Abstract
The whole world does nothing but talk about Artificial Intelligence, but at the same time - what a paradox! - the definition, in fact, of "intelligence" (not even artificial, but in general) - there is still no generally accepted, understandable, logically structured and deep! Why not take the liberty of trying to find and propose such a definition? After all, definition is the foundation on which everything else is built, right? How do we build AI if everyone sees differently what should be the basis? Go…
Key words: intelligence, ability, property, object, adaptation, behavior, environment, preservation, survival.
To describe the existing definitions of intelligence, the article "A Collection of Definitions of Intelligence" (S. Legg, M. Hutter. A Collection of Definitions of Intelligence (2007), ), quotes from which are presented along with comments (italics).
Entry
This article (A Collection of…) is a review of a large number (over 70!) of informal definitions of the term "intelligence" that the authors have collected over the years. Naturally, compiling a complete list would be impossible, since many definitions of intelligence are buried deep in articles and books. Nevertheless, the definitions presented here are the largest selection, provided with detailed references ...
Despite a long history of research and debate, there is still no standard definition of intelligence. This has led some to think that intelligence can only be approximated, not fully defined. We believe that this degree of pessimism is too strong. Although there is no single reference definition, if one looks through the many proposed ones, it quickly becomes apparent that there is a strong similarity between the many definitions.
Definition of intelligence
Definitions from general sources (dictionaries, encyclopedias, etc.)
(there are 3 best definitions of intelligence out of 18 that are given in this section of the original article. The choice was made according to the criterion - the width and depth of coverage of properties - abilities, characteristics, parameters, etc., given in the definition).
- The ability to effectively adapt to the environment, either by making changes in oneself, or by changing the environment, or by finding a new one ...
- Intelligence is not a single mental process, but rather a combination of many mental processes aimed at effective adaptation to the environment.
Adaptation is the result of the manifestation of many unspecified properties that form intelligence. It is important that the environment is indicated - existing or even new.
- The ability to learn and understand or deal with new or difficult situations;
- Skillful use of the mind;
- The ability to apply knowledge to influence the environment, or the ability to think abstractly, as measured by objective criteria (when tested).
It is important that the environment is indicated! Flaws:
- Through the union "or" the categories of different quality are connected: "the ability to learn" and "to deal with new situations".
- And "skillful use of the mind" is not at all suitable for definition.
- People differ from each other in their ability to comprehend complex ideas, the effectiveness of adapting to the environment, learning from experience, engaging in various forms of reasoning, overcoming obstacles through reflection.
Well, at least people are indicated, that is, a person with abilities! The effectiveness of adaptability is indicated - this is important, but the adaptation itself is not included in the list! Overcoming obstacles is essentially problem solving.
Descriptions given by psychologists (best 3 out of 35 definitions given)
- I prefer to call intelligence "successful intelligence". And the reason is that the emphasis is on using the intellect to achieve success in life. Therefore, I define intelligence as a skill for achieving what one wants to achieve in life in a socio-cultural context, which means that people have different goals: for someone it is getting very good grades in school and passing tests, for someone, maybe , become a very good basketball player, or an actress, or a musician.
The goal is explicitly to achieve success in life, but that's all ...
From the most general point of view, intelligence is present where an individual animal or person is aware, albeit dimly, of the relevance of his behavior in relation to the goal. Of the many definitions taken by psychologists to define what is indefinable, more or less acceptable are:
- the ability to respond to new situations or learn to do so through new adaptive responses, and
- the ability to perform tests or solve problems, including grasping relationships, with the level of intelligence proportional to the complexity or abstraction, or both.
So, a hierarchy appeared: “From the most general point of view ...”, this is already good. But that's where all the good stuff ends...
- Tautology: Respond... with new adaptive responses. There is no difference - with the help of old or new reactions, the main thing is to react!
- Now for the tests… Capturing relationships is good, but far from enough!
- Intelligence is not a single ability, but a composite one, consisting of several functions. It means a combination of abilities necessary for survival and development within a particular culture.
Oh, survival by intelligence is finally indicated! But everything else is lost...
Descriptions given by AI researchers (best 3 out of 18)
- An intelligent agent does what] suits his circumstances and his purpose; it is flexible to changing conditions and to changing goals, it learns from experience and makes appropriate choices based on perceptual limitations and processing capabilities.
Perhaps the best (of all presented here) definition of intelligence.
The goal is noted, however, but not specified.
Adaptability - both in terms of conditions and in terms of purpose. The latter means that there is no concept of the most important goal!
Learning is the identification (though not explicitly indicated) of the properties of the environment, memorization, use.
Choice means criteria are implied.
Restrictions - on perception and on influence.
- “The ability to learn something is the necessary, domain-independent skills needed to acquire a wide range of domain-specific knowledge. Achieving this with General AI requires a highly adaptive, general-purpose system that can independently acquire an extremely wide range of specific knowledge and skills, and can improve its own cognitive abilities through self-education.”
It seems that here the ability to learn something is the ultimate goal ... And the properties of General AI follow from it - high adaptability, universality ...
- Intelligent systems need to work, and work well, in many different environments. Their intelligence allows them to maximize the likelihood of success, even if there is no complete knowledge of the situation. The functioning of intelligent systems cannot be considered separately from the environment, from the specific situation, including the goal.
What is "doing a good job"? What is success?
Possibility of prefabricated description
If we “pull out” frequently occurring functions (features, characteristics, etc.) from the considered definitions, then we find that intelligence:
- It is a property that an individual agent has in its interaction with its environment/environments.
- This property refers to the ability of an agent to achieve success or benefit in relation to some goal or task.
- This property depends on how the agent is able and must adapt to different targets and environments.
Using these key attributes together gives us our informal definition of intelligence: Intelligence is measured by the ability of an agent to achieve goals under a wide range of conditions.
But wait, we need an answer to the question: what is intelligence, and not how (or by what) it is measured (estimated)! You can justify the authors of the article by saying that these definitions are almost thirteen years old, and expect that something should have changed over the next years - after all, the IT field is developing at a frantic pace ... But the following is an example from an article in 2012, (M. Hutter, One Decade of Universal Artificial Intelligence, ) where practically nothing has changed in the definition of intelligence:
Reasoning, creativity, association, generalization, pattern recognition, problem solving, memorization, planning, achievement of goals, learning, optimization, self-preservation, vision, language processing, classification, induction and deduction, knowledge acquisition and processing ... A precise definition of intelligence, including each of its aspect seems to be difficult to give.
Again, the same problems (even more) with the definition as 8 years ago: manifestations of intelligence are given in the form of an unstructured list of characteristics!
Wikipedia definition of intelligence (accessed May 22, 2016):
“Intellect (from Latin intellectus - sensation, perception, understanding, understanding, concept, reason) is the quality of the psyche, consisting of the ability to adapt to new situations, the ability to learn from experience, understand and apply abstract concepts and use your knowledge to manage environment. The general ability to know and solve problems, which unites all the cognitive abilities of a person: sensation, perception, memory, representation, thinking, imagination.
The same Wikipedia, but in the most recent edition on January 24, 2020:
“Intellect (from Latin intellectus “perception”, “understanding”, “understanding”, “concept”, “reason”) or mind is a quality of the psyche, consisting of the ability to adapt to new situations, the ability to learn and remember based on experience, understanding and applying abstract concepts, and using their knowledge to manage the human environment. General ability to know and solve problems, which combines cognitive abilities: sensation, perception, memory, representation, thinking, imagination, as well as attention, will and reflection.
So many years have passed, but we see everything the same - a set of characteristics without any structure ... And with an indication of a person - the bearer of intelligence, only at the very end of the text. That is, it is not possible to make a replacement: “abstract Object with intelligence —> Human with intelligence” with subsequent identification in this definition: “What does a Human need to become intelligent?” is not possible. Or does this substitution lead to banal wishes: In order to become intelligent, a person needs to acquire the ability to adapt to new situations, to learn from experience, to understand and apply abstract concepts and to use their knowledge to control the environment, etc. In short , that's how you can become smart, and not remain stupid ...
So, on the basis of the foregoing, the following definition is proposed, tied to the Object, since the intellect cannot "hang in the air", it is necessarily someone's abilities. The same applies to behaviors that only someone or something can have:
An Object's Intelligence is a set of abilities that are used to:
(1) Identification, formalization and memorization (in the form of a model) of the laws of state and / or behavior:
(1.1) Environment, and
(1.2) Internal environment of the Object.
(2) Forward modeling of states and/or behaviors:
(2.1) in the Environment, and
(2.2) Internal environment of the Object.
(3) Creating a description of the state and / or implementation of the behavior of the Object, adapted:
(3.1) to the environment, and
(3.2) to the Internal Environment of the Object
subject to maximizing the ratio Object Behavior / Behavior Cost
Object in order to preserve (existence, duration, existence) of the Object in the Environment
environment.
This is how it looks on the diagram:
»
Now about the application of the definition ... Truth, as they say, is always concrete. Therefore, in order to check the logic of the definition, one should replace the Object with some well-known and understandable concrete system, for example, with… A car. So…
A car with intelligence is a car with a set of abilities that are used for:
(1) Identification, formalization and memorization (in the form of a model) of the laws of state and / or behavior:
(1.1) traffic conditions, and
(1.2) The internal environment of the Vehicle.
(2) Forward modeling of states and/or behaviors:
(2.1) in traffic conditions, and
(2.2) Vehicle interior environment
(3) Creating a description of the state and / or implementation of the behavior of the Vehicle, adapted:
(3.1) to the traffic situation, and
(3.2) to the Vehicle's Interior Environment
subject to maximizing the ratio (Car Behavior / Behavior Cost
Car) in order to preserve (existence, duration, existence) of the Car - both in the Road environment and the Internal environment of the Car.
Am I the only one who can see that we call a Car with such abilities intellectual? Then another question: would you notice the difference between driving in a car driven by a professional driver and driving in such an Intelligent Car?
The answer "NO" means:
- The correct definition of intelligence is given: when replacing “Object -> Car”, there were no logic failures or any inconsistencies in the description.
- A car with such abilities during the trip, as it were, passed Turing's "automobile" test: the passenger on the trip did not see any difference between a car with a professional driver and this Car. Or, if the wording of the Turing test is exactly followed: “If, during several passenger trips in an unmanned vehicle and in a car with a professional driver, the passenger cannot guess which car drove him, then in terms of the level of “thinking in road conditions”, an unmanned vehicle can be considered equal to a car with professional driver.
Those who wish are invited to “play” with this definition - substitute in it, instead of the impersonal word “Object”, the name of any, if desired, well-known system (natural, social, industrial, technical) and independently check, thereby, compatibility. Be sure to share the results and thoughts on the results of the experiment!
Definition of intelligence through its goals
(A. Zhdanov. "Autonomous Artificial Intelligence" (2012), 3rd edition, electronic, pp. 49-50):
The main goals that the nervous system of any organism strives for are:
- the survival of the organism;
- accumulation of knowledge by his nervous system.
These 2 points: survival and accumulation of knowledge - a generalized description of points 3 and 2, respectively!
As a Conclusion…
"Vicarious teaches the computer to use imagination"
("The computer learned to drive aggressively" )
“Life would be pretty boring without imagination. So probably the biggest problem with computers is that they don't really have any imagination. Startup Vicarious is creating a new way of processing data, inspired by how information is likely to flow through the brain. Company leaders say it will give computers something akin to the imagination, and that they hope will help make machines much smarter. The company introduced a new type of neural network algorithm, with properties borrowed from biology. One of them is the ability to imagine what the learned information would look like in different scenarios – a kind of digital imagination.”
Wow, what a coincidence! Exactly item (2) of the definition: advanced reflection is the digital imagination!
It doesn't happen often, but look what we find on the Web:
("The computer learned to drive aggressively" )
“Specialists from the Georgia Institute of Technology have assembled a model of an unmanned vehicle (1: 5 scale based on a commercial radio-controlled model chassis) capable of cornering using a controlled skid. The on-board computer is equipped with an Intel Skylake Quad-core i7 processor and an Nvidia GTX 750ti GPU and processes information from a gyroscope, wheel sensors, GPS and a pair of front cameras. Based on the data coming from the sensors, the control algorithm generates 2560 forward trajectories for the next two and a half seconds.”
The control algorithm contains the "picture of the world" of the car in the form of a set of possible trajectories of movement along a given track.
“Out of 2560 trajectories, the algorithm selects the most optimal one and, according to it, corrects the position of the wheels and speed. At the same time, all 2560 trajectories are built and updated 60 times per second.”
This is anticipatory reflection, artificial creativity or digital imagination! Choosing the optimal trajectory from 2560 pre-generated ones and adjusting the position of the wheels and speed (adaptation!) to stay on the track. Everything together is described by the presented scheme of intelligence!
“The whole process of learning the control algorithm took several minutes of driving around the track by an operator with little experience in management”
The learning process is the creation of a picture of the world!
“At the same time, the researchers note, during the training, a controlled skid was not used, his computer was “invented” on his own. During the tests, the car autonomously passed the track, trying to maintain a speed as close as possible to eight meters per second.
Controlled skidding is an element of the optimal strategy (that very maximization of the “Object Behavior / Behavior Costs” ratio) independently developed by the car.
“According to the authors, learning algorithms for aggressive driving can be useful for everyday driving of an unmanned vehicle in the same way that the ability to control a skid can be useful for a live driver. In the event of an unforeseen situation, for example, ice, the unmanned vehicle will be able to independently get out of the skid and prevent a possible accident.”
And this is the dissemination of the experience of the car ... Well, like a guardian bird (we recall a well-known story), having received a useful skill, immediately passed it on to everyone else.
Once again I will give the definition that is proposed for use:
An Object's Intelligence is a set of abilities that are used to:
(1) Identification, formalization and memorization (in the form of a model) of the laws of state and / or behavior:
(1.1) Environment, and
(1.2) Internal environment of the Object.
(2) Forward modeling of states and/or behaviors:
(2.1) in the Environment, and
(2.2) Internal environment of the Object.
(3) Creating a description of the state and / or implementation of the behavior of the Object, adapted:
(3.1) to the environment, and
(3.2) to the Internal Environment of the Object
subject to maximizing the ratio Object Behavior / Behavior Cost
Object for the purpose of preserving (existence, duration, existence) of the Object in the Environment.
Thank you for attention. Comments, remarks - absolutely welcome.
PS But we can talk about "... a highly adaptive, universal system that has the ability to independently acquire an extremely wide range of specific knowledge and skills" and which is required to create AGI, we can talk separately - this is a very interesting topic. Unless, of course, there is interest from readers. 🙂
Source: habr.com