Very interesting question... I would love to have an answer that would be important for the field of ML.

There is some interesting new research into this question from the perspective of cognitive science and neural engineering by Paul Thagard and Terrence Stewart.

Thagard, P., & Stewart, T. C. (2011). The Aha! experience: Creativity through emergent binding in neural networks. Cognitive Science, 35, 1-33. PDF

As to whether a machine could be imbued with a sense of discovery, I would say that in principle it would be possible, given that we have sufficient understanding of its nature and components (e.g., meta-awareness of newness of discovered solution, "frustration" at an impasse, etc.) that we can implement it. We already have implementations of the cognitive aspect of discovery (e.g., Jeff Shraget, Pat Langley, the late Herb Simon).

Very interesting question. But I don't have the answer. I was also having a similar type of question. Can we build machines with emotions.

Maybe it is the capacity of agent to find an important data that represent an essencial goal for this agent. I think that a possible approach it is developing a dynamical cost function

The first problem is to motivate despite uncertainty.

The second problem is to recognize novelty

The third problem is to have goal setting behavior

The fourth problem is to have predictions as to whether or not the goal will be reached.

The fifth problem is to adjust motivation so that obviously faulty goals are abandoned and better goals attract effort

The sixth problem is to recognize achievable goals

The seventh problem is to project beyond recognized achievable goals some meta-goal that might also be achievable

The eighth problem is how to motivate to achieve meta-goals.

I think: there is a triple: internal intensions, based on utility function, situation (external), action. Agent collects such triples in memory and tries to compare the new situations with old ones to select adequate action. Agent's efforts to find up essential features of triples for comparing old and new situations implies "discovering" by mean analyzing the differences.

Niklas Luhmanns system theory might be helpful, to pursue your question. His theory explains how structures and meaning evolve within communication systems. The key of his concept is to define communication as a continuous process and distinguishing between the phenomenons information (what), message (how) and understanding (why).

During the process of communication two communicating parties are varying and selecting what they consider “important”. This requires from each party an act of understanding.

Understanding involves recognizing the difference between information and message and to accept the information as making sense or to reject the information of being meaningless. The information carries meaning (is understood) if it can be integrated into

a network of previously acquired meaningful information (the party's memory).

The act of understanding is key to either continue or terminate the communication process.

Luhmann's concept assumes that each communicating party has the capability of a so called autopoietic system. An autopoietic system is able to create and maintain all its functions according to an existing intrinsic program. Luhmanns dynamic system concept is far more than a description of a system by means of its elements and relations.

Each autopoietic total system has the three subsystems programsystem, interactionsystem and functionsystem. The subsystems interact with each other and are open for variations, selection of variations and rejection of variation.

---> programsystem – interactionsystem – funktionsystem ---

I \ / \ / I

I variation selektion I

I I

------------------------------------------------------------------------------------

Each such autpoietic system has the intrinsic capabilities

1. to recognize variations

2. to select variations

3. to integrate selected variations

4. the functionsystem reacts on coupling with other functionsystems

Variations are considered “irritations” by an existing autopoietic system independent whether they are random (mutations) or intentional. An autopoietic system is considered

“alive” as long as the communication process is continued (reproduction).

The magic behind the word "discovery" isn't backed by simply finding out something you didn't know. For example, I discovered that the lights were left on all night.

The magic of discovery is the significance of new information as compared to the previous knowledge. Here two terms are of importance: "significance" and "compared".

If we can find a metric system for comparing two datums of "knowledge" then setting a threshold for the difference of a new datum would classify its significance.

I don't have a suggestion on how this metric can be designed.

I think that this issue comes to many issues, which do not have a direct connection between them. On the concept of discovery, for example, it does not depend on fundamentals of biology or even the orders of preserving life for a definition of its role in cognitive and intellectual activity. Thus, the process of "discovery" is not the driving force of the living entities, but it could be understood as a result of unvarying conservation orders and balance responsible for the gradual phylogenetic evolution and ontogenetic development.

Piaget gives us the lesson that the child invents and organizes his universe, step by step, structure objects, the space, the time and causality, constructing a logic of reality which is applied in direct relation to the world and its processes of transformation.

Thus, the determining factor or the nature of the orders of preserving life, reflected in immediate and instinctive behaviors, such as the drive for food, sex, social interaction, is not related to the concept of "discovery" because this is the process of an intellectual activity that results from a cognitive relationship with the world, in which the primate or thinking subject perceives the manifestations of nature, threats, needs and changes ("innovations" or “novelty”). Thus, the orders of maintenance and preservation are more related to the regulative laws that govern organic and cognitive processes.

The last point addresses the issue to the problem that involving the relationship between organic intelligence and artificial intelligence. I wrote an article two years ago addressing this issue. A binary computer, not quantum, does not deal with probabilities. This is the first question. In the world of human intelligence, creativity often comes from the error, which the subject insists because he believes, because he insists, because he finds news, and not of the hit consecutive and uninterrupted defined by the program already established. This would be another structural difference that prevents a direct relationship between organic and cybernetic mechanisms.

There is an irreversibility in the mechanisms of life so that playback of synthetic intelligence is the product of a model and will never be the model itself.

Let us consider mappings: datum->action->utility. Utility increment may the metrics.

What I am suggesting is that Utility may only be determined with respect to a goal

The problem of self-motivation towards new "discoveries" is ill-defined. It can be formulated in terms of an endless list of multi-disciplinary approaches.

With that being said, taking an information theoretic perspective, I'd say what you are looking for can be defined mathematically as a trade-off between data compressibility (i.e., the agent's ability to consistently "summarize" and "recall" events and patterns) and surprise (i.e., how unlike is the observed event given the agent's current internal model of the world surrounding it).

One cannot be motivated towards "discovering" facts or patterns already known. On the other hand, one must be able to "compress" the new knowledge otherwise it is likely regard as "too complex" to be worth analysing, given the agent's current learning capabilities.

Therefore, I'd say this problem cannot be put in absolute terms, but only in the context of the learning model an agent has access to.

See: Formal Theory of Creativity, Fun, and Intrinsic Motivation (1990–2010)

Autonomous Mental Development, IEEE Transactions on

Date of Publication: Sept. 2010

Author(s): Schmidhuber, J.

Dalle Molle Inst. for Artificial Intell., Univ. of Lugano, Manno, Switzerland

Volume: 2 , Issue: 3

Page(s): 230 - 247

http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5508364&tag=1

Or intensions

The goal's reachability is a problem of initial state of the agent and sufficent resources for plan execution

I think, all of our behaviors are based on reflex By our mind, each reflex can be coordinated for special purpose Program for special purpose can be made and can be instilled to a machine/robot.

Reaching towards a goal, how do you determine how much further to go?

In the goal state a definite set of conditions (or assertions) is satisfied, in the intermediate state some but not all of them are satisfied, you or agent may just count the number of them may be with weights according to their significances, it'll be distance.

Considering this discoverer's story from Wikipedia: “During his first voyage, instead of reaching Japan as he had intended, Columbus landed in the Bahamas archipelago (…) inaugurating a period of European exploration and colonization of foreign lands that lasted for several centuries. … the discovered place was not India, as Columbus always believed, but a new continent …”;

discovery seems to be related to utility while utility is not necessarily (discoverer's ) goal based. Where am I wrong?

There are no useless dicoveries with zero or negative utility.

There are four units: first generates different ideas, second gives apriori utility value, third estimates aposteriori utility after effect of action, fourth collects ideas and its utility value and use data mining methods for discriminating bad ah good ideas, due to feeback from 4 to 1 units the last from enough period of learning emits only good ideas.

negative feedback I mean

Dr Victor Romanov: “There are four units (…)”.

While the four units seem to be in compliance with classical (rational) artificial intelligence planning, I’m wondering if they can be applied to the case of “accidental discovery” (which may be a redundant term since I think discovery to be a question of chance) ?

Here is another story of (very) useful discovery: “It took place in 1928 when Alexander Fleming left a culture plate smeared with Staphylococcus bacteria on his lab bench while he went on a two-week holiday. He came home to see that the culture had been contaminated by a fungus, which stopped the bacteria growing. He had discovered an antibiotic.”

He found out only experimental confirmation of his vague hypothesis, Unprepared man would not see there nothing but garbage.

the concept of discovery, I consider, may not neglect the structure of causality in between source domain and target domain around with scientific discovery.

"discovery" is a concept and a behaviour. we can use the concept "discovery" as a subclass of activities like "exploration" (synonym) and "exploitation" (antonym?). But the definition of "discovery" is more a behaviour/scenario described by the "program" of this activity: use your previous concepts and behaviours to find a new concept or a new behaviour, and if you are sure it's new, name it and decribe it. if we can name it, it can exist in mind. job done!

I think naming is an important step.

The name is not as important itself as having the label by which to find the concept thereafter. Many researchers had been sloppy and had destroyed evidence of antibiotics before one researcher gave the ring of contamination a name and thus was able to study the effects separately.

I think Discovery has a lot to do with the Declarative Memory System, and how we find memories we know that we have. By labelling an experience as a "What" by giving it a name, we open up the What based index for finding it, so that we are not limited to episodal memory (when and where) for finding it. It becomes not that thing that happened after we flipped the finger at Fred, but instead "Antibiotic activity" in our garbage.

It may have been because Fred complained about us leaving our garbage out on the counter all weekend, that we took the time to really look to see what had happened to our cultures, but giving it a name such as "Anti-biotic" means that whenever we see these clear rings again we will look to see what contaminated the sample, and study that.

Yes Graeme. The name itself is not important,but it should be an unique reference! Moreover, it can have multiple meanings in multiple contexts, so semantic relationships between concepts, behaviours and values are also important.

Assuming a competitive nature of declarative and nondeclarative memory systems, don’t we get closer to “a dynamic of consciousness” as the initial question presented discovery?

So, here we might consider a distinction between the internal indices of a species' acquired knowledge (the present forum being an exchange of this, if you will), and external "unknowns". So, then, is the problem of "discovery" resolved by the agency of intelligence -- the behavior of factoring in what is recognized and nameable with what is unrecognized and has no name? Curious as to how living tissue appears to thrive, "one way or another", as a result.

I "discovered" that dynamic aspects related to consciousness and declarative memory system are well introduced in another topic.

Excluding Discovery from such a dynamic makes it a usual learning process (ie Observing, Classifying, Measuring and Inferring).

As to "naming" in the discovery process, one can cite Higgs giving his own name to a particle he never believed it can be discovered!

I think that really discovering is an essential part of human being or artificial agent's attempt to consruct the theory (explanation) of the surrounding world and it (he) is trying to introduce all elements of this theory (or belief) signature, that are: relations, object its names, signs etc.

.

You can see inquiry based learning to teach scientific activities in classroom and you will find out that : „discovery learning is an active process of inquiry-based instruction in learning...”

http://halshs.archives-ouvertes.fr/docs/00/19/06/09/PDF/A101_Edelson_etal_99.pdf

Quick true story: a zoo keeper explained that the lemurs were not thriving in their new natural habitat inclosure until one day, a keeper tripped over their food bowl, scattering seeds, cut fruit, etc. mixing the stuff with the dirt and under logs and in bushes. The lethargic lemurs went crazy as they searched out and ate all the food, even ate the vitamins. Now the keepers hide most of the lemur food, especially the vitamins. Does this really extrapolate to astrophysics and quantum mechanics?

I consider that black hole and black matter may explain this story in astrophysics, because the lemurs were not thriving in their natural habit. of course, a keeper tripped over their food bowl, scattering seeds, and cut fruit can seem as some matters enter into universe, the lemurs do not seek food will drop into black hole. these food selected by the lemurs, will become a part in black matter.

It can be by chance or coincidentally, or intentionally If it is coincidentally, firstly maybe the purpose of the discovered thing not identified. If it is intentionally, the purpose of the discovered thing already known or identified.

I think the Lemurs case is a point in motivation, food prepared and presented in the same old food bowl from day to day is boring, but food scattered across the cage, some of it hidden, requires that you discover it, which is no longer boring. In the wild, you can eat from a single banana tree, or you can search for variation in diet, and in doing so, get a wider range of nutrients. It is entirely likely that lemurs are wired not to eat too often from the same tree.

Think of the lemurs being something like adolescents, who get bored if things are same way all the time. The difference is that lemurs can get bored to death, since they won't eat boring food. This might suggest a lesson for education, in that presentation of knowledge might bring out the discoverer in students that would otherwise be bored silly with normal schooling. Kick over the food bowl and let the students search for their vitamins.

The lemurs anecdote made me consider the development of the human neural complex, both from childhood to maturity, but, also, evolutionarily, that, as the sensory array that develops would suggest in that relationship, there are qualities of response, such as"discovery", and "belief" that are difficult to quantify in the discussions about somehow creating these qualities in AI.

Since everything is around and already present I like to think of "discovery" as the first person to recognize, comprehend, analyze, and communicate to others their new understanding of how things thought they worked or functioned yesterday as compared to after the communication. I believe it is a multi-facted uncovering of tomorrow's unknowns.

@william: To discover you also need to first poke the anthill with a stick to see what comes out.

From a review of responses to my question, so far, I see a trend to recognition of an somewhat indecipherable element of melding of thinking and feeling, pairing "learning" with the the concept of "discovery" in living and artificial systems. As the human-centered entities Watson and IBM learn from doing "their thing" in ever more complex data systems, hopefully, the simple questions do not go unrecognized, unasked, unconsidered.

Whether the simple questions are unrecognized, unasked, and unconsidered depends very definitely on the granularity of the search algorithm spanning the search space these tools are applied to. The higher the granularity, the faster the search runs, but the more options are left unconsidered. These days searches have variable granularity starting out as highly granular, then cutting down the granularity as they progress, in the assumption that most similar answers will be local to each other. This only really works in pre-organized search spaces.

Hence, discovery, in the abstract, may depend upon the nature of the search.

Let's compare discovery with innovation. Any innovative solution is built on an obscure (i.e., infrequently-noticed or never-before noticed) feature of the problem. If the key feature is commonly noticed, then there would not likely be a discontinuous leap or aha moment. The key feature needs to be obscure for a discontinuous noticing and solving to take place. If you think a discovery is similar to an innovation, then this description might be helpful. This is the basic of a new cognitive theory of innovation by Tony McCaffrey published in Psychological Science "Innovation Relies on the Obscure" and recently published in July/August issue of Scientific American Mind.

We have reflected with a colleague (Jean Sallantin) about theory construction considered a synonym of scientific discovery. This obviously does not include artistic creation. We both believe that there are many similarities with learning, and in particular both are mainly serendipitous events. This calls for a profound revision of our practice in education (see eg: http://www.youtube.com/watch?v=zDZFcDGpL4U ) but this is another story ... I include our contribution to the Encyclopedia of the Sciences of Learning.

Sorry, as a novice to RG I find it difficult to apload a short paper ... trying again ...

I would like to add the concept of Surprise as a learning stage, involving the transfer of information between implicit and declarative memory, and the differential in latency between the directions involved. Implicit memory is routinely transferred to declarative memory but due to limits in the transfer to Short-term/Working memory, there is a bottleneck that limits transfer in that direction. Declarative memory is transferred to implicit memory, as part of consolidation, but part of that is a 2 year latency, that slows down the integration of knowledge levels.

As a result of this limited cross pollination between the implicit/declarative systems, information can be known at one level of memory and not at another. When such a secret gets revealed to the other level of memory, there is a moment of discovery when other previously known knowledge gets reassessed. This Surprise makes formerly obscure elements discontinuously observable. And can cause a cascade of new interpretations that is actually physically painful, the so called AHA phenomena.

Re: Stefan: In the cognitive sciences, by examining origins of the artifact naming of a particular event, we should seek to limit or pare down on the number of meanings, and arrive at descriptors that are more useful for the tasks at hand. This concept that has the label "discovery" -- seems to be at the heart of a great field of intelligence-related events -- are these events related in some way (causality: does "discovery" mark the result of a process, or is "discovery" a behavioral marker of something that is a continuum?) My question is framed to evoke a survey of expert opinion about what we understand as the thing that is "discovery". The question arose from another discussion on this forum regarding the nature of "consciousness". It seemed to me that one particular marker of "consciousness" might be what we understand to be "discovery". Frankly, it is quite apparent from a variety of responses that it was, what I call, "a good question." I am an educator, and am constantly aware of this state of mind in my students, but am also experiencing states of being that I call "discovery", a topic of interest to education as a motivator. As to AI, could "discovery" be such that it can be an existing element in the field, as yet unrecognized? We seem to mimic this behavior when we name machines "Discovery". So, to qualify, I am seeking as many definitions as possible -- not just for me...

There are several literal meanings of word *discovery* like a few as given below:

a.To be the first, or the first of one's group or kind, to find, learn of, or observe something not found or observed earlier.

b. Finding out or ascertaining something previously unknown or unrecognized; as, Harvey's discovery of the circulation of the blood.

c. The process of learning something that was not known before, or of finding someone or something that was missing or hidden.

However, the context in which you have used discovery amounts to *scientific discovery*.

According to Popper:

“The question how it happens that a new idea occurs to a man… may be of great interest to empirical psychology; but it is irrelevant to the logical analysis of scientific knowledge. This latter is concerned… only with question of justification or validity (The Logic of Scientific Discovery, 1959). Thus, scientific method is deemed as inapt to generate new ideas. However, scientists are under pressure for *novelty* i.e. to generate new ideas. But, novelty (discovery) is hard to accomplish.

A mode of discovery as suggested by Györgyi is seeing what everybody else has seen and thinking what nobody else has thought.

It is dilemma of our scientific methods data or experimental data are theory-laden, therefore, we end up what is already known (anticipated results). I don’t know much about much hyped experiments being conducted at CERN, if people there are interested in a particular aspects, it is possible that their researched should be highly focussed and lot of data of no concern of theirs may have gone waste, if a proper protocol was not at place to record as much data as possible whether relevant or irrelevant.

However, in a focussed research unexpected experimental results are generally let go and emphasis is placed on expected results. But, in fact these are unexpected results which lead to important findings and something new is discovered.

Charles Peirce calls it *abduction*. It acts as a powerful general mechanism by which novelty (discovery) is accomplished. It is why a number of inventions and scientific discoveries are either results of some accident or some unexpected events in the process of experiments. It should be noted that even during normal course of scientific process, additional possibilities (beyond focus) arise and noticed earlier but not recognised as important. Therefore, noticing unexpected and recognising it is the key to discovery.

If an artificial brain is trained to notice unexpected or allowed to err like humans, certainly, it may make discoveries.

As to conducting a more formal survey, rather than entertain a discussion, I have to say that I lack the descriptors for this on this kind of forum, so wide is the sample (for some reason, my thinking blends so many diverse disciplines -- enormously grounded in erudite and esoteric technical knowledge such as neuroscience, and maddeningly dynamic and political as "education", I lack the resources -- I'm a poor classroom teacher, for the time being, nestled between deep digital canyons of student essays. If anyone has the resources, be my guest. Hopefully, this question has generated a few of those descriptors. I have to commend RG for keeping so many broad and narrow discussions moving along -- following them has brought me to ask my question. At the bottom of my essential question is a need to gain some perspective on the diverse qualities of consciousness by stepping aside and asking some other "simple" questions. It was quite gratifying to read that the concept of "Surprise" has been introduced to this thread, because (great minds think alike?) that was going to be my next ventured question. My own research project has to do with how individuals are motivated to learn -- and I am very interested in the relationship between "feeling" and "knowing". "Discovery" seems to be a bridge between. And it all seems to develop along the lines of maturity of neural systems.

I can see only 34 of supposedly 57 answers, so I apologise if I duplicate something already here.

My initial thought went in a rather different direction. Inglis wrote a series of papers on the information primacy model, which assumes that reducing uncertainty is a basic motivation of animals, but with a lower priority than, for example, hunger. In a 2001 paper in Animal Behaviour, Inglis et al presented a fuzzy logic model, in which they assumed that "Uncertainty is reduced by visiting a location and gathering fresh information about both the stimulus characteristics of that location and the food it contains."

Skimming through Inglis' papers again, it is not immediately clear to me whether he makes a distinction I found in a different field of research, between surprise and ambiguity. In a review in Behavior, Inglis cites Staddon who clearly makes that distinction: “Animals learn only when something violates their expectations, or when they have no expectations (as in a novel situation).” However, Inglis seems to focus on violation of expectation, something that Itti and Baldi have quantified in their definition of Bayesian surprise, which is a function of the difference between prior and posterior probabilities.

I have tried to work out what an equivalent definition of ignorance would be. A single level Bayesian model has to have priors for all possible events. However, you may not be sure about priors, and may need to represent your expectation as a distribution of probabilities. If you write a programme that produces 0s and 1s with a fixed probability, and ask me to work out what that probability is, I can't offer a very precise estimate after a few samples. I can only offer a range of possible probabilities. I have been told that range is known as a distribution of hyperpriors.

Relationships among events are represented by likelihoods, and those need to be estimated from empirical data as well. Therefore I can know likelihoods only with limited precision. I suppose that means I need hyperlikelihoods.

That description of ignorance assumes that at least I know the possible outcomes, and I am merely uncertain about the probabilities of outcomes, which would be ambiguity. Being uncertain only about the outcomes, but knowing the probabilities precisely (for a given threshold value of precisely), would be risk. Pushkarskaya et al defined a more profound state of ignorance, where even the outcomes are unknown, as sample space ignorance. I don't have the maths to work out whether in a hierarchical Bayesian model this could be described as a wide distribution over a suitable space of conceivable outcomes. I think that should be possible.

A reduction in ignorance then might be quantified by how much narrower the distributions of various probabilities become at several levels of a hierarchical Bayesian model. I definitely don't have the maths to propose a formula. I can only offer a hypothesis that a sense of discovery is related to the rate at which ignorance decreases. Even a small reduction in ignorance might induce an AHA experience if the insight is sudden enough. That hypothesis is necessarily rather vague in the absence of a quantitative measure of ignorance.

Regards

Robert Biegler

A very tough question.

1. every learning process contains elements of discovery. People and higher animals have the knowledge instinct driving us to learn, or to create mental cognitive representations. Animals learning is limited by inborn predispositions and by behavior of surrounding animals, mostly conspecifics. Learning from books or teachers is a process of creating personal mental cognitive representations from language representations. Language representations exist 'ready made' in books and in surrounding language.

2. But cognitive representations do not exist 'ready-made' in any experience. To make real discoveries, one have to combine the knowledge instinct and language. See:

Perlovsky, L.I. & Levine, D. (2012). The Drive for Creativity and the Escape from Creativity: Neurocognitive Mechanisms. Cognitive Computation, DOI 10.1007/s12559-012-9154-3.

Levine, D.S. & Perlovsky, L.I. (2010). Emotion in the pursuit of understanding. International Journal of Synthetic Emotions, 1(2), 1-11.

Levine, D.S., Perlovsky, L.I. (2008). Neuroscientific Insights on Biblical Myths: Simplifying Heuristics versus Careful Thinking: Scientific Analysis of Millennial Spiritual Issues. Zygon, Journal of Science and Religion, 43(4), 797-821.

Let me know if it is helpful

Leonid

I think there is something missing in this discussion so far, and that is the role of implicit memory as a storage of priors. The implicit image of a stimulus, is therefore a model of that stimulus from priors. It is the declarative memory image, that sets up the expectation of a new synthesis, that sets the probabilities for the new image that will exist after processing is complete. Surprise is when both the implicit and declarative images turn out to be wrong, and a new synthesis is achieved.

The surprise is born in the need of the brain to resolve conflicts in the data between the two models. As a prelude to consciousness, the brain tasks itself with the need to resolve the differences in the two models, and that task being neither implicit nor declarative in nature, can come to a third solution, which is neither of the other two.

When that synthesis results in a cascade effect, causing a reshuffling of explicit assumptions, the AHA phenomena is born. It is not only the new image, but the shock of having memories realigned that denotes discovery.

I think that it is instructive that one of the references would be to the difference between careful thought, and heuristics, because to some extent the so called careful thought, is equivalent to the declarative model of the stimulus, and the heuristic is equivalent to the third model that defines the AHA moment. Heuristics often are discovered because there is an incompatibility between so called careful thought, and scientific, observation, and so a new fusion is required to reconcile the differences between the models. To imply that they are somehow less carefully thought out, than some other approach, is just another example of the laplacian fallacy, where it is assumed that a proof from first principles is always better than a derived proof, even if the first principles used don't fit the original observations.

Dear Graeme Smith,

I like what you wrote, but I am not sure I really understood you. Let me try to formulate your thought in terms of my model and may be you will comment, which part of it makes sense to you.

My model combines two hierarhies of language representations and cognition representations. Language representations exist 'ready-made' in the surrounding language and can be learned early in life throughout the hierarchy. Cognitive representations (implicit images) do not exist 'ready made' in the surrounding world. The reason is that whenever one looks in any directions there are hundreds of images, and only a small part of them corresponds to 'features', or objects, or scenes worth remembering.

The number of combinations of features, or objects is larger than the number of all particles in the Universe, say combinations of 200 features is 200 power 200, this number is many many times larger than the Universe, practically infinite. Therefore the question arise which "implicit images" or cognitive representations we learn to remember?

My model of the dual hierarchies suggest that most of the time we learn cognitive representations in correspondence with language representations (ideas) existing in surrounding language-culture. Rarely a creative process occurs in which a new "implicit images" or cognitive representations is created in unconsciousness, and it is so powerful and important that it riches consciousness, is expressed in language, and is gradually accepted by surrounding culture and becomes a part of language.

As I understand you, heuristics are language representations (ideas) learned from surrounding language, but not deeply thought through, there are no corresponding 'implicit images' or in my model, the corresponding cognitive representations are vague and barely conscious. Careful thoughts on the opposite correspond to 'implicit images' or cognitive representations that are crisp and conscious, and they can be clearly and consciously expressed in language.

Often people think in terms of heuristics, ready-made language representations without real cognitive understanding. Rarely we think creatively so that language ideas correpond to original cognitive representations.

Does it makes sense?

Leonid

@leonid: Let's take it back to the original system you are trying to model.

Lets agree that when we look in any number of directions, there are hundreds of images and only a portion of them correspond to 'features', worth remembering.

Let me postulate the implicit case, where a sampling of previous (Prior) images have taken on a correspondence to 'features' worth remembering. These images may have no correspondence to language, or cognitive images, but do have 'importance' enough to be recognized and remembered in a special memory type called 'Implicit'

In essence they have importance but no semantics, and so are not noticed to have meaning.

Now, let us assume that with processing, and often in response to language there is another type of image, one with semantics, and a derived meaning that can either be formed as a result of processing, or of re-evaluation in light of language constructs. This image type is seated in the Declarative Memory, and is subjective at first glance.

I believe you call this the "Cognitive" image, but I could be wrong.

Finally let us assume that there is a process, some call reasoning, that further processes the Cognitive image to find out more information or semantic meaning than would be available with the wild-type cognitive image.

Your Careful thought would seem to capture the reasoning process, but what I find objectional in your assumptions is the idea that heuristics, because they often take on the cachet of metaphors, are somehow still cognitive images without reasoning. Instead, I suggest that heuristics are only created as a result of reasoning, when the semantics of the implicit and explicit images are so different that a cognitive dissonance is created, that demands another approach to the problem, that may lie outside the language constructs and therefore may need to be expressed in the less obviously reasoned metaphorical approach called heuristics.

One way of looking at heuristics is that they are examples of systems that work in the proposed manner, rather than conclusions directly of reasoning. In essence the Reasoning has stalled, because more is known than can be modelled in the language constructs/reasoning patterns hitherfore determined. In short "Careful Thought" has broken down, because it is not flexible enough to encapture the necessary detail.

I have noticed that to "Classically Trained" minds, this thought pattern is opaque, often impossible to get across, because they have no language constructs to build it out of, and are unwilling to accept any substitutes. I have a hope that you will not be this hidebound but wait eagerly to hear your response.

LP: Graeme, I like what you say and I published few papers where I tried to make it a bit more precise. My comments below can be a step towards matching these papers to your thoughts, by emphasizing where could be difficulties in this process.

GS: Lets agree that when we look in any number of directions, there are hundreds of images and only a portion of them correspond to 'features', worth remembering.

LP: to be more specific, there are no hundreds of definite images, but a continuous undifferentiated “percept.” It has no useful “portion” unless an agent finds it purposefully. Nothing worth remembering could appear by chance. This is fundamental.

GS: Let me postulate the implicit case, where a sampling of previous (Prior) images have taken on a correspondence to 'features' worth remembering. These images may have no correspondence to language, or cognitive images, but do have 'importance' enough to be recognized and remembered in a special memory type called 'Implicit'

LP: OK, we can assume that an evolution resulted in features = implicit images, but how a person would create an implicit image out of the near infinite number of raw percepts?

GS: In essence they have importance but no semantics, and so are not noticed to have meaning.

LP: likely, this is impossible, what has no meaning does not evolve. You are right: the difficulty is how to make more meaning out of less meaning.

GS: Now, let us assume that with processing, and often in response to language there is another type of image, one with semantics, and a derived meaning that can either be formed as a result of processing, or of re-evaluation in light of language constructs. This image type is seated in the Declarative Memory, and is subjective at first glance.

I believe you call this the "Cognitive" image, but I could be wrong.

LP: OK

GS: Finally let us assume that there is a process, some call reasoning, that further processes the Cognitive image to find out more information or semantic meaning than would be available with the wild-type cognitive image.LP: OK, and the mechanism of this process IS of interest)

LP: Graeme, I’d prefer if you split your thoughts in short understandable sentences. Now I’ll have to do it the way how I understand. Please correct if you disagree.

GS: Your Careful thought would seem to capture the reasoning process, but what I find objectionable in your assumptions is the idea that heuristics, because they often take on the cachet of metaphors

LP, this might take a half page discussion. I disagree, heuristics are opposite of metaphors; and there are no scientific definition of a metaphor

GS:, are somehow still cognitive images without reasoning

LP, When Archimedes cried out “Eureka” he meant an original discovery; today “heuristics” means the opposite: a ready-made language rule used without understanding, that is without cognition.

GS: Instead, I suggest that heuristics are only created as a result of reasoning

LP: when they have been created the first time, it was “Eureka” a cognitive-language creative process of discovery; when they are used without thinking they are hollow words, sometimes with aura of usefulness but also irrational according to Tversky-Kahneman, and this is called today “heuristics”),

GS: when the semantics of the implicit and explicit images are so different

LP: how there could be any difference? any explicit image is a product of implicit images and language

GS: that a cognitive dissonance is created

LP: that would be great but it is exactly what is difficult to understand, I think it is the greatest mystery,

GS: that demands another approach to the problem, that may lie outside the language constructs

LP: how to explain creativity? – I’d love to know, where could it lie? In my papers I try to explore this and developed a math. model of this process, this process goes “from vague to crisp” and from “unconscious to conscious”; and it was experimentally proven to occur in the mind.

GS: and therefore may need to be expressed in the less obviously reasoned metaphorical approach called heuristics

LP: this statement is difficult to make precise enough: what exactly is a mechanism of metaphor creation? It is true that some people have this ability, and this IS what we would like to understand.

GS: One way of looking at heuristics is that they are examples of systems that work in the proposed manner

LP: metaphoric, or hollow-language? If metaphoric, we do not know how it works,

GS: rather than conclusions directly of reasoning. In essence the Reasoning has stalled, because more is known than can be modelled in the language constructs/reasoning patterns hitherfore determined. In short "Careful Thought" has broken down, because it is not flexible enough to encapture the necessary detail.

I have noticed that to "Classically Trained" minds, this thought pattern is opaque, often impossible to get across, because they have no language constructs to build it out of, and are unwilling to accept any substitutes. I have a hope that you will not be this hidebound but wait eagerly to hear your response.

LP: In my papers on creativity I tried to address some of these, also papers on interaction between language and cognition address it.

@leonid:

GS: "In essence they have importance but no semantics, and so are not noticed to have meaning."

LP: "likely, this is impossible, what has no meaning does not evolve. You are right: the difficulty is how to make more meaning out of less meaning."

I didn't say that they had no meaning, I said they are NOT NOTICED TO HAVE MEANING in the sense that declarative semantics are meaning. Consider them "Out of Band" information not included in the declarative function. The semantics of the implicit memory are all tied up in context, where as declarative memory declares the semantics alongside the information itself.

LP: Graeme, I’d prefer if you split your thoughts in short understandable sentences. Now I’ll have to do it the way how I understand. Please correct if you disagree.

Ah here we get into one of the limitations of language, any truly abstract concept, must perforce either be a long-run-on sentence, or it would be concrete, and easily chopped into little bits.

GS: when the semantics of the implicit and explicit images are so different

LP: how there could be any difference? any explicit image is a product of implicit images and language

Sorry should have used declarative instead of explicit there. A hold over from before my realization that a portion of the philosophy was taking explicit to mean conscious.

Here we are getting into functional difficulty, in that declarative memory has an explicit role that is different from that of implicit knowledge and as a result has indeed a completely different image than the same element in the implicit memory, not that there is necessarily even a correlation between the storage elements in one memory type and the storage elements in the other. In fact the transition between the two types of memory is more than just a language interface.

GS: One way of looking at heuristics is that they are examples of systems that work in the proposed manner

LP: metaphoric, or hollow-language? If metaphoric, we do not know how it works,

Well, we might not know exactly how it works. For instance the heuristic of Evolution, Evolution is a natural effect that we might not know exactly how it works.

The Evolution Heuristic is a method of estimating what the results would be if a system was evolved in a certain manner. We know exactly how the heuristic works, and that it seems to work in much the same way that Evolution itself works, but the details are fuzzy because no one has proven how evolution works.

The heuristic steps in to supply a rationale for why processing is done in a particular manner, instead of another more classical manner.

Cue the orchestra. That is, "music" in the broadest sense, in one form or another, generated as a response to an environmental stimulus, a spontaneous artifact created by behavioral activities, hence -- "AHA", a vocalization associated with "discovery". So, might endocrine action effect be involved? One might purposefully anticipate a discovery, and one might be taken by surprise. Or a little of both as the mind resolves itself to accommodate a new "reality'. Listen to the tone of the "AHA". Sounds like intake, then exhale. Stress/release stress.

Music is an interesting suggestion. Aristotle listed the power of music among unanswered problems. Darwin thought is is the "greatest mystery." May be there is something to your suggestion that would explain both music and discovery?

The greater the abstraction of the stimulus -- metaphorical language, for example -- the greater difficulty accounting for the resultant response. "Learning", for example, as measurable behavioral response has a layer of complexity due to the inherent abstractions and ambiguity of linguistics! But considering the tactile nature of generating a sound as response to sound -- "echoing" for example -- would seem to offer some possibility to isolate the concrete nature of whatever is going on during the act of discovery. Harmonics? I bring up the Helen Keller story only to suggest that there is most assuredly some connection with touch and learning, and that there are quite a few tactile learners. I would suggest that language evolved from music -- again, very broadly, authentic body language, especially vocalizations -- and so to go around the problem of complexity of the variety of stimulus from the environment, consider that "tonality" may offer better clues to isolating and identifying the basis of this type of event -- at least this would seem more primal an agent and to the responder.

Leonid:

I must confess that I just discovered the abstracts of your research on music. Certainly resolved some dissonance for me, as the processes of venturing into uncharted territories of AI and neuroscience can be pleasing but daunting.

Thank you Jonathan

I have reached a conclusion about discovery: with each click of the Enter button, and, as the screen is refreshed, a shot of sweet discovery is delivered to the brain; equal only to "hunger" and "danger", this cheap and plentiful fuel for "curiosity" drives humanity into the 21st century.

Most of under 3 year children activities are discoverung.

The brain is an anticipation machine. To anticipate situations we need a lot of information for a statistical analysis. For that we have to prospect for those informations. Therefore we are curious.

Music and emotions are siblings. Archimedes "Eureka" was also a very emotional shout. We should also note this dimension not only the cognitive one.

There is an ability of the mind to predict failure or error.

There is an emotional charge to such prediction, and this causes cognitive dissonance to form creating an emotional tension, or stress, which when released gives us a pain to pleasure transition. Thus the discovery moment brings us a stress/release emotional charge that results in extreme cases with a verbal marker such as A-hah or Eureka!.

Part of the strength of the emotion has to do with how long the tension has developed. And how important the answer is to the individual.

Somrthing new is interesting. Discovering a new knowledge that is still uncover is always interesting.

The history of science is full of cases that can be classified as discovery. The original concept of discovery was widely used on its origin. These days, it is considered as a special form of creativity, or one of many techniques for developing the creative potential of an adult, which combines perseverance, intelligence and sense of observation.

@ Graeme Smith Agree and  add  to the strength of the emotion the feeling of how beautiful the solution is. There can be an aesthetic component also.

@Colin

Elegance, or the beauty of a solution, depends on it's relative simplicity in comparison to former solutions. this is related to a memory cascade reaction, which is quite painful, and represents a review of previous work. The elegant solution such as Einsteins e=mc2 represents a remapping of previous work, that results in a much simpler solution. In essence the old long answer is ugly, or painful, the new simpler answer is beautiful, and elegant.

I like to contribute some metaphorical sentences: If a discovery takes place a puzzle is solved. The cognitive dissonance disapears and a mental representation of the problem becomes clear. The gap is bridged. 

This situation relaxes and results in a posive emotional state.

discovery is when the problem is rephrased and hence a solution can be found.

So intstead of solving the original problem, discovery is to identify the problem behind the problem and solve this one. Often easier than the original problem. 

Good question.. Please share me the best answer might you trust...

Regards…

Good question.. Please share me the best answer might you trust...

Regards…

the question should use curiosity instead of discovery. the latter is a consequence. humans used to explore environment and found food or new useful things so reward was very appreciated, giving satisfaction and pleasure. and it is now in our dna.

Alan Turing wrote that the machine can be a carrier of the Soul, as well as the human body. So the car can have all the features of the human psyche. It should be noted that the machine can not simulate the human psyche. his emotions, intelligence, thirst for discovery, etc., and really possess them. http://2045.ru, http://lc.kubagro.ru/aidos/Credo/Credo.htm