Defining AGI
Gotcha! In this post, I am not going to define AGI. Its very first letter A stands for "artificial", which implies that it is a model for what the last letter I stands for - intelligence. So instead of defining AGI, I will try to accomplish a smaller task and define only intelligence.
What is a definition?
According to [1], "definition" is "the act of defining, or of making something definite, distinct, or clear."
For now, let's focus on the word "distinct" and keep it in mind. Why do I think that it's the key word about definitions? Simply because any definition should encircle phenomena that qualify and filter out phenomena that don't. The key role of a definition is to differentiate the instances of the defined class from the instances of the other classes.
Let's now see what we differentiate when we define intelligence.
Definitions of Intelligence
I will analyze some definitions from [2]. Below I will list them all but will comment only those that introduce new terms or angles.
Collective definitions
1. “The ability to use memory, knowledge, experience, understanding, reasoning,
imagination and judgement in order to solve problems and adapt to new
situations.” AllWords Dictionary, 2006
The very first definition from that paper uses a lot of common suspects, usual components of definitions of intelligence. Analyzing the following definitions, we will see many of these words.
Memory and knowledge are related because the former stores the latter and allows for retrieval of the pieces of knowledge as needed. But note that we are not so interested in the same joke repeated many times or the same fact encountered more than once. The number of repetitions may be a fact of its own but the underlying piece of knowledge is stored in memory once. We store similar facts only if they have some details that are significantly different. On the other hand, it may point at the modularity of facts which are stored as a collection of separate but connected pieces.
We may use a formula metaphor for facts. As components play different roles in a formula, constituents play different roles in sentences, which may represent facts in memory. Note that each constituent may participate in multiple facts creating an interconnected mesh of knowledge.
When we query memory for some constituent, we provide keys to enable the search and also validation. This is where the formula metaphor no longer serves us well. In formulas, we calculate the missing component based on the provided ones, for example, a distance is the time multiplied by the speed. When we answer questions, we use the substitution method. Given all the known facts, we check if any of them has all the constituents provided in the question in the proper roles. We may find a good fit, a weak fit, or no fit at all. Calculating a perfect fit is not good when that fit is unavailable.
Note the use of comparisons and the importance of establishing differences or "thesameness" in the process described above.
Experience and learning are also related to knowledge. We are born without knowledge but fully equipped with the perception machinery. Every living organism has a boundary separating it from the environment and differentiating "self". We have needs and we register the constant exchange of heat, food, air, information, etc. as either satisfying those needs or increasing them. We introduce the labels "pleasure", "pain", and others to differentiate our estimates of the outcomes of those exchanges.
We learn not only about how the environment can affect us but also how we can affect the environment. We notice many different dimensions along which we can make or experience a difference. We learn about actions affecting those dimensions. We create and test multiple hypotheses along the way. Sometimes we may act on the basis of false ones but we are good at quickly eliminating the inconsistent ones. The core scientific principle used for analyzing experiments and theories respects the complicated effects of multiple dimensions and requires changing one and keeping others equal.
Note again the respect for differences.
Understanding is also related to knowledge. It relies on the existing knowledge in the process of acquiring new knowledge. Encountering new phenomena, we try to categorize them according to the concepts we already know. A child who never saw a cow but saw pigs upon encountering a cow for the first time may exclaim, "A pig with horns!" The horns will not be the only differences observed and remembered along with a new label "cow".
A new phenomenon that is miscategorized at first will be treated according to the knowledge about that mistaken category. Observing differences and updating actions to deal with the new phenomenon we learn how we can interact with all the known dimensions packaged differently in this new category. At some point, we become confident about our understanding. Like Newton was confident about his complete, final understanding of the laws of nature. The ultimate knowledge is unreachable, at any point in time, we proceed with the best hypotheses we have so far.
Please note that phenomena are multi-dimensional. We understand apple when we know what to expect from it along those multiple dimensions. What will happen when I eat an apple? throw it? draw it? peel it? put it in ground? keep it in a box for a year? What will be destroyed if I press an apple against berries? Understanding is multi-dimensional and never complete.
Reasoning, as I understand it, applies our current knowledge of concepts/categories and actions affecting them to explain the observed facts and fill in the blanks. Observing a dead body and a number of people who in theory could commit the murder, a detective tries to build a consistent theory about how suspects behaved and whether or not each of them is innocent. The substitution method is in action again. The clues are used as constraints to filter out innocent people from the set of suspects. Alibis work because there are known limits on speeds, forces, etc.
Imagination is a huge topic. It can be about the mundane business meeting and what arguments will work best in favor of our proposal. It can be about a hypothetical subatomic particle that may stand behind the gravity holding galaxies together. It can even be about impossible, fantastic worlds/creatures/events.
"What if" is another useful tool of scientific research. Mental simulations allow us to run experiments and based on the existing knowledge figure out likely outcomes. You can view imagination as the experience and learning only virtual.
Judgment is similar to object recognition but applies more to actions and the categories used are more often moral or emotional. Note that judgments are easy to make only in simplified scenarios. Take into account more actors, broader contexts, and more time frames, and no codex will be able to list all the rules to take all the possible nuances into account.
Solving problems is the most frequent phrase encountered in definitions of intelligence. However, I consider it to be too general to enable any practical implementation. Different types of tasks are solved using different algorithms. Besides, how do we decide which type of task we have at any time? How do we decide what can and what cannot serve as a solution? How do we determine which solution to prefer? How do we decide that the task is solved?
Think for a second about the phrase "the second best solution" or the phrase "two birds with one stone". Whatever intelligent system we implement, it should be capable of coming up with such solutions as implied by those phrases. And more.
Adaptation continues the topic of "self vs environment". It is about forecasting the current or future effects of the environment and counteracting them with our actions. On the one hand, adaptation requires recognizing actors in the environment, their strategies and actions, our resources and capabilities, and selecting proper counteractions. On the other hand, we can recall task-solving and consider adaptation as another task to solve.
"New situations" are often mentioned without any explanation or analysis. I think that even the word "new" alone deserves much more respect. According to Heraclitus, not only a river and man but all objects are unique across time and space. But for practical purposes, we consider some things interchangeable. But the question remains, "What makes a situation "new" or "already known"?" It seems we decide that by some process of recognition, which requires comparisons. Should I reiterate and mention that "new" in some sense is the same as "different"? (Pun intended - the same as "different")
2. “The capacity to acquire and apply knowledge.” The American Heritage Dictionary, fourth edition, 2000
This definition has nothing that we did not discuss above. When you see such a comment it is up to you to decide if this is so.
3. “Individuals differ from one another in their ability to understand complex ideas, to adapt effectively to the environment, to learn from experience, to engage in various forms of reasoning, to overcome obstacles by taking thought.” American Psychological Association
Complex ideas and their understanding rely more on the existing knowledge than on analytical skills. You may consider the complexity of ideas as a skyscraper - building each additional floor requires the completion of the previous one. Ideas like actions have prerequisites and apply to some domains. The same domain may be addressed by multiple ideas, which are somehow different from each other. Complex phenomena may be looked at from different angles or along different dimensions.
Overcoming obstacles can be considered as taking into account additional constraints or as the need to invest more resources to reach a goal. It is a kind of adaptation discussed above.
Thought may have many definitions but I propose to view it as the process of applying the substitution method. Solving any task may involve solving subtasks or lead to solving other tasks. Entertaining any thought in your head is like saying a sentence. With each word from that sentence, we may play the association game or ask any questions about it. As a result, we hop from node to node in memory.
4. “The ability to learn, understand and make judgments or have opinions that are based on reason” Cambridge Advance Learner’s Dictionary, 2006
This definition has nothing that we did not discuss above.
5. “Intelligence is a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience.” Common statement with 52 expert signatories
Thinking abstractly or generalizing is the opposite of specialization. We may think of preparing tea. Or we may think about taking spring water and boiling it in a clay teapot to the stage of "the string of pearls" when we will add young tea leaf buds to produce a delicate white tea. Do you see the difference? Introducing details is specialization. Forgetting differences is generalization. Both processes rely on the use of comparisons and respect for differences.
6. “The ability to learn facts and skills and apply them, especially when this ability is highly developed.” Encarta World English Dictionary, 2006
This definition has nothing that we did not discuss above.
7. “. . . ability to adapt effectively to the environment, either by making a change in oneself or by changing the environment or finding a new one . . . intelligence is not a single mental process, but rather a combination of many mental processes directed toward effective adaptation to the environment.” Encyclopedia Britannica, 2006
A combination of many mental processes is not wrong but definitely lacks specifics and details to enable implementation.
8. “the general mental ability involved in calculating, reasoning, perceiving relationships and analogies, learning quickly, storing and retrieving information, using language fluently, classifying, generalizing, and adjusting to new situations.” Columbia Encyclopedia, sixth edition, 2006
Perceiving relationships implies the existence of multiple forms of relationships and differences between them. They may involve different properties - "further" or "earlier", to name a few.
With analogies, everything is more complicated. Consider the shades of red - for example, ruby and scarlet. To most men in general both will be just red and similar/analogous as can be. But to a professional designer, they will be different. Who decides? If a tired person wants to sit down, differences in the chair color or material or shape will play no role whatsoever. They will to an interior designer.
"learning quickly" - did humans arrive at Newton's laws "quickly"? did humans accept those laws "quickly" after their discovery? did humans "quickly" realize their flaws? Who determines what is or is not "quick"? Just as with analogies, speed may be context- or task-dependent.
"using language" - language is both a tool and a product of intelligence. Many think that its role is to communicate information. I disagree with that. It is not language that communicates information or the major part of it. The role of language is to guide readers'/listeners' attention to relevant phenomena in context. Those phenomena are differentiated from the context using the same substitution method. Information is not provided in language, it is picked up by perception. As you can see, language uses the same core algorithm.
Classifying is all about differentiation. To understand why, consider first wrong theories.
There are theories of classification/recognition (prototype and exemplar theories) that rely on analogies/similarities. Fodor claimed that categories/concepts are atomic, independent of each other. If concepts are atomic and we recognize objects by comparing them to some idealized representative (prototype or exemplar) of that concept, how long will it take to check all known concepts? Is it feasible in the real-time?
With respect to concepts, everything is recognized in comparison. Consider the game 20 Questions and note its performance in terms of recognition. This game relies on the substitution method, by the way. It also explains how generalization works.
Generalizing is the same as abstraction and relies on forgetting differences.
9. “Capacity for learning, reasoning, understanding, and similar forms of mental activity; aptitude in grasping truths, relationships, facts, meanings, etc.” Random House Unabridged Dictionary, 2006
Truth should not be considered as something global and ultimate, there is just no such thing. If we refer to the truthfulness of language then we should just check if some sentence properly reports the relevant phenomena from a given context.
Meaning can be understood if we recall the role of language as pointing to the relevant phenomena in some context. Then meaning stands for those relevant objects in that context. The Latin phrase "intendere arcum in" ("pointing an arrow at") fits here well. It's the target of your reference.
There is one more use of the term "meaning" as the goal of an action. I accept that usage as well. As the first meaning (could not help it) of the term, this one is also about some "aiming".
10. “The ability to learn, understand, and think about things.” Longman Dictionary or Contemporary English, 2006
This definition has nothing that we did not discuss above.
11. “: the ability to learn or understand or to deal with new or trying situations : . . . the skilled use of reason (2) : the ability to apply knowledge to manipulate one’s environment or to think abstractly as measured by objective criteria (as tests)” Merriam-Webster Online Dictionary, 2006
Objective criteria are a myth. Recall the multi-dimensionality of phenomena. Take an apple. Is it an object, fruit, apple, or Golden Delicious? Do we move or walk? We can only talk about objective criteria when we limit our view of a phenomenon to just one of its dimensions. Then let's call them constrained criteria.
12. “The ability to acquire and apply knowledge and skills.” Compact Oxford English Dictionary, 2006
This definition has nothing that we did not discuss above.
13. “. . . the ability to adapt to the environment.” World Book Encyclopedia, 2006
This definition has nothing that we did not discuss above.
14. “Intelligence is a property of mind that encompasses many related mental abilities, such as the capacities to reason, plan, solve problems, think abstractly, comprehend ideas and language, and learn.” Wikipedia, 4 October, 2006
This definition has nothing that we did not discuss above.
15. “Capacity of mind, especially to understand principles, truths, facts or meanings, acquire knowledge, and apply it to practise; the ability to learn and comprehend.” Wiktionary, 4 October, 2006
This definition has nothing that we did not discuss above.
16. “The ability to learn and understand or to deal with problems.” Word Central Student Dictionary, 2006
This definition has nothing that we did not discuss above.
17. “The ability to comprehend; to understand and profit from experience.” Wordnet 2.1, 2006
This definition has nothing that we did not discuss above.
18. “The capacity to learn, reason, and understand.” Wordsmyth Dictionary, 2006
This definition has nothing that we did not discuss above.
Psychologist definitions
1. “Intelligence is not a single, unitary ability, but rather a composite of several functions. The term denotes that combination of abilities required for survival and advancement within a particular culture.” A. Anastasi
Survival is tricky, just as competition, success or victory. Can we confidently say that Neanderthals were not intelligent? Consider a wolf chasing a rabbit. If the rabbit escapes and the wolf dies of hunger and exhaustion, does that mean that it was not intelligent? If the wolf is successful in the chase, does that imply that the rabbit was not intelligent? Humans are intelligent and many of them die long before the old age. If that is not enough, think about the sacrifices some people sometimes make for others. No, survival only sounds cool as the necessary feature of intelligence.
Culture is different. Culture is the collection of all the knowledge collected and refined by humans. Its volume has long become too big for any single person to hold in memory. The invention of language to enable the storage and transfer of information was truly a wonderful gift from our intelligence. Still, knowledge and the ability to generate it are different things. Intelligence creates culture, intelligence is primary, culture is derivative.
2. “. . . that facet of mind underlying our capacity to think, to solve novel problems, to reason and to have knowledge of the world.” M. Anderson
This definition has nothing that we did not discuss above.
3. “It seems to us that in intelligence there is a fundamental faculty, the alteration or the lack of which, is of the utmost importance for practical life. This faculty is judgement, otherwise called good sense, practical sense, initiative, the faculty of adapting ones self to circumstances.” A. Binet
This definition has nothing that we did not discuss above.
4. “We shall use the term ‘intelligence’ to mean the ability of an organism to solve new problems . . . ” W. V. Bingham
This definition has nothing that we did not discuss above.
5. “Intelligence is what is measured by intelligence tests.” E. Boring
Intelligence tests measure a lot of manifestations of intelligence. Often those manifestations are related to only a subset of the dimensions humans learn to deal with in their lives using their intelligence. Often these tests can be passed by memorizing lots of example questions. Without understanding intelligence it is hard to develop good tests for it. Understanding intelligence, it is probably unnecessary to develop such tests.
6. “. . . a quality that is intellectual and not emotional or moral: in measuring it we try to rule out the effects of the child’s zeal, interest, industry, and the like. Secondly, it denotes a general capacity, a capacity that enters into everything the child says or does or thinks; any want of ’intelligence’ will therefore be revealed to some degree in almost all that he attempts;” C. L. Burt
"quality that is intellectual and not emotional or moral" - humans deal with multiple dimensions in their lives, be they intellectual or emotional. They approach them with the same substitution method. I would rather claim that it is non-intellectual to ignore the emotional side.
7. “A person possesses intelligence insofar as he has learned, or can learn, to adjust himself to his environment.” S. S. Colvin
This definition has nothing that we did not discuss above.
8. “. . . the ability to plan and structure one’s behavior with an end in view.” J. P. Das
This definition has nothing that we did not discuss above.
9. “The capacity to learn or to profit by experience.” W. F. Dearborn
This definition has nothing that we did not discuss above.
10. “. . . in its lowest terms intelligence is present where the individual animal, or human being, is aware, however dimly, of the relevance of his behaviour to an objective. Many definitions of what is indefinable have been attempted by psychologists, of which the least unsatisfactory are 1. the capacity to meet novel situations, or to learn to do so, by new adaptive responses and 2. the ability to perform tests or tasks, involving the grasping of relationships, the degree of intelligence being proportional to the complexity, or the abstractness, or both, of the relationship.” J. Drever
This definition has nothing that we did not discuss above.
11. “Intelligence A: the biological substrate of mental ability, the brains’ neuroanatomy and physiology; Intelligence B: the manifestation of intelligence A, and everything that influences its expression in real life behavior; Intelligence C: the level of performance on psychometric tests of cognitive ability.” H. J. Eysenck.
This definition has nothing that we did not discuss above.
12. “Sensory capacity, capacity for perceptual recognition, quickness, range or flexibility or association, facility and imagination, span of attention, quickness or alertness in response.” F. N. Freeman
Attention, as I see it, is different from intelligence. Intelligence is responsible for processing signals. Attention helps perception in collecting them. Intelligence and attention/perception work in tandem.
It may be that intelligence suggests to attention what to focus on or what to watch for. Recall that we use constraints to facilitate the substitution method. Attention may help in noticing them.
13. “. . . adjustment or adaptation of the individual to his total environment, or limited aspects thereof . . . the capacity to reorganize one’s behavior patterns so as to act more effectively and more appropriately in novel situations . . . the ability to learn . . . the extent to which a person is educable . . . the ability to carry on abstract thinking . . . the effective use of concepts and symbols in dealing with a problem to be solved . . . ” W. Freeman
This definition has nothing that we did not discuss above.
14. “An intelligence is the ability to solve problems, or to create products, that are valued within one or more cultural settings.” H. Gardner
Creation of products or tools may help us to achieve finer granularity in differentiating values along those multiple dimensions that are fundamental for our understanding of the world and interaction with it. As a result, we may satisfy more needs or collect more signals to process.
Using microscopes we collected information that opened our eyes on what the matter consists of and what we did not know about our bodies and what other life forms are out there. The more Lego pieces we feed to intelligence the wider variety it can address with them.
15. “. . . performing an operation on a specific type of content to produce a particular product.” J. P. Guilford
This definition has nothing that we did not discuss above.
16. “Sensation, perception, association, memory, imagination, discrimination, judgement and reasoning.” N. E. Haggerty
This definition has nothing that we did not discuss above.
17. “The capacity for knowledge, and knowledge possessed.” V. A. C. Henmon
This definition has nothing that we did not discuss above.
18. “. . . cognitive ability.” R. J. Herrnstein and C. Murray
This definition has nothing that we did not discuss above.
19. “. . . the resultant of the process of acquiring, storing in memory, retrieving, combining, comparing, and using in new contexts information and conceptual skills.” Humphreys
This definition has nothing that we did not discuss above.
20. “Intelligence is the ability to learn, exercise judgment, and be imaginative.” J. Huarte
This definition has nothing that we did not discuss above.
21. “Intelligence is a general factor that runs through all types of performance.” A. Jensen
This definition has nothing that we did not discuss above.
22. “Intelligence is assimilation to the extent that it incorporates all the given data of experience within its framework . . .There can be no doubt either, that mental life is also accommodation to the environment. Assimilation can never be pure because by incorporating new elements into its earlier schemata the intelligence constantly modifies the latter in order to adjust them to new elements.” J. Piaget
This definition has nothing that we did not discuss above.
23. “Ability to adapt oneself adequately to relatively new situations in life.” R. Pinter
This definition has nothing that we did not discuss above.
24. “A biological mechanism by which the effects of a complexity of stimuli are brought together and given a somewhat unified effect in behavior.” J. Peterson
This definition has nothing that we did not discuss above.
25. “. . . certain set of cognitive capacities that enable an individual to adapt and thrive in any given environment they find themselves in, and those cognitive capacities include things like memory and retrieval, and problem solving and so forth. There’s a cluster of cognitive abilities that lead to successful adaptation to a wide range of environments.” D. K. Simonton
This definition has nothing that we did not discuss above.
26. “Intelligence is part of the internal environment that shows through at the interface between person and external environment as a function of cognitive task demands.” R. E. Snow
This definition has nothing that we did not discuss above.
27. “. . . I prefer to refer to it as ‘successful intelligence.’ And the reason is that the emphasis is on the use of your intelligence to achieve success in your life. So I define it as your skill in achieving whatever it is you want to attain in your life within your sociocultural context — meaning that people have different goals for themselves, and for some it’s to get very good grades in school and to do well on tests, and for others it might be to become a very good basketball player or actress or musician.” R. J. Sternberg
Can we claim that unsuccessful people are not intelligent?
28. “. . . the ability to undertake activities that are characterized by (1) difficulty, (2) complexity, (3) abstractness, (4) economy, (5) adaptedness to goal, (6) social value, and (7) the emergence of originals, and to maintain such activities under conditions that demand a concentration of energy and a resistance to emotional forces.” Stoddard
Requiring intelligence to resist emotional forces is not intelligent, IMO.
29. “The ability to carry on abstract thinking.” L. M. Terman
This definition has nothing that we did not discuss above.
30. “Intelligence, considered as a mental trait, is the capacity to make impulses focal at their early, unfinished stage of formation. Intelligence is therefore the capacity for abstraction, which is an inhibitory process.” L. L. Thurstone
This definition has nothing that we did not discuss above.
31. “The capacity to inhibit an instinctive adjustment, the capacity to redefine the inhibited instinctive adjustment in the light of imaginally experienced trial and error, and the capacity to realise the modified instinctive adjustment in overt behavior to the advantage of the individual as a social animal.” L. L. Thurstone
The capacity to inhibit an instinctive adjustment is the ability to prefer System 2 over System 1 but both are parts of intelligence. How can we claim that intelligence is the ability to ask itself to think more?
32. “A global concept that involves an individual’s ability to act purposefully, think rationally, and deal effectively with the environment.” D. Wechsler
This definition has nothing that we did not discuss above.
33. “The capacity to acquire capacity.” H. Woodrow
This definition has nothing that we did not discuss above.
34. “. . . the term intelligence designates a complexly interrelated assemblage of functions, no one of which is completely or accurately known in man . . . ” R. M. Yerkes and A. W. Yerkes
This definition has nothing that we did not discuss above.
35. “. . . that faculty of mind by which order is perceived in a situation previously considered disordered.” R. W. Young
It is pattern recognition. When order is perceived in a situation previously considered disordered, it means that we learned about some dimension and its possible values and how they affect the results of some actions.
Without knowing the rules, team sports may often seem chaotic. After we learn the rules and become comfortable with them, the behavior of players starts to make sense.
4 AI researcher definitions
1. “. . . the ability of a system to act appropriately in an uncertain environment, where appropriate action is that which increases the probability of success, and success is the achievement of behavioral subgoals that support the system’s ultimate goal.” J. S. Albus
This definition has nothing that we did not discuss above.
2. “Any system . . . that generates adaptive behviour to meet goals in a range of environments can be said to be intelligent.” D. Fogel
This definition has nothing that we did not discuss above.
3. “Achieving complex goals in complex environments” B. Goertzel
This definition has nothing that we did not discuss above.
4. “Intelligent systems are expected to work, and work well, in many different environments. Their property of intelligence allows them to maximize the probability of success even if full knowledge of the situation is not available. Functioning of intelligent systems cannot be considered separately from the environment and the concrete situation including the goal.” R. R. Gudwin
This definition has nothing that we did not discuss above.
5. “[Performance intelligence is] the successful (i.e., goal-achieving) performance of the system in a complicated environment.” J. A. Horst
This definition has nothing that we did not discuss above.
6. “Intelligence is the ability to use optimally limited resources – including time – to achieve goals.” R. Kurzweil
I agree that intelligence may result in the optimum use of limited resources but it puts the horse before the cart. The same is true about "compression". Both are manifestations of intelligence but not parts of its mechanism.
7. “Intelligence is the power to rapidly find an adequate solution in what appears a priori (to observers) to be an immense search space.” D. Lenat and E. Feigenbaum
The power to find an adequate solution in an immense search space relies on respect for multiple dimensions. They can be combined in countless ways. All and any of those combinations may be decomposed along the known dimensions. Knowing how to handle each of them enables us to traverse that immense search space.
8. “Intelligence measures an agent’s ability to achieve goals in a wide range of environments.” S. Legg and M. Hutter
This definition has nothing that we did not discuss above.
9. “. . . doing well at a broad range of tasks is an empirical definition of ‘intelligence’ ” H. Masum
This definition has nothing that we did not discuss above.
10. “Intelligence is the computational part of the ability to achieve goals in the world. Varying kinds and degrees of intelligence occur in people, many animals and some machines.” J. McCarthy
The computational metaphor is great but without the details about the kind of computations used by our cognitive apparatus we cannot implement "a cognitive machine". Neither can we understand ourselves. Isn't it the ultimate goal of cognitive science?
I hope that by now you understand the role of comparisons in our cognition. Statistical methods pale in comparison.
11. “. . . the ability to solve hard problems.” M. Minsky
This definition has nothing that we did not discuss above.
12. “Intelligence is the ability to process information properly in a complex environment. The criteria of properness are not predefined and hence not available beforehand. They are acquired as a result of the information processing.” H. Nakashima
This definition has nothing that we did not discuss above.
13. “. . . in any real situation behavior appropriate to the ends of the system and adaptive to the demands of the environment can occur, within some limits of speed and complexity.” A. Newell and H. A. Simon
This definition has nothing that we did not discuss above.
14. “[An intelligent agent does what] is appropriate for its circumstances and its goal, it is flexible to changing environments and changing goals, it learns from experience, and it makes appropriate choices given perceptual limitations and finite computation.” D. Poole
This definition has nothing that we did not discuss above.
15. “Intelligence means getting better over time.” Schank
Getting better is the same as getting different. Which direction is "good" or "bad' is beyond the scope of this paper. What I can say at this point is that it is poetic to define intelligence that way. I am interested in a practical definition.
16. “Intelligence is the ability for an information processing system to adapt to its environment with insufficient knowledge and resources.” P. Wang
It is not "insufficient" but "available". When resources are insufficient and one cannot "adapt to its environment" it does not imply that one is not intelligent. It only implies that resources are inadequate for the task. Recognizing that is intelligent.
17. “. . . the mental ability to sustain successful life.” K. Warwick
This definition has nothing that we did not discuss above.
18. “. . . the essential, domain-independent skills necessary for acquiring a wide range of domain-specific knowledge – the ability to learn anything. Achieving this with ‘artificial general intelligence’ (AGI) requires a highly adaptive, general-purpose system that can autonomously acquire an extremely wide range of specific knowledge and skills and can improve its own cognitive ability through self-directed learning.” P. Voss
This definition has nothing that we did not discuss above.
My Definition of Intelligence
Intelligence is the ability to handle differences.
From that definition of intelligence, it follows that intelligence defines definitions. It does not occur in a vacuum. Rather each definition is a result of numerous observations and interactions with a defined phenomenon.
It relies on the use of comparable properties and their ranges also known as concepts/categories/classes.
Its core algorithm is illustrated by the substitution method mentioned above. It is about the selection of the most fitting option from the available ones respecting relevant constraints.
We can rephrase any definition or algorithm in terms of comparisons and differences. Consider the "task-solving".
In the object recognition task, available options are known categories and relevant constraints are the properties of an object at hand. The selection procedure reminds the binary search but it is a semantic binary search in this case. We start with checking the most general properties and with each answer (as in the game 20 Questions) we filter out roughly half of the remaining categories. Each answer leaves us with a current category - Tangible, Animate, Animal, Mammal, Feline, Tiger, etc.
In the question-answering task, available options are known facts, and relevant constraints are the constituents provided in the question.
In the planning task, available options are actions and relevant constraints are the differences between the current and the desired states. Given our knowledge about the prerequisites of each action and what properties they affect, we can select those actions that will promise to achieve our goal.
In the forecasting task, available options are possible results and relevant constraints are the current strategies and relative capabilities of the actors. It may be as simple as forecasting rain observing dark clouds or as hard as forecasting a stock price behavior after the elections.
The adaptation task combines the previous two.
In the reasoning task, for example, a detective investigation, available options are the suspects and relevant constraints are the clues or alibis.
As you can see, it is possible to frame any task as a selection, which can be then solved by semantically forming a set of options and filtering it using comparisons against the relevant constraints.
Differences from prediction
Currently, the prevalent paradigm for machine learning is based on statistical methods. It relies on the use of human-provided data to determine any correlations there. Those correlations are about co-occurrences and not necessarily meaningful ones. If defining features of a class are not present in the provided data, the system does not care.
LLMs recently became popular. They are considered by many as amazing tools for processing natural languages. They are based on statistical methods and are trained on massive amounts of text. Only text.
Imagine you are given a text in a language you don't know, you have no pictures associated with the text, no videos, no dictionary or grammar rules. Can you learn the language of the text? If you can answer some questions in that language as a result of noticing some co-occurrences in the text, what can you say about your understanding of the question and your answer? Will you be bored by a paragraph on accounting and excited about a love scene paragraph? Will you differentiate the two?
Differences from logic
Before statistical methods, there were expert systems and GOFAI based on formal logic. Let me ask you, "What is the logical value of a question?" Is it true or false? Is the "second-best" option true? If a character's name is Sarah Jane, is she lying saying, "I am Sarah"?
Please note that trivial cases like the mortality of Socrates are easily solved in the proposed model using differences. Specialization/generalization are perfectly explained via differences. Rules and exceptions are introduced at different levels of the specialization/generalization tree.
But the biggest advantage of the proposed theory is the ability to look at phenomena from multiple angles, along different dimensions. Not only does that make their treatment meaningful but it also enables us to break down complex tasks into manageable subtasks with known algorithms.
Intelligence
Intelligence is the opposite of insanity. Check out the definition of the latter from Albert Einstein. Intelligence is about two questions, "Does it make a difference?" and "What difference does it make?" Differences in inputs and outputs. Change. Flow. Progress.
References
1. https://www.dictionary.com/browse/definition
2. Legg, Shane & Hutter, Marcus. (2007). A Collection of Definitions of Intelligence. Advances in Artificial General Intelligence: Concepts, Architectures and Algorithms. 157.