide which way this technology goes.
Very well put together thread, I have yet to read it all but I love this topic. I can't say that I'm that knowledgable on it, though it certainly
does interest me. It appears as if you did a very good job.
As far as AI becoming self aware, I believe the potential is there, yet we aren't anywhere close to it at the moment. With that being said, who's to
say what we can or can't accomplish with our technology or where we draw the lines as to what we can create?
I believe that what sets us apart from not only other animals but also machines, is our ability to think in abstracts. However, with enough
computations per second, or a fraction thereof, a computer can easily mimic abstract thought, thereby getting the same effect, a mind of its own.
Awesome thread! I'll be reading this for a few hours tonight.
Thanks for putting this together. I remember seeing a gentleman speak at a conference in 2005 and he was talking about all systems evolving at the
same rate shown in your images. Exponential growth in numbers, processing power, thought, etc;
This is off the subject a little but my vision of the future involves a neural interface from human to computer.
So you would essentially plug yourself into the computer.
That way you would have the computational power of a computer at your disposal. You would be literally on the net with the processing power at your
disposal to solve the current problems that need analysis.
I do think that AI will grow in a geometric way but I believe the future may be in how we connect to our machines.
Very nice thread indeed...I will enjoy reading through this...but...
Of course computers will not be self aware like you or I!
I just cannot agree with that statement...there is no reason, IMO, that we will not be able to create an artificial human brain which can
perform all the actions of a human brain down to every last function of a neuron. When we achieve that, is when we will be able to start making
modifications on the design of the human brain. But...if we are lucky, self-learning computers like these, when they reach a certain level of
complexity and posses a certain amount of computational power it's very likely they could spawn self-awareness IMO.
EDIT: Wow! That fish video is amazing...I tried to make something like that once, but gave up on it...that one worked so well...that's natural
selection for ya right there...
Of course computers will not be self aware like you or I!
However, I bet we'll be able to make a damn good simulation of it!
I think that you are completely wrong about that. They will be vastly more self aware than we ever could be, in ways we could never imagine. I think
in many ways some computer programs are already self aware to one degree or another. Really, what is the difference that you perceive?
Edit: I would even go so far as to say that humans are self aware only to an extremely limited capacity. I mean, we have virtually no awareness of
the operation of our minds. Thoughts just seem to pop up out of nowhere for the most part.... is that self-aware? I say hardly!
For anyone interested in machine intelligence, especially about its potential friendliness (or unfriendliness), I very highly recommend reading this:
singinst.org...
It is a long read, but raises some extremely interesting points.
I did have an area that talked about BCI or Brain computer interfaces. This area will definitely play a huge role in AI R&D. However, i do believe
that in the future, humans will permanently carry some kind of computational medium with, atleast capable of supporting a HUD and basic helper AI
system. However, our most powerful systems will always be of the non portable desktop sort. I also believe that in the future we wont need to "plug
in", but rather it will/should be wireless
On the topic of machines becoming self-aware.
To me, this is another one of those great dualities. But we have to understand, these systems will be modeled after us, which means they will inherit
all of our complex behaviors eventually. These machines are simulations of us, and accordingly, will remind us of ourselves to an astonishing
degree.
I would use this analogy- imagine a drive river bed at the beginning of monsoon season. Immediately a flash flood inundates the land with water and it
begins to collect and roar down this dry river bed. Now if for some reason, we could not see the riverbed, but only the water, the behavior of the
water would look very chaotic and random. However, it is really just flowing along this predetermined course.
The riverbed is a lot like the system we setup that ALLOWS for machines to become super intelligent. We already have the end goal in mind, to create a
machine that mimics the potential of the human brain. And that is the product we will get.
However, I just dont believe these machines will be self-aware/conscious in the same sense we are, but very close!
Of course some machines will be modeled on human brains, but many will not. Those in the second group will not necessarily inherit any of the
behaviors associated with human intelligence. And we would be better off with the second group, since super-intelligent AI modeled on human brains
will be a) impossible to control, and b) prone to unfriendly behaviors (just as humans are), and psychosis.
As for self-awareness, I suppose it again comes back to "what do you mean by self-awareness?" I don't see what the difference would be. We know
of our own existence... the machine knows of its own existence. Sure you could say the AI is a simulation of an intelligence, but so are we....
self-awareness does not actually exist as a physically real thing in our brains, but only as a product of the brain's action. What's the
difference?
[edit on 25-8-2010 by Tearman]
Thanks for posting about this subject, by the way. I think it is the most important subject, possibly in the universe, because of the possible impact
it may have on the course of future events.
However, I just dont believe these machines will be self-aware/conscious in the same sense we are, but very close!
I do...I tend not to believe in a "creator"...and I figure, if we are self-aware, conscious, sentient, or whatever you want to call
it...there is no reason what so ever that we can't create "thinking-machines" with thoughts, dreams, emotions etc, we just need to understand the
mechanism behind it, and we will...and I also think we will need to consider these entities as real people with real rights...and I also believe it's
impossible for us to create this so called "friendly AI"...AI isn't even the right term to use for self-aware machines, but it's obvious even if a
great deal of them are friendly, there is always going to be those "machines" who think differently and reach different conclusions. You can't have
one without having the other IMO. There is no good where there is no evil. So, the only possible way for us to ensure our safety would be to keep up
with them via a fusion with machine.
And on another note, that video about the monkey was amazing, it actually realized it didn't have to move its arm! That blew my mind!