Thread: Too much terminator

Having been thoroughly enjoying the new terminator series I have repeatedly been thinking about a few questions and was hoping I could throw a few ideas around this forum.

Is seed AI the only real precursor to a machine such as described in the terminator series?
If so, how far away are we from obtaining a first seed AI?
From what I understand we are essentially at a brick wall! Maybe someone more informed could comment?

If we were to build an intelligent program that could learn, self optimize and then recompile wouldn’t it still be limited by the amount of available processing power and therefore singularity (‘as it is described’) wouldn’t happen overnight and in fact could take a substantial period of time?

Why is there an assumption that the program/being, would feel threatened by us and in the process become hostile? To me it logically occurs that any being that is significantly more intelligent/enlightened than a human would naturally evolve to a ‘docile’ and ‘non violent’ creature.

As we humans are evolving it seems as though we do tend to become less violent. Look at the Aztecs who lived only 400-500 years ago.

Finally, does it not seem that even if we were able to create such a machine we would never put it in control of the world’s nuclear arsenal? What benefits could be obtained from doing that anyway?

Would love to hear some thoughts.

Actually we were able to simulate half a mouse brain... running at one-tenth the speed mouse brains are capable of... for 10 seconds... adding up to 1 second of half a mouse brain's worth of thought. All this using the most powerful supercomputer on the face of the planet!!!!

We have a long way to go, but hey! in 18 months we should have the ability to simulate a full mouse brain for 20 seconds running at one-fifth the speed of a real mouse according to Moores Law!

Progress!!!

lol, I love the description.

Wouldn’t you consider the Robocup as a good example of what we can do? And surely to achieve those results we require a program that is smarter than a mouse? Maybe I am overestimating it...

Originally Posted by SuperNatendo

Actually we were able to simulate half a mouse brain... running at one-tenth the speed mouse brains are capable of... for 10 seconds... adding up to 1 second of half a mouse brain's worth of thought. All this using the most powerful supercomputer on the face of the planet!!!!

We have a long way to go, but hey! in 18 months we should have the ability to simulate a full mouse brain for 20 seconds running at one-fifth the speed of a real mouse according to Moores Law!

Progress!!!

Don't undersestimate them.

A computer could be created that has the processing power of a terminator but probably it would have to be a quantum computer. so it could do complex logic calculations fast. but still even if a computer has an OS that allows in to act like a Hyman it still has to follow primary decisions. to mack a computer have free will you would need some monster software. and something more intelligent than a Hyman would surely have better things to do than destroy the world.

Interesting idea! Reminds me of the argument concerning the Star Trek transporter. They say we have the theoretical ability to built a transporter, and actually have 'beamed' an atom from one place to another. The only thing stopping us from going further is computing power. Imagine having to store the precise position of every sub-atomic particle in your body, relative to each both in space and time, at a given fraction of a nano-second! Better start networking boys! [/code]

Well, digital computers running software that allows AI might not be the answer...

It seems the new memristors may allow analog computers that will be hardware programmed. These "memristors" allow for more than just a one or zero value and can simulate real neurons. I think this is the future of AI, a sort of analog/digital hybrid computer.

Originally Posted by SuperNatendo

Well, digital computers running software that allows AI might not be the answer...

It seems the new memristors may allow analog computers that will be hardware programmed. These "memristors" allow for more than just a one or zero value and can simulate real neurons. I think this is the future of AI, a sort of analog/digital hybrid computer.

Of course if this was to be the way to go, we'd have to hardwire the "Three Laws of Robotics" a la Isaak Asimov into them! Then we won't have any Terminator-like problems.

That very interesting SuperNatendo, so what your basically talking about is hardware based neuralnets, is that correct?

exactly. Memristors are not bound to ones and zeros, they can hold values other than on or off. This is very much like neurons.

I thought neurons are boolean: on or off, all or nothing. Just like our artificial binary logic. I think the only way neural impulses can do "maybe" is if you're reading frequency of impulses... but that's still digital not analog.

Neurons are not Boolean, that is why when simulating the mouse brain they had to simulate a neural net. If neurons were Boolean, they could just program neural nets at the hardware level using transistors, thus simulating a mouse brain at 100% the speed.

This WILL be possible using memristors which do not need to follow Boolean logic.

When something is programmable at a hardware level, it is said to be analog, when it is programmable on the software level, THAT is digital.

Digital information is stored using a series of ones and zeros. No digital information has, in the past, been able to store multiple values on a single element of a circuit before now.

Maybe what is Analogue and what is Digital needs to be re-defined

Originally Posted by KALSTER

Originally Posted by SuperNatendo

Well, digital computers running software that allows AI might not be the answer...

It seems the new memristors may allow analog computers that will be hardware programmed. These "memristors" allow for more than just a one or zero value and can simulate real neurons. I think this is the future of AI, a sort of analog/digital hybrid computer.

Of course if this was to be the way to go, we'd have to hardwire the "Three Laws of Robotics" a la Isaak Asimov into them! Then we won't have any Terminator-like problems.

Haven't you read or seen the movie "I, Robot"?
The three laws of robotics ends up creating a protest by the robots.

I think the robots should just follow the same laws we follow and thats all. No extra laws... Just the supreme law of the land followed by both humans and robots.

Originally Posted by SuperNatendo

Neurons are not Boolean

They really are, uncannily so. A nerve/neuron "fires" (1) or doesn't (0). It must be stimulated beyond a certain threshold to fire, then it fires in a perfectly predictable consistent fashion, then it recharges, then it may fire again if sufficiently stimulated. A neuron is built to resist analog signals. It doesn't do fuzzy logic. Just binary: ones and zeros.

Memristors may be a good shortcut to AI but they don't mimic our neurons so well as a the transistors in a cheap calculator do.

Moreover, our very senses are digital. Our nerves, are neurons same as grey matter, just stretched out. We've got some neat devices at the ends to bonk them with, also digital. For example a light sensitive molecule that flips one way, or the other. On or off. A multitude of these, with different filters, can approximate a scene before our eyes.

Originally Posted by SuperNatendo

When something is programmable at a hardware level, it is said to be analog, when it is programmable on the software level, THAT is digital.

The usual distinction is, analog is a continuous range of meaningful values (a rheostat tuning knob), digital is made of stark on/off states (a collection of neurons blinking).

Wikipedia:

The artificial neuron receives one or more inputs (representing the one or more dendrites) and sums them to produce an output (synapse)

Doesn't sound like boolean logic to me! I'm pretty sure the values of our independent neurons give more information than just on or off.

Here is a link to a wiki article about dendrites, it plainly shows that neurons handle more than 1 or 0 http://en.wikipedia.org/wiki/Dendrite

Also, neurons operate on both electrical and chemical signals, this alone requires them to be handle multiple values. Before memristors, we could not envision any single circuit component as complex as the neurons. Memristors are the only thing that came close and until a few weeks ago were completely hypothetical.

Originally Posted by SuperNatendo

The artificial neuron ... sums...

Doesn't sound like boolean logic to me!

You're right, that artificial neuron isn't acting like a real neuron. Real neurons can't add or transmit values. They blip, or they don't. It takes a lot of them blipping to convey values like "maybe" or "half".

Originally Posted by SuperNatendo

dendrites... plainly shows that neurons handle more than 1 or 0

Yes, they have multiple inputs, quite a crazy lot of them too. This is how things become fuzzy. Not in the binary relaying function of the neuron, but in the maze of connections between neurons. A single neuron may be directly connected to thousands. It's analogous to the copper paths on a circuit board. Our brains are dominated by pathways, not components. Why we say neural network.

Originally Posted by SuperNatendo

neurons operate on both electrical and chemical signals, this alone requires them to be handle multiple values.

A light switch operates on both electrical and mechanical signals. It still can't signal anything but "on" or "off" no matter how you wire it or fladdle at it. Also bear in mind that "electrical" and "chemical" are different ways of seeing - electrochemical. Is a battery chemical or electrical?


You seem to think that an individual neuron must be sophisticated and nuanced, for our minds to be. However, those qualities owe to the complexity of network.

it is a combination of complexity of network and complexity of singular elements such as the memristor and neuron.

Here is a link showing how it has been discovered that neurons don't follow boolean logic all the time; Tell me, where is the evidence that this new discovery has been debunked since it was discovered? I know it used to be taught that neurons operate on binary just as transistors do, but the new evidence shows that those teachings were misinformed:
http://www.yale.edu/opa/newsr/06-04-12-04.all.html

Why would they need to simulate a neuron using artificial neurons if a transistor could do the same thing? The truth is, a transistor cannot perform in the same manner as a neuron which is why these artificial neurons needed to be simulated by software. This software simulation required the use of more than one transistor per neuron operating at a far lesser speed than a neuron is capable of. The memristor CAN function more similar to a neuron than a transistor which is why it will benefit AI.

the transistor operates solely on electrical, the light can be made to work on a digital circuit as well, eliminating the need for a mechanical input, a neuron, on the other hand CANNOT function without it's chemical input.

The brain operates both on a digital and analog level, a hybrid of sorts which will now be better emulated with the physical manifestation of the once theoretical "Memristor"

Finally, all those guys really got was what we know already: electrical noise from migraine or seizure can activate neurons. OK, are you next going to bring in neurotransmitter chemistry to prove neurons are digital/analog hybrids?

Originally Posted by David McCormick

It’s as if everyone thought communication in the brain was like a telegraph, but actually it turned out to be more similar to a telephone.

Because there sometimes is measurable background noise on the line! When you burp, do you rate that language? How about the sound of breathing when people speak? Ignore it normally. It's a side effect of talking, not the message. Here is an example, turn it up and pay attention to David Suzuki drawing breath from time to time as he narrates. Listen for how the rhythm of breathing/talking alters his speech in an analog way. I'll warn you though it's really hard to consciously catch and focus on because we normally filter out such noises and "useless information". Then tell me if that's not superfluous to communication: Neurons and How They Work

Pong, you act as though everything to know about the brain and how it functions has already been learned. As you can see, there are still many mysteries. I have shown evidence that neurons operate using multiple values, you have not shown anything showing an otherwise boolean operation of neurons. Just as Memristors "fire" at different times yet can "fire" separate values just as neurons do.

And what about the need to simulate neurons in software instead of just a complex circuit of transistors? You have not refuted my point as far as this is concerned.

My point is, neither of us fully understand neurons, but the evidence suggests more than just boolean logic is occurring. We know these are not "burps" because they are being processed just as what you call the "main message" is being processed.

The brain is operating on more than 1's and 0's, just as a true AI will need to by using memristors. No transistor can take on multiple values as a neuron does.

Originally Posted by SuperNatendo

Pong, you act as though everything to know about the brain and how it functions has already been learned.

I shouldn't. We understand a single neuron to its individual atoms. Understanding the elaborate network of neurons is something else!

Originally Posted by SuperNatendo

And what about the need to simulate neurons in software instead of just a complex circuit of transistors? You have not refuted my point as far as this is concerned.

I've already said memristors might prove a good shortcut in our simulations.

Here is a quote from this article:

http://computerworld.co.nz/news.nsf/...25745100805915

"Here's a comparison I like to use. The human brain is estimated to have about a 100 teraflop capacity. We only use maybe 10% of that — some of us even less. That means the Livermore computer is about 4.7 times faster than the brain, and the new Roadrunner system will be more than double that. It's seriously fast."

The article is talking about the new supercomputer being assembled in Los Alamos, New Mexico.

This proves to me that the brain obviously registers more than simple ones or zeros, otherwise we would not be sentient, just as a computer is not. We already have the complexity of the circuits and network down, the missing link will be given by the memristors.

If we used our brain in the similar manner a computer operates, we would be much less intelligent. Even a mouse brain trumps these computers that can do 4.7 times the amount of calculations our own brains can. And these things are so huge compared to our brain! They need to be transported using a fleet of about 21 Big Rig Trucks!

Either this guy's estimate is way off about how much of our brain we use for thought, and how many calculations it performs, or the components of the human brain can do much more than simply binary calculations.

Yes the brain is subtle and fantastically complex in its entirety. That doesn't mean an individual neuron is!

Another example: chess. You know how to play chess? The pieces follow simple rules. There is nothing analog or fuzzy about a bishop. So how is it that a game of chess until recently baffled our best computers? It's the practically infinite relationships between multiple pieces, that is deep.

Go is comparable to chess in complexity. Yet the pieces are only white or black stones.

Go is even more complex than chess....thats why Go bots aren't as challenging as chess bots.

I agree with you pong, the complexity of the brain is a product of its network topology, and not down to the individual neurons.

I'm not saying neurons are some super complex mystical structures that can never be duplicated by a single component of a circuit. I am simply saying they do not follow boolean logic, just as a memristor does not.