Faisal Kadri: Artificial Psychology and Cybernetics
CADM was a great exchange of ideas and conversations, I enjoyed making new friends and hope to be in touch with all of you.
Here is my take on how Artificial Psychology (AP) could develop within Cybernetics, comments please!
Artificial Psychology as First and Second Order Cybernetics
Faisal L. Kadri
First order cybernetics is concerned with objects while second order cybernetics includes the observer in the role of the constructor of the image of reality. Artificial Psychology (AP) simulates personality with motivational and cognitive behavior, studying one AP model mirrors first order cybernetics; the personality model is the object. When two AP models are combined in an interactive duo, the system of the two forms one complex personality model: One acts as the ego, the other as the image of reality, mirroring second order cybernetics.
Introduction to first and second order Cybernetics
Different researchers describe first and second order cybernetics differently, here are some samples which hopefully will converge on concepts relevant to artificial psychology:
Fig. 1 Famous first and second order cybernetics schematic.
- The cybernetics of systems that are observed from the outside as opposed to the cybernetics of systems involving their observers (von Foerster), see Fig. 1.
- First-order cybernetics is concerned with circular causal processes, e.g., control, negative feedback, computing, adaptation (Krippendorff).
- What must be explained of second order cybernetics: How an individual constructs reality (Umpleby).
- Constructivism: the philosophy that models are not passive reflections of reality, but active constructions by the subject (Heylighen).
- The implication of constructivism is that since all models are constructed by some observer, this observer must be included in the model for it to be complete (Heylighen).
Clearly the observer in second order cybernetics constructs an active and adaptive image of the world, or as von Foerster put it: The world, as we perceive it, is our own invention. This is the main concept I will rely upon for developing Second Order Artificial Psychology (SOAP)!
What is Artificial Psychology (AP)?
AP is a simulation of human behavior that mimics motivation as well as cognition. AP can reference (objective) empirical personality measures for validation and viability. How does AP fit in with first and second order cybernetics, and what sets it apart from AI and other approaches? Here are some suggestions:
* The addition of motivation
Motivation, as its name root means, changes with time. Defining AP in terms of motivation implies dynamic behavior and internal change even in the absence of instructions from the outside. The path of AP development started from studying animal motivation as a continuous nonlinear dynamic process, which describes temporal behavior. In contrast, the simulation of human behavior as described in Herbert Simon’s influential work (1967) suggests that the emotions, motivations and drives, including “hunger-thirst”, can be simulated by interrupt systems driven by events and acting on (linear) negative feedback mechanisms. This concept fed much of current AI activity and designs of chatterbots, and provided for self-reference but no perceived possibility of validation for coherence and invariance. Whereas AP relied on continuous nonlinear systems model of priming and homeostasis, particularly the alternative multiplier feedback regulator, without the requirement for interrupts, and consequently found parallels in personality psychology and humor research which allowed for tests of validation through statistical analysis of survey data.
* Restricting the media
Unlike cognitive media often used as a starting point in AI, such as the word or the syllable, and other media not based on text such as music and the arts when seen as semiotic symbols, the sentence was found as the medium that can be classified according to motivational scales. The restriction of processing only sentences made it possible to proceed with currently available programming tools.
* Referencing empirical personality measures
The choice of the sentence as the media for AP provided commonality with personality measures as performed in profiling tests and surveys. Here, a test “instrument” often comprises sentences scored by the subjects and collectively define personality dimensions or types. By cross-referencing personality scores AP measures are compared with established concrete personality measures, therefore providing a path for validation. The model used by AP can also be directly referenced against psychological data when the model predicts a preference pattern which can be validated statistically against survey data. Here, the artificial personality model predicted an age preference pattern, which when applied to types of humorous sentences can be tested against humor appreciation survey data. The prediction was tested over several years and using three on-line surveys; in all tests the data analysis supported the prediction, and in a statistical measure of coherence a level of 75-93% probability of consistency was recorded. There was no reliance on self-referencing in the tests; if the statistical evidence did not support the AP model’s prediction then the model can be deemed wrong, which contrasts with AI models such as ALICE which relies on self-referencing; you cannot dispute ALICE’s correctness.
* The parallel with second order cybernetics
In the AP software implementation George and Mary (GaM), two artificial personalities exchange sentences and motivational re-enforcements. Looking at the personalities separately, each (simplex) personality can be seen as first order cybernetics where the personality model acts as an object or a self contained system. When two artificial personalities are combined in an interactive duo (duplex), such as in GaM, the duplex of the two can form one complex personality: One simplex acts as the ego and the other as a virtual imagination or the image of reality. Dialog emanating from the ego and addressing person A is copied to the virtual imagination where an image of person A resides. When person A responds, his/her sentence is checked against his/her image before being transmitted to the ego. The parallel with second order cybernetics is clear. The sentence checking process provides the means to recognize if person A is conversing out of character, in which case his image can be updated in an adaptive manner. The process can include parser based AI unit, parsers are an area of high research activity in AI, thus existing AI research complements AP, not replace it. Simulating normal psychology, the degree of adaptation or learning ability can be specified by a constant as suggested by Umpleby, the constant can be a characteristic of the duplex and has a value range of 0 to 1 to indicate the speed of learning. Here are three possible scenarios to simulate psychopathologies, Schizophrenia may be simulated in part by making the artificial imagination hyper active while stifling communication with the outside world. The ego which normally accesses only one repertoire of sentences could be made to have multiple, simulating multiple personality. Autism’s telltale symptom is having no empathy or imagination of the others and how they feel, this is a clear reference to the functioning of the imagination which could be modeled by suppressing the activity of the artificial imagination.
Artificial Psychology (AP) belongs to the field of cybernetics. Moreover, unlike Artificial Intelligence (AI) and other cybernetics models, AP can reference concrete empirical data and validate or disprove its models.
Having parallel with second order cybernetics, AP may have significant potential in modeling normal and abnormal psychology, and improve the understanding and teaching of otherwise difficult psychological conditions.