I think this is actually the starting point for any theory or practice of true artificial intelligence. Any system or algorithm (set of algorithms) for object detection and the related concept formation must take elemental sensory data that is essentially time and location data for real objects as input and build up a hierarchy of 'objects' by composition. I'm focusing on entropy as the fundamental discriminator for all data (signals). Detectable persistence in time and place together with stability of relationships between data points is my approach. I started this as a technique for natural language understanding, but realized it is fundamental and general to all sensory signals.
It should be mentioned that there is the same topic in Math Logic discipline Model theory [1, 2]. Where we have a) structures b) theories with concepts, relations, constants - primary and defined. And concept grounding may be treated as calculation of concept definition on particular structure.
So the main question may be What kind of structures AGI creates inside and what kind of manipulation (knowledge processing) it is doing?
It seems the main feature of the mind is processing of colored 3D figures, particulaly in motion.
In this case, structuring is not quite what (or not at all) what is meant in model theory.
Detection of objects in the environment (representation of a situation as a set of separate objects, the behavior of which can be predicted separately) is essentially a search task of potentially unlimited computational complexity.
Yes, this is a subtle topic. And for me if we take static for simplicity at-first, what we see is one large, variously curved and painted surface. But one. Reasoning for object separationis is a reasoning on this just one structure.
Consider the situation when all objects are glued and it turns out that, contrary to expectations, this is one such object.
Here is a nearest math but in Russian "Отображения гладких поверхностей на плоскость окружают нас со всех сторон. Действительно, большинство окружающих нас тел ограничено гладкими поверхностями. Видимые контуры тел - это проекции ограничивающих тела поверхностей на сетчатку глаза. Приглядываясь к окружающим нас телам, например к лицам людей, мы можем изучить особенности видимых контуров.
Уитни заметил, что в случаях "общего положения"* встречаются особенности лишь двух видов. Все другие особенности разрушаются при малом шевелении тел или направлений проектирования, в то время как особенности этих двух видов устойчивы и сохраняются при малых деформациях отображения."
We see those contours that are projections of the smooth surface of the body onto the "visual plane" due to the fact that they are illuminated. But lighting makes visible at the same time those features that are not projections of a smooth surface: irregularities that reflect light depending on the angle of incidence of light, edges, differences in color, etc.
I think this is actually the starting point for any theory or practice of true artificial intelligence. Any system or algorithm (set of algorithms) for object detection and the related concept formation must take elemental sensory data that is essentially time and location data for real objects as input and build up a hierarchy of 'objects' by composition. I'm focusing on entropy as the fundamental discriminator for all data (signals). Detectable persistence in time and place together with stability of relationships between data points is my approach. I started this as a technique for natural language understanding, but realized it is fundamental and general to all sensory signals.
It should be mentioned that there is the same topic in Math Logic discipline Model theory [1, 2]. Where we have a) structures b) theories with concepts, relations, constants - primary and defined. And concept grounding may be treated as calculation of concept definition on particular structure.
So the main question may be What kind of structures AGI creates inside and what kind of manipulation (knowledge processing) it is doing?
It seems the main feature of the mind is processing of colored 3D figures, particulaly in motion.
[1] https://en.wikipedia.org/wiki/Model_theory
[2] https://plato.stanford.edu/entries/model-theory/
In this case, structuring is not quite what (or not at all) what is meant in model theory.
Detection of objects in the environment (representation of a situation as a set of separate objects, the behavior of which can be predicted separately) is essentially a search task of potentially unlimited computational complexity.
Yes, this is a subtle topic. And for me if we take static for simplicity at-first, what we see is one large, variously curved and painted surface. But one. Reasoning for object separationis is a reasoning on this just one structure.
Consider the situation when all objects are glued and it turns out that, contrary to expectations, this is one such object.
Here is a nearest math but in Russian "Отображения гладких поверхностей на плоскость окружают нас со всех сторон. Действительно, большинство окружающих нас тел ограничено гладкими поверхностями. Видимые контуры тел - это проекции ограничивающих тела поверхностей на сетчатку глаза. Приглядываясь к окружающим нас телам, например к лицам людей, мы можем изучить особенности видимых контуров.
Уитни заметил, что в случаях "общего положения"* встречаются особенности лишь двух видов. Все другие особенности разрушаются при малом шевелении тел или направлений проектирования, в то время как особенности этих двух видов устойчивы и сохраняются при малых деформациях отображения."
http://mathemlib.ru/books/item/f00/s00/z0000032/st003.shtml
Thank for link!
We see those contours that are projections of the smooth surface of the body onto the "visual plane" due to the fact that they are illuminated. But lighting makes visible at the same time those features that are not projections of a smooth surface: irregularities that reflect light depending on the angle of incidence of light, edges, differences in color, etc.