The position of systems thinking among other forms of thinking: the intellect stack
There are different opinions on whether it is possible to call applied thinking (for example, the thinking of a rocket engineer, or the process of teaching people using the blended learning pedagogical method, or repairing toilets on spaceships) thinking. According to one viewpoint - of course, agents always think, but according to another opinion, thinking is only the function of generating a new applied method with its knowledge/disciplines serving as potential algorithms for changing the world for situations where encountered problems and tools for these changes, and as for calculations/reasoning with applied methods thinking is not really required, these thoughts go "automatically". The work of a calculator is not called "thinking," it calculates something "mechanically," and that's it. According to another opinion (which we adhere to as well) in situations of an applied method (even the work of a calculator, for example, a child painstakingly multiplying three-digit numbers in columns on paper) we can still talk about applied thinking. In some specified applied domain you may know well what to do in a typical situation- and do it "automatically," without thinking about the method of work, "not think". But there could be many objects in this applied domain, so you will need to:
- independently develop a method to solve a particular problem (explanations/knowledge/algorithms and tools that support them) of a specific problematic situation, a particular difficulty, even without going beyond the specific subject area (sometimes the development of a method for a certain problem situation is called strategizing, and the method found - a strategy),
- then plan the work (the method is only a way to perform work, the strategy does not include a plan with scheduled deadlines and resources in it), based on the available or expected resources (and if there is a lack of resources, then adjust the method, replacing it with one for which there are enough resources, or consider work by some methods of obtaining the necessary resources), and
- only then carry out the work using this method-strategy, and also
- monitor whether the results of the work being done meet, or if it is necessary to urgently adapt the method, since either the situation with the initial data or the required results has changed, or there were errors in previous steps.
So for elementary operations/actions in the world it seems that thinking is not needed, but thinking is needed to choose a explainable chain of actions (this is the method) when you need to consider many circumstances, keep your attention on a huge number of objects that change at each step following the working method—and that's where we tend to talk about applied thinking, not just about automatic execution of individual operations. Hitting with a hammer, meaning simply moving the muscles of the hammer—isn't a sign of great intelligence is needed, but to hit safely, at the right time, in the right place—that's where thinking is required, you need to discuss a method of hammer strikes, knowledge/disciplines and tools of this method usage.
Immediately mastering applied thinking to solve problems/difficulties in a working subject area, and then combining these different applied thinkings for different work methods in complex projects, involving hundreds of people, is not easy. But even after mastering narrow applied thinking for one method, it should be recognized that without supporting applied knowledge with fundamental/transdisciplinary knowledge it is difficult to act effectively in the real world. At the interfaces of work on any applied methods there will be situations not described in any textbook or regulation or standard work! People must simply use fundamental knowledge/explanations of human civilization because only they allow to connect in thinking knowledge of different applied methods, as well as knowledge about work on the civilization frontier: solving problems, methods of typical solutions for which no one on Earth has developed yet, therefore these solutions cannot be just googled, understood—and then applied without thinking. Although you can ask AI for a problem-solving method, but it is not guaranteed that this AI will be smart enough to provide something meaningful (ask AI today: how to become immortal? It's not guaranteed that you will receive a description of an appropriate method to solve this problem).
If we at least partially understand the structure of the world, this can significantly reduce the amount of mental calculations/thought in partially known subject areas. Is this a lot? For example, a problem P can be solved by the human brain in ten thousand years. This is more than the existence of human civilization, although you can reduce this time to ten years if a thousand brains work on it, and you have resources to support a thousand people's lives, and you know how to efficiently organize the division of labor among a thousand people. So it's better to make some educated, already known civilization assumptions about the structure of the task and its subject area. In our example, these assumptions helped reduce the mental work of one brain by ten thousand times, and the task was solved by one person in a year. This is approximately the difference in speeds of work between uneducated people and educated ones: uneducated people (savages) know little general explanations about the functioning of the world, and educated ones know a lot. You need to learn to solve problems quickly with your own brain. Alternative: you have to learn to push thisamount of mental work onto the brains of other people and computer toolsincluding AI.
Civilization, with the help of science based on the written accumulation of explanations/theories/knowledge/models, gives us ready-to-study guesses about the structure of the world, as well as teaches to formulate problems (which you don't know how to solve, the subject of intelligent work) and translate them into tasks (which are known to be solved, the subject of applied mastery). These guesses are the basis of education. Education, in this way, reinforces the capabilities of the mind by teaching methods of thinking of the intelligence stack. Education is thus - a specialization of training (education::training). Remember: training is the method to create mastery of performing work according to a target method that you are taught. Education allows to quickly find applied methods for turning a problem into tasks, i.e. translating a situation from "I don't know how to approach this" to "I know how to approach this method to get a result—what knowledge and tools to use".
The acquired intelligence::mastery during education allows to solve problems tens of thousands times faster than it could be done by an untrained natural intelligence of homo sapiens. A civilized brain is not "wild," it is a trained brain, fast in thinking, and modern brain even uses the brains of other people (collective thinking work) and computers (classical and with AI programs) to enhance the speed of thinking. At the same time, computers can be used in minimal versions even not at the expense of computer processing but just by assisting in organizing memory and attention. A computer as a "pen and paper" is also very effective for thinking! A smart and lazy educated person with a laptop can do much more than a crowd of active but uneducated fools-savages!
Mastering new skills and abilities comes to a person not through "natural ingenuity," but through "educated ingenuity," through knowledge/models/explanations/theories/disciplines about the structure of the world, the structure of problems and tasks, as well as knowledge/explanations/disciplines about what tools are available (the most often these are modelers) to support work. The same applies to AI. Manufactured computers for AI are dull, they can only perform basic operations like matrix multiplication. But after training with a huge amount of knowledge already accumulated in written form by civilization, these computers gain reasoning skills based on this knowledge, "large language models/LLMs", also known as "foundation models," referring to their transdisciplinary nature. This is similar to "education": learning to think and some engineering insights. And then such foundation models can be easily finetuned with applied knowledge or connected to these models with applied knowledge in the form of tools (for example, connecting Wolfram Mathematica for solving mathematical problems).
The ability and skill, or skillset, refer to an agent's proficiency in executing work according to a method, supported by theory/knowledge/explanations/algorithms/disciplines, where the execution of this method/work method is supported by certain tools. Intelligence is the skill of mastering a set of fundamental thinking methods needed to debate the methods when it is unclear which method to apply (possibly there isn't such a method yet—either there is no knowledge of it, or there is no tool that needs to be created).
Intelligence::mastery works with applied methods (and by extension their knowledge/disciplines/algorithms/theories) as objects of its work. It can be said that intelligence as mastery of fundamental thinking precisely creates and further develops applied methods, it is necessary for cognition, for the endless growth of knowledge (evolution of knowledge) and support tools for working with this ever-growing knowledge. More knowledge and tools supporting this knowledge—more transforming various problems into tasks. Infectious diseases were a problem, but then knowledge about microbes and soap hygiene practices substantially solved these problems, fighting infections became a list of tasks, not a problem: we know what to do, we just need to find resources, and then simply do it.
Natural intelligence allows people to be smarter than cats and monkeys, but the intelligence gained through education as a machine for learning applied disciplines is the learned part—it is a learning machine for formulating problems that are not solvable by any known methods into solvable tasks. This applies to both natural and artificial intelligence, as well as hybrid and collective intelligence.
Transdisciplines/"fundamental disciplines" is the explanations/theories/knowledge/models/algorithms about the structure of the world. They are convenient for quick thinking about the world, maintaining focus on calculations/reasoning/thinking about important matters, saving the brain (or computer) resource from wasting on thinking about unimportant matters. This kind of thinking based on the knowledge/algorithms of transdisciplines is then needed to create methods for changing the world for the better.
What might take a very clever savage half a lifetime, an educated person can do in a few hours, or even in a few seconds—especially considering that knowledge/models/algorithms of thinking include knowledge about engaging thinking tools (most often these are modelers). This also applies to AI-enabled computers, the true AI models, having learned a wealth of knowledge already available in written form by civilization, develop reasoning skills based on this knowledge, "large language models/LLMs"—which are otherwise known as "foundation models," referring to their transdisciplinary nature. This is analogous to "education": teaching thinking and various engineering insights. And then such fundamental models can be fine-tuned with applied knowledge (finetune) or connected to these models with applied knowledge in the form of certain tools (for example, integrating Wolfram Mathematica for solving mathematical problems).
The ability and skill, skillset, refer to an agent's proficiency in executing work according to a method and relying on theory/knowledge/explanations/algorithms/disciplines, where the execution of this method/work method is supported by some tools. Intelligence is mastery in the application of a whole stack of fundamental thinking methods, which support each other. This set of thinking methods (supported by fundamental disciplines and the tools that assist them) is what we call the intelligence stack.
In the fundamental thinking methods, disciplines/theories/knowledge will be only the "algorithmic" part. We confidently consider these disciplines/theories/knowledge as "algorithms" (descriptions of how to use the method in various circumstances/situations—just as algorithms can be used for computations with a variety of input data):
- There is plenty of evidence that constructive mathematics by fact is actually a transition from declarative (objects and relations) descriptions to descriptions through operations for constructing objects. This can be extended to all work with concepts (mental/mathematical objects).
- In computer science, there have long been results that allow to consider various types of algorithm representations, not just "step-by-step execution of imperative programs" (including the correspondence Curry-Howard[1] between the imperative algorithm and a set of logical statements). We interpret this result broadly enough.
- One should consider not just knowledge itself, but what the computer does with it — in this case it's about the skill in performing a method that uses knowledge for calculations (thinking) or even changing the world (thinking and using tools). In the theory of creators (constructor theory)[2] there is a generalization of the concept of "algorithm" to the description of transformations not only of information and superinformation (superinformation, in quantum computers, represented not in bits, but in qubits), but also theoretically any physical transformations.
Thinking methods, like any other methods, not only use concepts from theories/knowledge/explanations/algorithms/disciplines (including transdisciplines), but also use tools, understood as a set of body-extending tools/apparatus/equipment for the agent. In the case of thinking based on transdiscipline methods, the modeler is typically used as the tool (simple from pen and paper, or a computer program for modeling), and the consumable material for the modeler includes coffee for the human modeler and electricity for the computer modeling effort. Other tools in fundamental thinking methods of the intelligence stack are rare, but they exist. For instance, in conceptualization, the body is used, where some sensations are being sought that will then need to be transformed into thoughts, and the role of the executor of the conceptualization method is the "poet."
Despite the practical nature of thinking, it is more about modeling the world, that is, engaging in cognition/learning, creating models, rather than changing it directly in action—but remember, these are models, necessary for changing oneself and the world for the better, and intelligence makes decisions, whether to change the model of the world, the self model, the self or the world—interdependently.
In the case of transitioning to applied engineering (as changing the world) by the method of "trial and error" in old and well-known to the agent or even new less-known to the agent fields, for changing the world, the agent uses an abundance of different tools and applies various starting materials: machine tools, chemical reagents, trained animals, sunlight, water in a pond, clocks, a ballet platform, a quantum computer, etc.
There is some cunning in thinking by fundamental methods—it is purely a "mental act". Input/output in a computer (e.g., the human brain) is physical and requires tools/equipment (book printing, online courses, messengers for receiving problems and sending solutions), and even the computer itself is a physical object. As David Deutsch, a mathematician and astrophysicist, likes to remind us—computing machines are perfectly physical objects, and "mental work" necessitates the physicality of the worker! In the course, we use the concept of a "creator", which is an abstraction over a computer capable of performing a computation algorithm onto the creator/constructor from constructor theory, able to perform the knowledge/algorithm method like "algorithm of changing the physical environment"[3].
The work method, executed by the creator, includes the knowledge aspect (which the creator programmatically/hardware-wise implements as mastery of executing the method/algorithms/theories of the method), and the hardware aspect (mastery of supporting it with the agent's body can also be classified in this hardware part, and further goes the appliance of tools like sensors and actuators, as well as "exotels" as the platform for all these sensors and actuators, that is the equipment/apparatus that helps the mastery to make additional calculations and actions for the measurements in the physical world and changes in the physical world. For universal creators (intelligent agents), we can speak not just about calculations but immediately about thinking, and also consider situations when during the execution of the method, the creator first builds additional hardware—the tools (bootstrapping/bootstrapping).
Constructor theory provides a generalization for the concepts of
- an algorithm (the term remains the same) as descriptions/theories/disciplines,
- measurement (input, physical interaction to get data),
- change (processing information in the case of information, in the case of matter—transformation),
- output (in the case of creators, this operation is not separately considered, it's included in the transformation).
So it's possible to further discuss transformations not only of information but also of physical objects, and also the concept of a computer which implements "information processing according to some algorithm"::method can expand to the concept of a creator that implements "changing the physical world according to some algorithm"::method even further.
A universal computer, with sufficient computational resources (memory and time), can fundamentally perform any computation that a Turing machine can (known as the theory of computing), and a universal creator with the consideration of bootstrapping/bootstrapping (e.g., starting with smelting metal from ore, getting clean silicon for semiconductors from sand) can fundamentally perform any transformation of matter—and "any" is understood in a mathematical sense, here we are talking about the theoretical possibility. The practical possibility, however, will be limited by resources and the risks of some catastrophe (say an asteroid hits and destroys the creator, which has all the necessary resources—but that creator does not manage to finish its work.
At the same time, it is particularly emphasized that with one method (algorithm/theory/explanations/knowledge and the equipment/tools) the creator is capable of performing multiple calculations, remaining unchanged—it is roughly the same as a catalyst molecule (a simple creator) can perform multiple catalysis cycles, remaining unchanged.