In command systems there is a current world picture from which a situation assessment is made. This picture is a "context" which means that it has a validity based on the presence of objects which are in the locality defined by the sensors used to gather that picture. The picture must be continuously updated so that lost contacts are purged and new ones are added. Often other elements in the picture are also gathering data - these element may contribute (via communications) to a common world picture (eg Recognised Maritime Picture in naval command systems)

These systems are typified by being geographical, realtime and sensor derived and employ tracking and sensor-data fusion.

The presence of so much data is handled by means of command imperatives such as determining threat. Thus one might say that there is a "focus of attention" and that implies and set of prioriterised goals (to select the focus). There are also possible picture evolution (views or viewpoints ) which can be constructed which have the effect of a change of goal priority.

This architecture used to be known as a Blackboard architecture in the past. Typically the Hearsay II or HASP project are quoted see R. Englemore and A. Morgan, Blackboard Systems, Addison Wesley, 1988. After being distorted into parallel system with the blackboard being a kind of scheduler the architecture has declined in popularity, but as a real-time system for handling context bound problems it worth a look.

Another area which springs to mind is by J Barwise (80 and 90s) see http://en.wikipedia.org/wiki/Situation_semantics

I hope that these will be of use

Leonid, this reads as a provocative question because I cannot agree with your first statement even though the overall question you are raising is interesting. To me, situations are not just about the entities or objects but also the relationships between them, the perceptions involved in discerning these relationships into patterns, and the moments of conscious experience involved. When I recently moved to a new place to live I was noticing lots of things - where the closest shops were, the short cuts, the services I may need to access. But I was also actively trying to notice all these things, to learn quickly the important landmarks etc.

From a different line of response the research & development associated with the Semantic Web is all based upon representing entities & the relations between them.

But it's your title that is most provocative. Probably an open question ... I can figure out how I could find myself "in" a number of situations but situations themselves seems like a conundrum of semantics

Like jon I also think that some elements are missing in your conceptualization of situation. The most important is the perceptual element.

Anything that exists in the external world is limited to the mental categories that exist in the mind of the perceiver. If we do not have say the mental categories of ice, and snow then we would not be able to see it in the external world even though they exist for those who have the mental categories to relate it. For those without mental categories fo ice and snow, they will say that they are seeing a white rock that is very cold when we touch or something like this. But again this perception will emerge only to those who have the mental categories for rock and coldness.

So, the worldly objects exists for us because we have developed mental categories for that.

In Indian philosophy it is often quoted, "Yatha pndande tatha brahmande " (it is transliteration of Sanskrit version) . It means that what exists at the individual level consciousness only that exist at the Cosmic level or at the level of Universe.

Dear Jon,

I agree with you. Of course relations are important, I omitted them for brevity, and for three other reasons. First, relations might be similar to objects in some way. Second, objects make up possibly 90% of the situation meanings. Compare "Cat chase dog" and "Dog chase cat" - very different situations and (Cat, dog) misses a part of the meanings. But among millions of possible situations (Cat, dog) selects few possibilities (so, much better than 90%). Third, just considering objects is sufficient to illustrate the question that seems fundamental to both of us: how do we know what is an important situation worth understanding and remembering, and how do we discard a lot of other combinations, which are random collections (of objects and relations)?

Dear Rakesh

I enjoyed your discussion that we can only see what we know about (I am simplifying).

So what do we know about situations that we can see them? Where is this knowledge coming from?

Dear Johann,

Thank you.

Sometimes moving bright salient objects are sufficient to understand what is going on. But often non-salient objects are equally important.

Now my cat enters the room and sniffs an electrical outlet. Why? Any human being would not even notice it, unless specifically looking for it. How do we know this?

It seems your analysis assumes a lot of knowledge, such as what you need. Let me look at your paper.

Dear Leonid

Thanks for taking my off the track response in a positive way. As I am not much aware of the filed of AI, initially I failed to understand that 'situation' here is related to the field of AI. Even, I failed to notice the topic areas to which it has been tagged. But after your response to my answer and seeing other responses I came to notice that it is question related to very a specific interdisciplinary field.

Frankly speaking I can not contribute to this discussion.

Take it as one example (or case) how the human brain learns useful situations and sorts them from random collections of objects.

I would be following this thread and hope may develop some idea of this area that overlaps with the field of psychology also.

Regards

Rakesh,

I think it equally concerns AI and cognition.

@Leonid OK, I think I understand a little better what is is you are conceptualising, though I am still concerned with some of the semantic issues. Object modelling certainly has its strong points, and with your explanation the statistical weightings seem compelling. But a situation is not complete without the relations that might drive it - even though relations can also be seen as "objects" as they do in propositional logic. Also, while your dog & cat example makes sense I question whether it trivialises the notion of a situation. Most situations seem complex from a modelling perspective because there may be countless causes all combining. And, I am suspicious of situations being de-composed into just "descriptive" components (whether as objects, statements, or propositions). Often to be satisfied with comprehending or understanding a situation human beings also need a plausible "explanation", a rationale - not just a description - to actually "get it". The weakness of object modelling is that it just describes and doesn't explain. So, if we want to know how the brain learns about the benefits of some situations over others then we are going to want a fairly robust explanation.

In trying to re-think your initial question I'm also thinking that it is "learning" that teaches us to recognise some things more than others in the environment. At a neurological level "pruning" takes place that strengthens some pathways into memorable ones.

But this doesn't really answer your question! To get to your core question: is it to do with "how" the brain discerns & learns as opposed to the "why"? Actually, both these questions interest me.

Love makes you know what you want, and doubts make you take the next step, and obligations guide your feet. (art?)

El AMOR te hace saber lo que quieres y las DUDAS te hacen dar el siguiente paso, y las OBLIGACIONES te llevan los pies.

Joachim,

I agree, it is easy if you know which objects are important. But a "dog chase cat" scene could be not about a dog or a cat, but about e.g. a nuclear plant, or about a type of grass, or even about something less noticeable. In an opera hall, how do you know to ignore a shape of chairs for understanding the meaning of a situation?

Dear Jon,

I agree with you, both how and why are important. If you know why it is 99% of a solution. But people are capable walking down the street and understand what is going on without any specific why. How our mind does it?

I think I look first persons.

Dear Ana Maria,

Certainly love solves the problem. But how do you know what to love? Man certainly are stupid enough, very few know what to love, while thinking we are very smart. In the US the last three weeks TV shows a trial of a woman who killed her lover with 30 knife stubs. According to his friends he was a nice guy and they warned him against her. I am sure he thought he knew what to look for. Are woman smarter?

I think that all person (men and women) look for the situations whose make we feel loved. This make our feelings and likes. Our goals. The problems we must learn to achieve our goals make our "look for" the next situations (our doubts). (If we want to grow more). (If we want to be better we want to grow). With bad people all the normals rules not run, because they go backwards (not with love in the goal, but the lie). (I think).

About women, I think that we know more about feelings, emotions. For that we possibly recognize they more quickly. (The feelings of another woman even better still). (Women may be fooled by men more, too, I think. But this is because other problem. Our different necessity).

If we have empathy and the need to show what we are, we need to look for the feelings of other. If we need to help persons we need to know their feelings. We look this more, and learn. To know about feelings is a learning like other. Possibly women needs more to know about others, (and men are more distracted). (I think).

Dear Ana Maria,

Love and how to recognize it is a great topic, but I am not sure science is ready.

I think the training of brain is going on after the birth under guidance. From that point different objects which the child can view, under the supervision his neurons can distinguish the meaningful or good and bad. before this training all the objects r similar to him. thats why the child can eat soil, garbage or any other thing at the beginning. slowly he refrains from taking those not good objects. as he grows up then he comes across more objects.for all those objects now he may not get the guidance but he uses his previously gathered knowledge of other objects may be similar or sometimes he tries to apply his other organs like ear, tounge, etc. to get somethng meaningful information abt the new object and store it. Now when he is fully grown up person then his brain contains lot of object information, using all those information he tries to determine the meaningful info of the new object. But one interesting point to note that, all those new assessment may not be correct always, that depends on the prev stored information or training of his brain.

Problem solving involves multiple brain facilities. Definitely in includes memory, experience, imagination and emotions (which in itself are not independent). I agree with Ana that emotion might be more important, also in a scientific context, because our memory and ability to combine unexpected properties is so dependent on experience and emotions associated to those experiences. My point is that we should avoid looking at problem solving as a technical issue associated to abilities to memorize etc, but more as complex human processes like those recognized in creation of art.

Joachim,

It would be simple if you know the answer. But how do you know what to filter out and what to filter in?

In command systems there is a current world picture from which a situation assessment is made. This picture is a "context" which means that it has a validity based on the presence of objects which are in the locality defined by the sensors used to gather that picture. The picture must be continuously updated so that lost contacts are purged and new ones are added. Often other elements in the picture are also gathering data - these element may contribute (via communications) to a common world picture (eg Recognised Maritime Picture in naval command systems)

These systems are typified by being geographical, realtime and sensor derived and employ tracking and sensor-data fusion.

The presence of so much data is handled by means of command imperatives such as determining threat. Thus one might say that there is a "focus of attention" and that implies and set of prioriterised goals (to select the focus). There are also possible picture evolution (views or viewpoints ) which can be constructed which have the effect of a change of goal priority.

This architecture used to be known as a Blackboard architecture in the past. Typically the Hearsay II or HASP project are quoted see R. Englemore and A. Morgan, Blackboard Systems, Addison Wesley, 1988. After being distorted into parallel system with the blackboard being a kind of scheduler the architecture has declined in popularity, but as a real-time system for handling context bound problems it worth a look.

Another area which springs to mind is by J Barwise (80 and 90s) see http://en.wikipedia.org/wiki/Situation_semantics

I hope that these will be of use

Dear John,

These architectures are better than nothing, still they are far away from the human mind capability. The reason is the ubiquitous combinatorial complexity of all AI systems and all approaches to modeling the mind - all these approaches need to consider combinations. They do not really work for the same reason the logic does not work - as proven by Godel. Many know about Godel, but think that their areas are no affected by this fundamental limitation of logic.

Combinatorial complexity is similar to Godelian inconsistency for a finite system (like computer).

Leonid,

- Formal Logic is not everything - to me its a restriction imposed on a heuristic process (thought) for the convenience of .... best not said.

Define Complexity

... I believe that complexity only applies to models.

... Combinatorial complexity may be tautological

What is wrong with starting with a few practical solutions and building on them?

Godel's analysis and formal logic belong to the world of models as does any symbolic proof. I am an Empiricist I don't give up just because some theoretician says something is impossible. I simply assume that their model was wrong and look at the real world for answers.

Start with Context in a real-time system. Understand the nature of control in environmental-sensor rich systems. Scrap the logic and computing orientated approaches and re-visit an old-stager like Norbert Wiener (1950s) try his book " Cybernetics". Somewhere along the line, so-called AI got lost - we took a wrong turn.

John,

An excellent comment.

Complexity I always use as the number of operations that an algorithm requires. This is pretty standard.

I agree about Wiener, he is a smart guy and I learned few interesting ideas from him, but he is a mathematician and eventually lost "intelligence" while retaining artificial.

About logic, empiricism. and models. Here is my idea of what is science as a method (I think pretty standard):

1) intuition about the world and self, 2) a model that explains it (or a significant part), 3) makes experimentally testable predictions, and 4) experiments testing these predictions.

What you think?

Threat --> focus of attention. (John)

I think this is very important. A brain (a person) look for different things if feel threat or not ( in peace). The moments of doubt or threat are registered stronger than the sure moments. And possible determine our look for.

Leonid,

To comment briefly on the combinatorial complexity argument, proposed here (if I understood properly) as a fundamental limit of formalized AI systems, I must point out that:

1) Combinatorial explosion does not imply an intrinsic necessity to examine every possible states (actually, almost any polynomial-time optimization algorithm works on exponential search spaces), and it is not currently proven (unless somebody just became very rich overnight, by settling the "P=NP?" question) whether such a combinatorial explosion would necessarily impact a "formalized/mechanized brain".

2) "We know that logic does not work since Godel". Again, a bold interpretation: Godel shows that formal systems have "blind spots", but what makes you believe the "real brain" does not?

I personally root for formally expressible models, if only because they possess the merit of being falsifiable.

I would like to answer some parts of your question indirectly.

In real-time machine vision applications we are also facing similar problems: we must filter out as many irrelevant information as we can. The short answer: it depends on the purpose of your actions (or the context in terms of John). As I stated before, logics have deficiencies like describing intentions, context, and multi-dimensional data. To overcome these problems in machine vision applications, you must formulate first what are you doing and how it would be affected by the environment. I mean formulating in the sense you look at a great amount of input data first, and make some observations. For example, in pedestrian recognition problems you can take a bunch of images with pedestrians and many others with no pedestrian on it, and you make an "average" image for both cases. Thus, you find some clues (or else you do some other trick until you find one) which helps you to realize if there is a pedestrian on the image and if so, is she coming towards you or not. The fact is, by using some heuristic design like this one you always find a relevant pattern. Can you tell the computer what to see apriori? Absolutely not, but you can make expirements to see some patterns which helps you to find an answer, and make the computer understand. In data mining, that is why clustering based classification is the most popular way to build models: clustering techniques find similar patterns to break down possible "contexts" (scenarios) and classification techniques are recognizing the sort of "danger" in them.

Is it different for human brain? I think not at all... In early childhood one makes several experiments to see how some things work. Red usually means beware, metallic elements are cold or hot, and you can see how a man/car/etc can be in a collision course. We are clustering information (using analogies). What to see? Simple patterns. They might (and should) be different for individuals - at least in complex situations, however, there will be a frequent one. For collision avoidance, you use peripheric vision receptors that tell you if there is something coming in your way. But, in order to avoid to crash a child running out to the street from behind a car some people learn to look for shoes or feet below parking cars (common among taxi drivers), some are looking so ahead that they can prevent such an awful situation (common among manager-like people), and so on. The key is always patterns (stored, filtered knowledge acquired by observations or education), analogies, and induction...

Dear Joachim,

These are all important examples.

Dear Ana María,

I agree, understanding of threats are inborn, otherwise an organism does not survive. My question is how cultural knowledge accumulates before its importance becomes established?

Dear Yann,

Real brains are not logical devices. Aristotle knew this. And it is obvious to everybody. Still one has legal rights to ignore it and to study what he or she likes. You are right, in noticing that some of my statements are not exactly logically formulated. Aristotle wrote about this that "if we would formulate all statements exactly logically, it would be ridiculous. Language has the appropriate vagueness for most purposes."

Dear Zsolt,

Machine vision is an area where this problem is important. You are right in saying that a human analyst formulating the problem solves the most difficult part. But how to do it without the human mind? Or equivalently, how the human mind does it?

"clustering techniques find similar patterns" - if you have a correct measure of similarity - ho do you know it?

"Is it different for human brain?"

I think human brain uses knowledge accumulated in culture with the help of language. Accumulation of knowledge is not as simple - otherwise there will be no differences among cultures.

I think words are the more important system of entrance of knowledge in human, but this knowledge is not only in words in our mind. In our mind knowledge is a sensation, a sense, a complete feeling more than words.

How to store our experiences in brain is the question. Possible by feelings.

"Consider situations as collections of objects. When looking in any direction we see hundreds of objects. Most of these objects are not related to any meaningful situation."

Biological vision system took hundred million of years to evolve and they evolved at the same time as most of these natural objects evolved. These objects are all related and the evolution of these vision systems is the evolution of the discoveries of the relation within forms and between forms. So when a human infant is borned and open his/her eyes, he/she see with this hundred of million of years of learning and learn from there.

Louis,

You are right,

Whichever nonhuman animals can comprehend in the world around is based on evolutionary inborn mechanisms. But their learning abilities are limited. Chimps and bonobos trained for years by psychologists can learn objects and sequences. But even most capable ones cannot learn situations. Apparently language is necessary for learning situations and abstract concepts.

Leonid

Dear Leonid,

What is uniquely human is one thing: ARTISTIC SENSUAL LEARNING or IMAGINATION and this has allowed all othe r uniquely human characteristic such as language which is the art of theatrical enactment. Because of this we are the most cultural integrated species and it is the process of the completion of this integration that is the major driving force of history. Language as all the other arts allow us to subjectively create an pseudo-objective world of communication which allow us to think together. Learning is below language and directly in the sensual/imaginative layer. All abstract concept originate first as images from this layer and can later unfolded into the communicative pseudo objective world.

Regards,

Feelings not only depends of outside, if you are afraid you look different than if you are safe.

Taking an ecological perspective to 'situations', one way is to think in terms of a maze, and given stimuli, to proceed by muddling through, eventually leading to a solution/mapping/sorting. Of course the muddling through must be coupled with feedback to appraise the situation in terms of affordances given objects and obstacles in the navigation space (or any space consciously-investigated). So a theory based on interaction, may lead to the incremental but thorough parsing of the solution space.Intelligence may be of a low level type-based on impulse more than look forward (the scenario of 'infinite objects'may already reduce this to zero?)

http://ai.ato.ms/MITECS/Entry/gibson1.html

https://www.google.com.my/url?sa=t&source=web&rct=j&ei=cAQWVb_EKc-iugTej4HIDg&url=http://www.jstor.org/stable/973677&ved=0CJABEBYwCw&usg=AFQjCNEFUPbD2UjyBdIAPh-CypOUdxrXiQ&sig2=dhS7fb2UX6BRSUpnmY3b6w

https://www.google.com.my/url?sa=t&source=web&rct=j&ei=SQUWVYGEJIOBuwTelYLIDg&url=http://folk.uio.no/alexanje/phd/jensenius-phd.pdf&ved=0CCsQFjAD&usg=AFQjCNHcxEj6vZlz5cR0qKHB5AzoWSU04w&sig2=4RgsposqmrIGtTKl0kwE8Q