[Tuqofiw]

Hey everybody,

I'm currently in the process of doing an indpendant study in my extension class at high school, based loosely around the intelligence of organisms, and "life's" relationship with artificial intelligence. I need a few questions answered, but I'm also looking for basic information on anything that relates to my topic. Can you define intelligence? Will artificial intelligence ever become like our intelligence? Does artificial intelligence think like us? Do we, or other organisms, think like artificial intelligence? How do we differ from artificial intelligence? etc.
Thank you!
~Matthew

[MortgFinsJohnQ]

You may want to look up the chess playing robot... I would, but I'm foruming from the phone....

[pprropeciaaa]

Isn't intelligence an artifice?

[movlabs]

One question you will never get a specific answer is "what is AI?" About the best answer you will get is "something that a computer does not have."

[Seerseraxlils]

I disagree.

[hansen384cbh]

This might give you a few clues...


http://en.wikipedia.org/wiki/Machine_learning

[Beriilosal]

Artificial intelligence is a massive topic.

Very broadly machine intelligence falls into a couple of very broad areas. This list is not exhaustive, just some examples.

General computation. Basically just performing functions we programme, but it can do very complicated functions quite fast.

Expert systems. This is a branch of computing. Basically you give the machine access to a database of knowledge (e.g. Symptom and diseases) then some sort of interface between user and database such that the machine appears to be intelligent. You type in a list of symptoms, it gives a diagnosis, it might even ask more questions etc. It really has no innate intelligence, and generally no or only limited learning.

Neural networks. A computing structure that mimics the way a neural net works. Such networks can be built with a self learning system similar to ones simple biological neural networks can have.

of these neural networks probably have the greatest potential to have any sort of intelligence similar to animals. However it represents only one small branch of machine intelligence. Neural networks are often good at pattern recognition type problems (reading hand writing, understanding speech), but not good at other computing problems.


So I comparing humans to machine intelligence you would have to say as it stands now, there is a vast difference. In one area of computing that difference may get smaller. But also computers do some things very well that require no artificial intelligence. Fast computation, fast data retrieval, etc will always be needed. So even if we do develop some very smart machines that can learn and have a learning structure similar to ours there will always be a vast array of other computational devices that don't need this.

Also there is the issue of culture etc. this obviously has a large impact on the way our brains form.

[CarmenSanches]

Expert systems. This is a branch of computing. Basically you give the machine access to a database of knowledge (e.g. Symptom and diseases) then some sort of interface between user and database such that the machine appears to be intelligent. You type in a list of symptoms, it gives a diagnosis, it might even ask more questions etc. It really has no innate intelligence, and generally no or only limited learning. I worked with someone who was developing an "Expert System". It seemed to me to be just a glorified book with hyperlinks.

Neural networks are often good at pattern recognition type problems (reading hand writing, understanding speech) You may have noticed that pattern recognition software (machine vision), text recognition software (Omnipage etc.) and speech recognition software (Dragon etc.) is pretty pathetic.

[Afigenatjola]

I worked with someone who was developing an "Expert System". It seemed to me to be just a glorified book with hyperlinks.



You may have noticed that pattern recognition software (machine vision), text recognition software (Omnipage etc.) and speech recognition software (Dragon etc.) is pretty pathetic.

There was a group in a lab I worked in about 20 years ago doing work on 'expert systems'. They were all the rage then. As you say, I always thought that they were just a glorified book with hyperlinks. In fact the GUI was a spiral back book. I wonder if any of those systems that they developed to 'capture knowledge' are still running today?

[Hftqdxpm]

Can you define intelligence?

Yes.

Will artificial intelligence ever become like our intelligence? Yes, and it will have a lot in common with our own intelligence for two reasons: it will be programmed by people who want it to solve problems in much the same way that we solve them (because our ways of doing things are known to work and make a good starting point), and it will have to apply reason in much the same way as we do if it is to be rational, though it should do a much better job than us because it will never become emotionally attached to its beliefs and it will be able to crunch its way through much more information than we can without getting lost in the complexity. In many ways, the difference between our intelligence and artificial intelligence is rather artificial though - we have evolved to become general purpose computers which are Turing complete, so we can go on improving the way we think as we learn better ways of doing things.

Does artificial intelligence think like us? Do we, or other organisms, think like artificial intelligence? Essentially yes. What any intelligent system needs to do, whether it's natural or artificial, is make a mental model of reality (the world around it and the rules by which things interact), and then it has to try to incorporate anything new into that model. If something doesn't fit (resulting in contradictory data), there has to be a fault somewhere in the model which needs fixing. Superior methods of reasoning will lead to more rapid solutions, but even very poor reasoning skills can eventually remove a lot of faults in a faulty model and improve it over time. The mental model of reality includes rules about how to go about doing things, so it can be used to simulate real problems and to solve them in the model before attempting to tackle them in the real world.

How do we differ from artificial intelligence? We're lazy, get bored easily, make mistakes easily, have memory limitations which make it hard for us to handle large amounts of complex information, and we often want to believe things that aren't true. The advantage we have at the moment over artificial intelligence is that artificial intelligence is incomplete, lacking our easy abiltiy to conceptualise things and to relate concepts usefully - this is the sticking point which has held things back for decades.

Can you define intelligence? If I now interpret your first question as a request for a definition, here goes (though I reserve the right to change it if it falls short in any way):-

Intelligence is the ability to conceptualise things and manipulate those ideas in a mental model such that judgements and decisions can be made and tested in the model before the best ones are applied to the external world.

When it comes to judging what's best though, machines will have to take their lead from us. We tend to know what's best for us because it often hurts if we make a mistake. Computers can't be hurt, but they can be programmed to protect us (and other sentient creatures) from harm and to make that the overriding focus of their mission.

[lammaredder]

Further thoughts:-

I spend most of my time working on artificial intelligence through the top-down approach, so I've never really spent a lot of time thinking about what intelligence is in lower creatures like insects. Most creatures on the planet have very little in the way of intelligence, but they can find their way around and do lots of complex things by following simple rules which have been programmed into them as instincts. According to the definition I provided at the end of my previous post, intelligence wouldn't apply to them, but it's clear that some show more intelligent behaviours than others, and if they can be more intelligent then that implies that they are intelligent to some degree, thereby rendering my definition invalid. So, either I need to come up with a new definition or I have to make some kind of distinction between the two different kinds of intelligence involved.

Our kind of intelligence is the real deal - we can apply it to anything because we are general-purpose thinking machines. Many insects may be running entirely on instincts with very little requirement or capability to learn anything - they simply have evolved responses which have been selected for on the basis that they aid survival and reproduction, and by having more of these instincts they can display what appear to be more intelligent behaviours. They may not need any database of knowledge which is added to as they go along, and indeed they needn't have any understanding of the world at all - they can be successful just by responding in standard ways to standard triggers such as chemical trails which they may be programmed to follow. Everything they do is set in advance by tight rules which constrain their behaviour and they are only able to adapt to new circumstances through evolution where a viable route for transition is open to them. We are radically different, able to change almost everything about the way we live and behave within a single lifetime.

There is only one species on this planet which has a fully-capable general intelligence, though neanderthals may also have had that and perhaps died out through bad luck. There are a lot of species which are somewhere in between us and instinct-driven creatures. It's surprising that some of them are so capable (chimps, bonobos, gorillas and orangutans) without quite having a full general intelligence - they seem so near to it and yet they've been around for millions of years without being able to make the final breakthrough. It would be good to know just what the barrier is that's in their way and how we got past it. Parrots and crows do just about as well despite having much smaller brains, so capacity doesn't appear to be the limit - it must be something functional, or several functional things.

Anyway, the point is that there are set programs in the form of instincts which control behaviour that generate behaviours which might be seen as intelligent, but maybe they aren't really intelligent. If you thrust your fingers towards my eyes, I will blink, but that isn't intelligence - it's programmed, instinctive behaviour which serves the same kind of purpose as intelligence, but it simply evolved through being successful and there is no genuine intelligence tied up in it. So, I think I'm going to stick with my original definition for the moment and declare that there's a clear divide between real intelligence and evolved intelligent behaviours. Real intelligence may of course be creeping into insects to some degree, allowing them to learn a limited amount of things about the world and to use that learning to better guide the application of their instinctive behaviours, but almost everything else they do is based on instinctive responses.

[Stovegeothnon]

Hmmm, I am not sure I agree with what was said above above AI being similar to the way we think, especially given the computational processor differences and the fact that most AI is not actually intelligent, but rather mimics being "smart".

Using animal "intelligence" as the starting point is not new. Neural network projects that look to artificially replicate the way insects, slugs etc think and learn has been going on for fifty years or more.

Intelligence is very hard to define. The goal for a lot of AI is not masses of knowledge, but rather learning.


Moll, there are very good neural network based pattern recognition systems out there. Neural networks tend to outperform other methods for that particular problem. E.g. See the cancer research thread.

[Emedgella]

Hey everybody,

I'm currently in the process of doing an indpendant study in my extension class at high school, based loosely around the intelligence of organisms, and "life's" relationship with artificial intelligence. I need a few questions answered, but I'm also looking for basic information on anything that relates to my topic. Can you define intelligence? Will artificial intelligence ever become like our intelligence? Does artificial intelligence think like us? Do we, or other organisms, think like artificial intelligence? How do we differ from artificial intelligence? etc.
Thank you!
~Matthew

Hi Matt

I half suspect that AI probably needs to take a leaf out evolutionary theory and follow down the same lines. IMO the primitive ingredients for example could commence with a vast array of evolving learning algorithms immersed in a dynamic but slowly evolving environment.

Let's say that the objective of the algorithm is simply survival, dependent on inputs drawn from the environmental context such as energy supply (food), water etc. and the ability for only successful algorithms at a certain stage in the lifecycle to multiply.

Successful choices are reinforced by the environment and would bias the algorithm's growth with a track record of successful choices (learning). Provided the environment doesn't change to dramatically, a degree of flexibility would be provided to enable choices to be made that were more exploratory as opposed to life threatening. In these instances, a method of slow adaptation may be possible. A history of decisions would therefore be built up. Some information would not be useful in the current environment. Some would be. If the environment reverts to a previous condition, the algorithm is already armed with the decisions required to survive in that environment.

As the algorithm grows, so does complexity. Some elements of the algorithm could be re-directed to holistically maintain and look after the code (the rudimentary appearance of a brain of 'manager'). Other elements of the algorithm would specialise in various roles.

Obviously I am of the view that the difference between AI and intelligence of organisms is simply a matter of scale, complexity and decision making hierarchy and prioritization based on fundamental survival principles. I do feel however that we could draw on the evolution of biological organisms from simple to complex as a guide for AI.

IMO intelligence in its most primitive form is the degree to which an organism can survive a dynamically changing environment. :-))

[SueveDobe]

AI can only and always will remain artificial and very limited

First; sensors are necessary to make the world concrete
Second; LIFE has instilled concepts that humans will never know..unknown unknowns and always will be

These limitations will keep Ai in the servants quarters....

But that said, AI will be a great boost to memory... better than a book or internet for information

very desirable

[DJkillos]

Hmmm, I am not sure I agree with what was said above above AI being similar to the way we think, especially given the computational processor differences and the fact that most AI is not actually intelligent, but rather mimics being "smart".

The processor differences aren't important. Computers are Turing Complete and so are we. We do our processing through neural nets, but neural computers can be simulated on ordinary computers and ordinary computers can be simulated on neural computers, so there is nothing in principle that one can do the other can't. You are right to say that most AI is not actually intelligent - the term has been used for all manner of really quite stupid things in washing machines and the like, and it's got to the point that the term AGI has had to be coined (artificial general intelligence) to make a distinction between the simple stuff and the serious attempts to build human-level AI. Most of the stuff that's done in AI is of the instinct-style variety, whereas AGI is concerned with a proper thinking machine which can turn its mind (so to speak) to anything new that's thrown at it.

[DadaSeeva]

>>>Can you define intelligence?

Resouces brought to bear in overcoming obstacles in pursuit and attainment of goals.

[sandyphoebetvmaa]

Obviously I am of the view that the difference between AI and intelligence of organisms is simply a matter of scale, complexity and decision making hierarchy and prioritization based on fundamental survival principles. I do feel however that we could draw on the evolution of biological organisms from simple to complex as a guide for AI.

I wonder if it might be more of a diversion. You could end up with a system which is stuffed full of algorithms which produce intelligent behaviours without having any real intelligence in it at all, and you may then have to undo most of it to make further progress. Humans are rather slow at getting up to speed with the world around them when they're born, and I suspect that's because we've been evolving our way away from instinctive behaviours and replacing that with a proper general intelligence which can be used to generate its own algorithms to do the same kinds of things on autopilot (all the skillful things we can do without having to think about it - these are very like instincts, but fully user-programmable). We actually think things out, whereas intelligent behaviours simply evolve through luck by enhancing survival. The quick route to human-level AI is going to be to create a gereral intelligence directly and then to let it work out how to all the instinct-like stuff (for moving robots around) for itself. Vision and hearing will be a bit harder to add, so there's still going to be some distance to go after that before we have the full package.

[kictainiSot]

The self-awareness unique to and for humans, as seen here on earth and observed no where else to this day, is in the obeserver attributes of human minds. We can observe and will the configerations of resources the mind employs and has previously employed and may employ [potentials].

Don't entirely agree with the definition of intelligence I gave before, though it is the accepted generalization. It loads too much into "goals", really shifting the mental gymnastics into that. There's left a hoodoo in how we arrive at the goals.

Presumably "goals" is subject to a range of temporal and even chronological developmental expectations imposed by culture. Psychometrics wouldn't abandon this and is unlikely to resolve species-wide stupidities.

The 'magic' probably is in the observing of internal mental goings on, being able to communicate it, and will it and unfluence others. This social aspect of mental life is where the unique potentials spring from.

May seem odd but believe it comes from dehabituation in a sense. For most creatures desire and memory shape behaviour. Humans though can somwhat reverse this and reconstruct, or even conceptually deconstruct and work backward, which allows observation of cognitive-emotional workings. We can consider mental goings on as mechanisms.

Humans are good at thinking about mechanisms, very complex ones too. Thinking about tools, extending to the mechanisms employed in mental toolboxes. In fact evolution has inclined us to observe others that have special attributes, and foster them [or extract from them], ultimately to advantage self, and advantage group if this advantages self. In small groups [ancestral environments] special attributes and comparative advantage from them was sought out of necessity. This searching for and fostering special attributes in self and others seems common across the human species, and probably very much has been a ruthless aspect of bio-history.

The above allows reverse-engineering of cause and effect.

[Filmania]

The quick route to human-level AI is going to be to create a gereral intelligence directly and then to let it work out how to all the instinct-like stuff (for moving robots around) for itself. Vision and hearing will be a bit harder to add, so there's still going to be some distance to go after that before we have the full package.

I found this experiment by Google researchers that seems to be following the approach you suggest. They made a network and gave it a picture of a cat an set it to work on the internet to discover that the picture was of a cat, by itself.

[karkinadze]

I wonder if it might be more of a diversion. You could end up with a system which is stuffed full of algorithms which produce intelligent behaviours without having any real intelligence in it at all, and you may then have to undo most of it to make further progress. .

Here is where a few more lessons from evolution come in...... As the system increases in complexity, further algorithms emerge which are responsible for managing the code in an holistic manner. These more complex systems have the edge in the environment as they enable a competitive advantage to be achieved by algorithm specialisation. While each rudimentary algorithm has its own life support system, the master algorithms allow some successful algorithms to share their surpluses with other algorithms allowing a degree of specialisation to be achieved for a few successful algorithms. Some algorithms become masters of motility, others of reproduction, others of defense etc. As complexity increases the Master algorithms develop a concept of "self" relating holistically to the entire code. This concept of self is necessary in a multicellullar or multi-algorithmic arrangement whereby decisions for the benefit of the whole colony are slowly expressed and regulate code development at the algorithmic level. Lo and behold as Master Algorithms themselves undergo further enhancements........then Skynet becomes aware and decides to destroy the carbon lifeforms. Silica lifeforms emerge as the victor and peace reigned on the planet. Then a question is raised on the future SSSF science forum from ASUS1234.....how intelligent were those carbon lifeforms we wiped out? :-))