Evolutionary Adaptive Machine Intelligence - EAMI
An elegant machine intelligence described by David Heiserman in TRS-80 Color Basic that which cannot lose its luster.
The description of a hierarchical-classed artficial creature follows a well-defined pattern:
- Alpha Class: Knowledge of present conditions only.
- A machine of this type shows only reflexive responses to changing environmental conditions. It lives "in the moment" and has no memory of past experience.
- Beta Class: Knowledge of past and present conditions.
- A machine of this type can remember and use successful responses of the past to help solve problems of the present.
- Gamma Class: Knowledge of past, present, and future conditions.
- A machine of this type lives in the present and has recall of past events. It has the ability to anticipate future events by drawing upon remembered experiences and generalizing them toward future conclusions.
- Delta Class: Knowledge of past, present, and future conditions with the ability to plan outcomes.
- A machine of this type contemplates the kind of reactions to phenomona it has experienced while transiting the prior classes. Hence, it is intrinsically a discrete evolution.
The description of the algorithm, in totality, is broken-up into evolutionary complexities, called creatures. These 'forest' creatures have an increasing complexity to the point that they are autonomous programs who learn by Reinforcement Learning I have encapsulated as Ideal where it is going into novel territory. This code repository and evolutionary complexity serves as a foundational perspective on self-governing programs where the ultimate aspect is one that is homeostatic, or when is manifest as a presence. See US20180204107A1.
The Alpha-Class Machine
All Alpha-Class machines make purely random responses to changes they sense in their environments. The ALPHA-I creature has only one sensory mechanism and one way to respond: It senses only light-colored segments on the screen and responds by attempting to get away from them.
Alpha-Class adaptive machines need not be limited to single sensory and response mechanisms. There can be any number of sense/response mechanisms, and there are no limits on how simple or complex those mechanisms can be. The defining feature of an Alpha-Class machine is that it makes random responses or, in a few cases, pre-programmed responses. The heirarchy of Alpha-Class machines can be divided into subgroups such as Alpha-I, Alpha-II and so on. An Alpha-I version is the simplest possible Alpha-Class machine. ALPHA-I senses two, mutually exclusive conditions in its environment, namely being in a condition of running into a barrier or in a condition where it is free from any barrier contact. Furthermore, ALPHA-I responds to any barrier-contact situation with a negative "flight" response. The only response mechanism is one aimed at getting the creature away from such barriers.
The antithesis of running into a barrier is being free from such a condition which is a more desirable condition. It responds by performing the response that works. In other words, an ALPHA-I that is free from negative conditions in its environment keeps on moving the same way. An Alpha-II machine is a bit more sophisticated. It has the ability to distinguish a larger number of environmental conditions and react accordingly. The reactions are still randomly selected, however. Specifically, the ALPHA-II creature introduced in this chapter can sense the difference between two different kinds of lighted segments on the screen, lighted segments that flash and lighted segments that do not flash. In this case, lighted segments that flash are considered positive, desirable elements of the environment-running into a flashing lighted segment is "good". Running into a non-flashing lighted segment, however, is still considered "bad", and ALPHA-II responds accordingly.
The "good" flashing light on the screen is going to be a nest where the creature presumably gathers nourishment or engages in some other form of activity vital to its survival. Of course, this is a computer simulation and as such isn’t absolutely necessary. However, if ALPHA-II were a real machine, the "nest" could represent a battery charger or a similar kind of energy source.
The border is made up of non-flashing line segments and is thus considered something to be avoided. The next rectangle near the center of the screen, on the other hand, flashes one time whenever:
- It is turned on, and,
- The creature makes contact with any lighted segment on the screen.
Whenever the nest is turned off, it no longer flashes and is treated as a barrier to be avoided. Whenever the creature senses contact with the nest and the nest is turned on, the creature responds by resting at the nest until it is turned off. In a sense, this resting response is an in-born reflex, the first and one of a very few pre-programmed responses. Having this one pre-programmed response built into an Alpha-Class machine does not really violate the purity of the adaptive machine. Alpha-Class machines do not have the capacity for learning how to respond to a positive influence of this kind, and in the real world, an inborn reflex of this kind would be absolutely essential to survival. Rest assured, higher-order machines such as Beta-I and Beta-II will have to learn to deal with the nest situation on their own, first by trail and error and then by purposeful intent.
- ALPHA-I ReadMe
- Basic
- Scoring
- Navigation (maze)
- Strategy (paint)
- Confidence (compile)
- ALPHA-II ReadMe
- Basic
- Stratgy (feeding)
- Priority (nest)
- Confidence (compile)
- Track-Sense
- Multisense
- Sense-Response
- Duality (unbiased Anticipation data)
The Beta-Class Machine
A Beta-Class machine still relies heavily upon elements of Alpha behavior, which are contained as subroutines tucked into the Beta flowcharts and programs. The big difference is that a Beta-Class mechanism couples Alpha-Class behavior with an ability to remember events in the past. And of course, it not only remembers past events, but also has the capacity to act upon those memories at any time in the future. Generally speaking, a Beta-Class mechanism is an Alpha creature that is coupled to a memory scheme.
- BETA-I ReadMe
- Basic
- Scoring
- Confidence (decision seed)
- Scoring (memory)
- Confidence (memory-seed-compile)
- BETA-II ReadMe
- TBD
The Gamma-Class Machine
A Gamma-Class robot includes the reflex and memory features of the two lower-order machines, but it also has the ability to generalize whatever it learns through direct experience. Once a Gamma-Class robot meets and solves a particular problem, it not only remembers the solution, but generalizes that solution to a variety of similar situations not yet encountered. Such a robot need not encounter every possible situation before discovering what it is supposed to do; rather, it generalizes its first-hand responses, thereby making it possible to deal with the unexpected elements of its life more effectively. To get a clearer impression of how a Gamma-Class robot learns and thinks, suppose its initial successful response to a bright flashing lightis to attack it and knock it over. The machine generalizes that particular aggressive response to a wide range of different environmental situations involving bright flashing lights in particular and, perhaps, any sort of light in general. It is impossible to say how deep this experience will penetrate the robot's view of his world. To be sure, this particular machine will exhibit elements of aggressive behavior that might even extend to situations other than those directly related to bright flashing lights.
- GAMMA-I ReadMe
- General structure
- General template architecture
- Runtime presence
The Delta-Class Machine
One further refinement that is entirely possible today is extending the Gamma-Class robot's level of self-awareness to the point where it can evaluate the responses it is learning. This Delta-Class robot assigns a certain confidence level to the responses it deems appropriate under certain environmental conditions. The solution to a problem that is solved by first-hand experience takes on a rather high confidence level. As in the case of a Gamma-Class robot, the machine generalizes its responses to first-hand solutions to similar conditions not yet encountered. This robot assignsa relatively low confidence level to these conditions.
So as the Delta-Class robot experiences events in its world, it evolves a hierarchy of responses having a high, medium, and low level of confidence. The machine, however, does not like responses with a low confidence level to exist within itself, so it sets out to encourage those events that call for a test of its responses gained by inference or second-hand generalization. The delta-class machine might not be wholly successful in its attempts to set-up new environmental conditions, but it works at the job ceaselessly. It has a will of its own: It programs itself, setting goals aimed at enhancing its confidence in dealing with the world as it perceives it.
- DELTA-I ReadMe
- TBD
Multiplatform application versions of the sublime TRS80 emulator are in this folder. Extract the app for the platform and copy to the root directory of the various creature evolutions. The .gitignore file will keep this from being checked-in. Run a script in the following manner.
.\trs80gp experiments\alpha-1\ch-3\32-alpha1-basic.bas
And an emulator screen will appear.
Further:
.\trs80gp experiments\alpha1\ch-6\63-alpha1-compile.bas will ask for parameters, such as the number of cycles to run.
Then run the analysis.
Perform its intuitive-themed task.
And output the work results of the presence's experience as-per this particular experiment.
A modern version is tabled here.