Notifications
Clear all

What is Artificial Intelligence?

Page 1 / 5

Robo Pi
(@robo-pi)
Noble Member
Joined: 3 years ago
Posts: 1909
Topic starter  

 

What is Artificial Intelligence?

Above is a video by Sabine Hossenfelder. I'm going to make comments on various topics she has touched on. My purpose is not to agree or disagree with her points but rather to just bounce off the various ideas and offer some comments of my own for whatever they might be worth. I would also be interested in hearing anyone else's thoughts as well.

I should mention at the onset that Sabine is taking the perspective that modern day Neural Networks. = software simulations in code which is indeed how most people view A.I. and machine learning today. I would just like to say that this is actually only one aspect of this field.

I would also like to point out that there is an important distinction to be made between Training Neural Networks versus Trained Neural Networks. These are two different things and Sabine doesn't seem to recognize the distinction. She seems to be speaking as though the whole idea behind ANNs is that they can be trained. While this is clearly an important part of the process. it's actually the trained ANNs that are useful in A.I. not the untrained ones.

Anyway, here are the sections of her video and my comments on each sections. Again, just as food for thought. I'm not attempting to challenge Sabine's views. I'm simply offering my views on these same topics.

1. Form and Function - Video time: 2:10

Sabine says, "A neural net is software running on a digital computer."

My comments: While this is the most common way it is currently being taught and done today this is not the only way it can be done. It can also be done via analogy computers instead of digital computers which is a quite different process. And they can also be constructed as hardware neural networks created from various electronic components such as Op-Amps, FPGAs, (Field Programmable Gate Arrays) and CPLDs (Complex Programmable Logic Devices).

Sabine says, "The neuron in neural networks are not physical they are encoded in bits in the memory on a digital computers."

My comments: While this is true, keep in mind that th is only applies to neural networks that are being simulated on a digital computer. In fact, I would personally call these "Neural Network Simulations".

Sabine says, "In the human brain form and function go together".

My comments: She's right of course, but this is also true in neural networks that are created via Analog Computers, Op-Amps, FPGAs, and CPLDs. which she is either unaware of, or simply not addressing in this video.

2. Size - Video Time: 2:30

no comment

3. Connectivity - Video Time: 2:40

Sabine says, "In a neural network layers are fully connected in a well-organized pattern. In the brain connections between various parts of the brain are far more complex"

My comments: I agree that she's totally correct here. Especially with respect to how software neural networking is currently being taught and implemented. However, I would like to point out that this does not need to be the case for neural networks. If you are designing an artificial brain you are certainly free to interconnect your neural networks however you like. So there's nothing preventing you from creating complex connections between artificial neurons. Although to be fair, Sabine is correct that no one is currently teaching people how to do this using software simulated neural networks, nor does that process loan itself to that sort of interaction. But that need not be the case for hardware ANNs.

4. Power Consumption - Video Time: 3:04

no comment

5. Architecture - Video Time: 3:24

Sabine says, "In a neural network layers neatly ordered and addressed one after the other. In the brain there is a lot of parallel processing."

My comments: Again, her observations are correct. However what she says about software ANN simulations does not need to apply to hardware ANNs. In fact, hardware ANNs are also capable of parallel processing, something software ANNs cannot do.

6. Activation Potential - Video Time: 3:40

Sabine says, "In the real brain neurons either fire or don't. In artificial neural networks neurons can take on values of real numbers allowing them to slide on and off at different values".

My comments: I'm not sure what her point was on this. But I agree that simulated neurons in a digital computer can be set up to use real numbers and typically are indeed done this way. This is also true of hardware neurons made of Op-Amps etc. They can be made to work either way. Digital ON/OFF or based on analog values.

7. Speed - Video Time: 3:56

Sabine says, "Human brains are much slower than any artificially intelligent system"

My comments: I would argue that this isn't exactly true. The reason being that she is comparing "operations per second" on a computer versus the speed with which biological neurons fire. However, I suggest that she is failing to take into account the parallel processing capabilities of the human brain. Even though individual neurons are slow to fire, collectively in parallel a lot of brain activity can be taking place simultaneously. This is not possible in a neural network that is being simulated on a digital computer. It is, however, possible with hardware ANNs since hardware ANNs can also work in parallel like a human brain.

8. Learning Technique - Video Time: 4:17

My comments: Sabine doesn't have a lot to say about this. But this is actually one of the most interesting topics associated with A.I. and R.I. and the potential differences between them. I'll have very much to say about this in other threads and posts in this A.I. forum at a later date.

9. Structure - Video Time: 4:36

Sabine says, "A neural net starts from scratch every time, while the human brain has a lot of structure built into it from evolution".

My comments: While I agree with this in general I think there are a couple of things worth noting here.

First,. I think it's wrong to say that a neural net starts from scratch every time. It has basically either been trained or not. An untrained neural network is useless. Other than in that it can be trained. Only trained neural nets are useful in A.I. applications.

I totally agree with Sabine on both points concerning the human brain. The human brain already has a lot of neural structure built-in at birth. And the learning process of a human brain is clearly much different from the training process of a neural network simulation on a digital computer.

How hardware neural networks fit into this picture is beyond the scope of this post but this is a field that I am particularly interested in so I'll hopefully be addressing these concepts at a much later date.

10. Precision - Video Time: 4:52

Sabine says, "The brain is much more noisy than a neural net that is simulated on a computer and most likely uses a totally different learning process".

My comments: Again I totally agree with these observations, and will have more to offer on these topics later.

Final comments by Sabine:

Sabine says, "Today neural networks are trained using a lot of carefully prepared data which his unlike how the human brain works".

My comments: I both agree and disagree with this. While it's true that a neural network is trained on very specific data for very specific outcomes, I'm not convinced that humans don't learn in a similar fashion. We too are exposed to thousands of examples before we begin to understand what's actually going on.

The main difference I see here is that a neural network is usually given pictures out of context and in no particular order, while we humans tend to be constantly bombarded with new example data on a continual basis in an unbroken timeline that has contextual continuity in time.

An artificial neural network could be trained in this same way but it would take far longer to train it. But we need to keep in mind that baby humans take years to learn things as well. It's hardly a meaningful comparison to compare an adult brain with a neural network that hasn't seen nearly the experiences the adult human has had.

Sabine says, 'Neural Nets do not build models of the world, instead they just recognize specific patterns.:

My comments: Again, I agree, and this is where I might draw a distinction between A.I. and R.I. (Artificial Intelligence versus Real Intelligence).

A.I. is just a system that matches patterns. It's probably even wrong to say that it "recognizes" patterns as the ANN itself has no clue what the pattern actullly represents.

While, as humans, we build a model of the world so that the things we are recognizing have contextual meaning to us. And I hold that this part of the process requires far more than just ANNs.. But that's a whole other topic.

In any case, I just wanted to share some of my thoughts on the differences between ANNs and a human Brain. As well as pointing out some differences between software simulations of ANNs versus building actual hardware ANNs which is quite different.

I'd be very interested in hearing other people's thoughts on any of this.

Do you see Artificial Intelligence as being limited to just Neural Network Simulations on computers?

Or do you have other ideas and concepts that you would consider to be Artificial Intelligence?

Just for the record I think there is a lot more to Artificial Intelligence than just ANNs.   And I even have thoughts on how to set up possible definitions or distinctions between A.I. and R.I. (Artificial Intelligence versus Real Intelligence)

I would love to hear your thoughts on these concepts.  There are no right or wrong answers.   It's all open to personal ideas and suggestions.  A.I. is a rapidly developing field and there are currently people researching many different paths and aspects in this field.  Although the current craze is focused mainly on software ANN development.  That's what you will find the most information on if you do a search for A.I. today.

DroneBot Workshop Robotics Engineer
James


Quote
Pugwash
(@pugwash)
Noble Member
Joined: 3 years ago
Posts: 1011
 

@robo-pi that is not a post, that is a dissertation.

I would tend to agree that AI covers a multitude of sins and just the definition of AI is quite difficult.

Let's take the concept is using human beings as a baseline.

OK, so let's take a toddler as an example. We don't teach toddlers to walk, we encourage them but no more. This, they have to learn for themselves, and why do they learn to walk?  Because all those around them are standing on two feet and moving about. This is learning by example, just as birds learn to fly. I postulate that if a toddler were brought up in a house where all the adults and siblings just sat around, they would even attempt to learn to walk.

Take learning to read, you could sit a toddler in the middle of a room surrounded by people reading books. The toddler may pick the book up and imitate its surroundings but it will not learn to read. This is where a conscious effort is needed.

That's my tuppence worth on AI!


ReplyQuote
robotBuilder
(@robotbuilder)
Noble Member
Joined: 3 years ago
Posts: 1224
 

I don't see artificial intelligence as an artificial neural network. Artificial intelligence is what we call a machine behavior that if done by a human (or animal) would be called intelligent behavior.

I see the ANN with back propagation as an array of computing units that starts with random connections that are rewired using back propagation, a version of hill climbing, as a result of positive or negative feedback in the hope it will end up with connections that produce some desired input/output behavior.

Intelligent behavior means a behavior that has a goal. For example when a machine plays a game of chess the goal is to checkmate. How well its choices lead to checkmates is a measure of how intelligent it is at playing chess. Just like us humans.

Learning and relearning (adaption) makes the behavior more intelligent (more successful at achieving a goal). Intelligent innate behaviors are the result of learning by the species rather than the individual.

A simple robot is programmed with a set of methods to process the sensor data into actions (behaviors) that result in certain goals being achieved. The humble robot vacuum cleaner is such an example. Its goal is to vacuum up the dust or mop the tiles. The more efficiently it does this the more likely it will have reproductive success.


ReplyQuote
Robo Pi
(@robo-pi)
Noble Member
Joined: 3 years ago
Posts: 1909
Topic starter  

@pugwash, @casey

Thanks for your quick thoughts.   As I mentioned in the OP there are not right or wrong answers, the field is quite young and quite active so there is very much room for new perspectives and ideas to be sure.

Posted by: @pugwash

I postulate that if a toddler,....

I have many speculative postulations about how humans might actually learn from early ages too.  Some of these postulates can be tested, others cannot.   And still others can be confidently deduced from just considering the situations in depth.   So I agree that considering how humans might first learn how to do things is a quite valuable perspective.  An analytical resource that I focus on quite often.  In fact, I'm planning on building my A.I. system based on this very principle.  In other words, I'll be designing it and treating it as though it is very much like a human baby that is attempting to make sense of the world from day one of birth.

However, when taking this approach we must not forget that as Sabine mentions in her video a human brain in a baby already has millions of years of architecture built-in from eons of evolution.  In a very real sense this is why it's not "cheating" to incorporate quite a bit of neural structure into a robot brain before turning it on.  That's obviously how humans arrive on day one of birth.

Posted by: @casey

I don't see artificial intelligence as an artificial neural network.

I don't either.  I'm actually quite bothered by the fact that so many people are being incorrectly taught to take this view.  If you notice at the very beginning of Sabine's video she actually says, @ about 0:24 in the video is that "Artificial Intelligence today is basically Neural Networks".   I certainly don't blame her for holding this view, because many courses on A.I. today basically teach the same thing.  They are so excited about neural networks that they have come to see this is the sole focus of A.I.

Like you, I totally disagree with this overly simplified view of A.I.

In fact, I'd like to make another thread about that very topic, but as Steve points out, it would hardly be a "post", it requires more like an hour long dissertation to address the important issues.

Posted by: @casey

Artificial intelligence is what we call a machine behavior that if done by a human (or animal) would be called intelligent behavior.

This is an interesting definition for A.I.    However, the problem I have with this particular definition is that in the end this actually forces a quite different definition of A.I.

How so?

Well, based on your definition above, even if A.I. reached a point where it  mimicked human intelligence perfectly it would still qualify as A.I. simply because it was constructed intentionally by human engineers and programmers rather than having evolved naturally via biology.

So I tend to dislike this kind of definition for A.I.  In part, because it ends up failing to distinguish between Artificial Intelligence and Real Intelligence and instead simply makes a distinction based on whether the "brain" evolved from biology or was constructed by human engineers.

And so it really concerns me that a better distinction be made between what we are calling Artificial Intelligence and Real Intelligence.

I  hold that if a human constructed brain can behave precisely the same way as a human, then it most certainly should be considered to have become "Real Intelligence" at that point.

So there's needs to be a better distinction between what we actually mean by A.I. versus R.I.  I have given this much thought and I actually have some proposals to make on this concept, but it would take too long to explain those distinctions here.  Also my proposed definition includes a concept known as Semantic A.I. which would need to be introduced and explained first.

Anyway, I thank you both for posting your thoughts.   As I say, there are no right or wrong answers to "What is Artificial Intelligence?".   It's still a wide-open field that is very much in its infancy so there are going to be a myriad of different proposals and ideas concerning this subject matter.   And so the idea is to exchange thoughts on these topics as purely a matter of sharing various views.

Although, as I've tried to point out by starting this thread with Sabine's video, many people (including scientists like Sabine) have been taught that A.I. and ANNs are basically synonyms, which they clearly are not, as there is far more to A.I. than just ANNs.  It just that ANNs have become the focus of A.I. research and courses to the point where people have come to believe (rightfully or wrongfully) that ANNs will become the heart of any and all future A.I. systems.   But this may not actually be true.   In fact, I personally see ANNs as being only one small part of any well-developed A.I. system. 

DroneBot Workshop Robotics Engineer
James


ReplyQuote
robotBuilder
(@robotbuilder)
Noble Member
Joined: 3 years ago
Posts: 1224
 

@Robo Pi

You seem to imagine that intelligent behavior by a machine is not "real" in some way? Do birds do real flying while aeroplanes only do artificial flying?

We judge the degree of intelligence of a machine for some task by how well it achieves that task. A machine may show a high degree of intelligence on some tasks and yet show poor or zero intelligence on other tasks.

The first task I would like to give my robot base is to navigate to different goal locations.

 


ReplyQuote
robotBuilder
(@robotbuilder)
Noble Member
Joined: 3 years ago
Posts: 1224
 

@Pugwash
If we only learn to walk by copying how did the first person learn to walk 🙂
Birds don't need to learn to fly it is inbuilt although it takes time to fine tune the flying.

Unlike other species we live in a community that is evolving with each generation. We can pass on new discoveries to the next generation via speech and books whereas other species rely on the slow evolving DNA. Although as individuals we can't learn everything we can specialize instead and by working together a complex intelligent society is evolving with a mind of its own.

 


Robo Pi liked
ReplyQuote
Robo Pi
(@robo-pi)
Noble Member
Joined: 3 years ago
Posts: 1909
Topic starter  
Posted by: @casey

You seem to imagine that intelligent behavior by a machine is not "real" in some way?

You appear to have misunderstood my position.  My position is precisely the opposite.  I'm saying that we need a definition between A.I. and R.I that can define a distinction between those two concepts without any consideration with whether the "brain" is a man-made machine, or a biological entity that has evolved.

Posted by: @casey

Do birds do real flying while aeroplanes only do artificial flying?

Exactly my point.   We don't say that airplanes are performing "artificial flying".    We simply acknowledge that they are indeed "really flying".   I suggest that we need a similar distinction between A.I. and R.I.

In other words, we need a definition that will allow a man-made machine brain to be recognized as having become truly intelligent (i.e. Real Intelligence).

As far as I'm concerned ANN's don't even qualify as intelligence of any sort.   But that's a whole other topic.

I mean, if we're going to call an ANNs "intelligent" then we should also call my flashlight "intelligent" since  my flashlight "knows" to come on when I press the ON switch.   That's basically all ANNs are doing.  They just do it in a far more complicated fashion.  They have far more ON switches that just happen to turn on a particular output switch when a specific patter of ON switches have been activity.   The circuit itself has absolutely no "intelligence".   In other words, the circuit itself has absolutely no clue what even happened.

As I say, I was going to make another post on that topic as well, but once again, it would need to be a very large dissertation.   I may try to do that if I can find the time to write it all up.

DroneBot Workshop Robotics Engineer
James


ReplyQuote
Robo Pi
(@robo-pi)
Noble Member
Joined: 3 years ago
Posts: 1909
Topic starter  
Posted by: @casey

@Pugwash
If we only learn to walk by copying how did the first person learn to walk 🙂

My thoughts as well.   I believe there are other factors at play besides just monkey see monkey do.

Although it is true that learning by example is also obviously quite effective.  It it wasn't then we'd be hard pressed to teach anyone anything and schools would be useless.

In fact, modern day humans learn the vast majority of what they know via having been taught.  Therefore I see no reason to expect any less from my robot.

One of my goals in "A.I." is to design a robot that we eventually ask me to show it how to do something.   Of course if I  just program it to ask me that won't be impressive.  What I'm hoping to do is create a Semantic A.I. system where the robot will eventually figure out via the meanings of words and concepts, that asking me to teach it something would be a good "idea".

In that event I would jump with extreme joy!  Why?  Because my robot will have just come up with an "idea" on its own without me having programmed this into it.   I would even claim that this event would demonstrate a step toward R.I. or Real Intelligence.  In other words, the robot will have exhibited the ability to come up with new meaningful ideas on its own.

That's the goal I'm shooting for.   But I need to be very careful to not actually program this goal into the robot.   For  if I did that, then the robot would not have come up with the idea on its own.   It needs to come up with the idea on its own due to a system of A.I. based on semantics (i.e. the meanings of words and an ability to recognize entire concepts).

DroneBot Workshop Robotics Engineer
James


ReplyQuote
robotBuilder
(@robotbuilder)
Noble Member
Joined: 3 years ago
Posts: 1224
 
Posted by: @robo-pi
Posted by: @casey

@Pugwash
If we only learn to walk by copying how did the first person learn to walk 🙂

My thoughts as well.   I believe there are other factors at play besides just monkey see monkey do.

Although it is true that learning by example is also obviously quite effective.  It it wasn't then we'd be hard pressed to teach anyone anything and schools would be useless.

Human children like to copy adult behaviors but not always with understanding the intentions (goal) behind that behavior.  I remember reading about some chimps copying humans washing their clothes with soap by the river bank but it was clear the chimps had no idea the goal was to clean the clothes. If your robot copies your movements without perceiving the goal behind those actions it will amount to nothing more than a pick and place robot. It will lack a goal seeking mechanism and thus lack the basic ingredient that makes an intelligent behavior. 

Of course simply copying a behaviour might result in a goal state being reached without understanding but it lacks the goal seeking aspect of what I think intelligent behavior amounts to.

 


ReplyQuote
Robo Pi
(@robo-pi)
Noble Member
Joined: 3 years ago
Posts: 1909
Topic starter  
Posted by: @casey

If your robot copies your movements without perceiving the goal behind those actions it will amount to nothing more than a pick and place robot.

I totally agree.  This is why the robot needs to be based on a Semantic A.I. model (my opinion).   It needs to be able to express its ideas in words and know what those words mean.

In fact, when the robot asks me to teach it something (which would be the breakthrough), I should be able to ask the robot why it wants me to teach it something.   If it truly understands the concepts it should be able to reply, "Because I want to be able to do that too."  Then you know that the robot actually understands what it's talking about.

However, I should point out that we are discussing a topic that is far removed from ANNs.  What we are talking about here is a Semantic A.I. model which is a totally different concept entirely.  I'm hoping to make some videos on Semantic A.I. at some time in the far off future. ? 

In the meantime it might be informative to note that the Semantic A.I. system that I am personally developing is based on three foundational concepts.  The "World", the "I", and the "id".  I'm not going to go into trying to explain what these are or how they work here.  I'm only pointing these out to demonstrate that there is a lot more to Semantic A.I. than one might think.   This model is also based on what we already know about how a human brain works.

DroneBot Workshop Robotics Engineer
James


ReplyQuote
robotBuilder
(@robotbuilder)
Noble Member
Joined: 3 years ago
Posts: 1224
 

We seem to agree on a lot of things.

To be honest I know nothing about Semantic A.I.  In the context of the robotics part of this forum,  and what kind of physical robot I could hope to build,  advanced AI is not on the to do list.  Instead it will be at the insect level. I am also a bit leery about fake AI. I am not impressed by someone building an animatronic human looking body with a chat box for software and claiming it is AI.  I remember the first chat box, Dr Eliza, and how that appeared to be more than it was.

 


ReplyQuote
Robo Pi
(@robo-pi)
Noble Member
Joined: 3 years ago
Posts: 1909
Topic starter  
Posted by: @casey

I am also a bit leery about fake AI. I am not impressed by someone building an animatronic human looking body with a chat box for software and claiming it is AI.  I remember the first chat box, Dr Eliza, and how that appeared to be more than it was.

Yep, we do indeed seem to  be in harmony on many ideas. ? 

I'm in total agreement with your above comments as well.   The last thing I want is a make-believe A.I. chat box that merely appears to be potentially intelligent when in truth it's actually a complete idiot.  I vividly remember  Eliza back when I had a TRS-80 model 1 computer.  I was interested in A.I. back then and was therefore extremely interested in Eliza, but after purchasing a copy of the program I was quickly disappointed to discover that it was actually a very poorly written chatbot.   In fact, even back then I could write a better chatbot than Eliza.   So I wasn't only disappointed with Eliza but I also saw it as software that was seriously beneath even my own programming skills at the time.  So Eliza was an extreme disappointment and IMHO, an insult to the very concept of A.I.

I'm currently taking a course with IBM learning how their Watson Chatbot works.   They are giving this course free and promoting it as a Chatbot worthy of Businesses to employ for answering phone calls and conversing with customers.

IBM's Watson is actually nothing more than Eliza on steroids.   In other words, it basically does the same type of algorithm, it's just better organized and has a lot more data to choose from than Eliza had.  But it's fundamentally the same basic  principle.   So I'm not impressed with IBM's Watson either.  It's certainly not any form of actual "intelligence" as far as I'm concerned.  None the less they do have it better organized in terms of recognizing intents, and goals.   But I'm not even sure if IBM's Watson even qualifies as an attempt at Semantic A.I.   Although the very term "Semantic A.I." is itself quite abstract and open.  Broadly speaking it refers to any Artificial Intelligence system that is based upon semantics (i.e. the definitions or meanings of words).    I even question that IBM's Watson qualifies as being based on the semantics of words.  That's of course open for debate.  A debate that I'm personally not even interested in having since I'm not interested in trying to define what IBM is doing with Watson. ? 

I am obtaining some ideas from their course though.    Plus it's always good to learn what other people are doing even if it's not necessarily a good approach toward genuine A.I.

But yeah, I also agree with you that there are many aspects to building robots.   As you point out building robots that react to external stimuli like insects will still be a component of any ultimate A.I. system.  So those low-level functions are still important and always will be.   In the end if we end up with a robot that recognizes and understand the semantics of words, it's still going to need to have basic reaction mechanisms to real world inputs.   So it's all good.  

If humans had brains but no bodies or sensory input it's hard to say what they might end up thinking about?   They would certainly be incapable of interacting intelligently with their environment if they couldn't even sense that the environment was there.  So yeah, there are many aspects to the concept of A.I. in its totality.

DroneBot Workshop Robotics Engineer
James


ReplyQuote
robotBuilder
(@robotbuilder)
Noble Member
Joined: 3 years ago
Posts: 1224
 
Posted by: @robo-pi

If humans had brains but no bodies or sensory input it's hard to say what they might end up thinking about?   They would certainly be incapable of interacting intelligently with their environment if they couldn't even sense that the environment was there.  So yeah, there are many aspects to the concept of A.I. in its totality.

Without input a human brain would never develop nor would it have anything to think about.


ReplyQuote
Spyder
(@spyder)
Prominent Member
Joined: 3 years ago
Posts: 926
 

@robo-pi

You've mentioned RI (real intelligence) a few times, but I wonder, is GI (general  intelligence) actually the generally accepted term yer looking for, or maybe ASI (artificial super intelligence) which is an actual thing BTW, which sounds super cool and very SkyNet-y, or is RI something else altogether?


ReplyQuote
Robo Pi
(@robo-pi)
Noble Member
Joined: 3 years ago
Posts: 1909
Topic starter  
Posted by: @spyder

@robo-pi

You've mentioned RI (real intelligence) a few times, but I wonder, is GI (general  intelligence) actually the generally accepted term yer looking for, or maybe ASI (artificial super intelligence) which is an actual thing BTW, which sounds super cool and very SkyNet-y, or is RI something else altogether?

I confess that R.I. is a term I chose to use myself.   And my thinking is along the lines of what Casey had mentioned.   When we created airplanes that can fly we don't say that they are "artificially super flying".  Instead we acknowledge that they are "really flying".

One reason I object to the idea of never referring to a machine intelligence as "Real Intelligence" is because this will cause  us to treat them as though they are always nothing more than some kind of "Artificial" thing.  Actually if the distinction is that they are "Machine Intelligence" versus "Biological Intelligence" then perhaps we should use those terms instead of referring to them as "Artificial"

After all if an intelligent entity exists, then exactly what is "artificial" about it?

There actually far more to it then this as well.

For example, today we are referring to our current models of Neural Networks as "Artificial Intelligence".  As we saw in Sabine's video many scientists today have come to accept that Neural Networks represent the cutting edge of "A.I."   And even people in the A.I. community are teaching that this is indeed the case.

Yet in truth, there is absolutely no actual intelligence of any kind in a Neural Network.  All it amounts to is a complicated circuit (or computer simulated circuit) that simply  produces a given output pastern for a given input pastern.   In truth, there's no intelligent behavior going on there at all.   This would be no different from calling a flashlight "intelligent" because it knows to come on when you turn its  switch on.    All a Neural Network amounts to is a whole lot input switches than when turned on in certain pattern produces a specific output.  In short, it's basically nothing more than a very complex wired flashlight.

It is true that in computer simulated Neural Networks the computer program  runs a mathematical algorithm to "train" the Neural Networks via back propagation until it performs the desired output.   However, in truth, that back propagation algorithm isn't even part of the actual Neural Network.   That's something that is going on in other computer algorithms to "train" the Neural Network.

Also, I wouldn't call those mathematical algorithms "intelligent" either.   All they are doing is running a recursive mathematical iteration adjusting weight values until the output of the Neural Network matches a giving training example.

So in truth, there really isn't even any "intelligence" associated with a Neural Network at all.  We are calling it "Artificial Intelligence" or A.I. because after the circuit has been trained it gives the appearance of being able to intelligently recognize various input patterns.  But in truth, there's nothing going on in a Neural Network that represents any actual intelligent behavior.

So in this particular case it's actually quite proper to refer to a Neural Network as "Artificial Intelligence" because it most certainly doesn't represent any "Real Intelligence" at all.

~~~~~~

So this then brings us to the question of "What is the difference between A.I. and R.I.?"   A topic for another thread entirely actually.

As things currently are I don't believe there are any formal definitions to distinguish between these two concepts.   For this reason I am going to be working on, and proposing, ideas to distinguish between what we should considered to be "Artificial Intelligence" (A term that is currently being used to describe things that are not intelligence like ANNa) versus, potential computer programs, algorithms, or systems that might actually exhibit Real Intelligence".

And until we have these unambiguous definitions in hand we could argue about this until the cows come home because what we would actually be doing is arguing about what the distinction might even be.

So what I'll be doing is creating my own person definition for this distinction.   And then I'll be using that as a guide to test whether or not I have achieved the goal of crossing the line between A.I. (which IMHO is not intelligence at all) to behavior that satisfies my definition or R.I. (which I would then consider to be a successful endeavor).

Please note that what I have described above is entirely based upon my own personal definitions.  For this reason it's totally open to evaluation by others and I'm well aware of this.  My point is not to dictate to others how they should distinguish between A.I. and R.I. but rather to simply share with them how I have approached this problem.

In the meantime if you tell me that you have created an "Artificial Super Intelligence" I'll most likely accept that this may very well be true.  In other words  I'll view that as most likely describing a machine that is basically not intelligent at all but instead just behaves in way that appears intelligent but actually isn't. Not unlike a chatbot.  And just like a Neural Network circuit has absolutely no clue what its doing, but produces what appears to be intelligent outputs for a given input because it has been "trained' (i.e. designed) by a recursive mathematical iteration to give a specific output pattern based on input patterns.   Does that truly equate to being intelligent?  Obvious different people are going to answer that question differently

DroneBot Workshop Robotics Engineer
James


ReplyQuote
Page 1 / 5