The field of computational intelligence has enjoyed much success in developing a variety of algorithms that emulate human cognition. However, a framework to tie these algorithms together in a coherent manner to create a machines that possess the full spectrum of human-like personalities is still needed. To date, research on artificial personality synthesis has focused on using the Big Five model from the field of personality psychology. The overlooked Achilles heel of Big Five (BF) is that it is purely datadriven model, and thus offers only marginal guidance on how a machine with a personality might actually be created. In this work an alternative computational personality framework is presented based on the work of Carl Jung. There are two key insights that suggest a Jungian type-based framework is suitable for synthesizing an artificial personality. First, the cognitive functions which form the building blocks of the Jungian personality model can be mapped to classes of algorithms used to emulate cognition. Second, the Jungian framework suggests that at any given time humans are only using one of the cognitive functions. This suggests that a human personality could be emulated using a state machine with each state implemented using the appropriate class of algorithms.
• Human-centered computing➝Interaction concepts and paradigms design theory,
The advent of ubiquitous computing has increased interest in techniques for endowing a machine with a human-like personality. Vinciarelli [1] provides an overview of personality computing research which shows that since 2006 a marked increase in research papers with the word “personality,” included in the title. Vinciarelli makes one particularly interesting statement with regards to the nature of the work that has been undertaken thus far, “Trait based models are widely accepted in the computing community as well. All of the works surveyed in this article adopt personality traits (the BF in 76 cases out of 81) and, to the best of our knowledge, no other theories were ever adopted in computing oriented research. On one hand, this barely reflects the dominant position of trait based models in personality psychology. On the other hand, trait-based models represent personality in terms of numerical values, a form particularly suitable for computer processing. [1].” This statement raises an interesting question. Why has computational personality research thus far restricted itself to exploring trait-based models of personality? Vinciarelli points out a number of competing personality models including: psychoanalytic, cognitive, and behaviorist and biological. Arguably, personality models such as the biological model may currently not be adequately understood to be implemented on a computer, and as Vinciarelli points out, the numerical nature of trait-based models may be amenable to computer implementation to a degree. However, the nature of the Jungian type-based model also lends itself to implementation on a computer, but to the best knowledge of the author has not been explored to date. This work will outline how Jungian psychological type theory can be used as guidance to synthesize an artificial personality.
This Jungian type-based framework possesses a number of attractive properties. First, a very large fraction of algorithms emulating cognitive processes can be utilized by the framework in a coherent manner. Second, the framework can arguably synthesize the full spectrum of human personalities. Third, Jungian-based personality theory is very popular among laypeople. Modern personality psychologists might argue that this is due to the cognitive bias known as the Barnum effect. From a Turing test perspective however this is irrelevant, and perhaps even an advantage because it would facilitate the illusion of a machine having a human-like personality For practical applications all that really matters is that the machine can convince a human that it has a personality.
Carl Jung is probably most popularly known for introducing the concept of the “introvert,” and “extrovert.” Jung detailed his theory of psychological types in 1921 [2], [3]. Jung’s model for personality is based on the idea of “cognitive functions.” Jung identified two fundamentally different kinds of cognitive functions known as “perception,” and “judgement. [4]” Perception describes how a person takes in information, and judgment pertains to decision-making. Jung then broke these classes down further. Jung asserted that perception came in two main forms: “sensation” and “intuition.” Sensation is focused on physical reality:. It tends to focus on the present and past. Intuition is primarily focused on finding meaning, patterns, or possibilities in information. The tendency is to focus more on the future [5]. Likewise, Jung identified two forms of judgement which he referred to as “thinking,” and “feeling.” Thinking refers to decisionmaking processes that focus more on the application of basic truths/principles. It tends to impersonal in the sense that it resists allowing personal value judgements, or the value judgements of others influence decision making. Conversely, “Feeling,” puts significant weight on values. These values can be either personal or shared by a community. It tends to prefer decisions that will result in harmony [6]. Jung then further refined his cognitive functions by asserting that each of the four functions has an introverted and extroverted orientation. An introverted orientation implies a tendency towards a person’s interior world of thoughts, ideas, feelings and memories. An extroverted orientation focuses on people or experiences external to the self [4]. Jung’s clinical observations and reflection ultimately resulted in a total of 8 cognitive functions. For completeness the eight cognitive functions are:
Extroverted Thinking Introverted Thinking Extroverted Feeling Introverted Feeling Extroverted Sensing Introverted Sensing Extroverted Intuition Introverted Intuition (Te) (Ti) (Fe) (Fi) (Se) (Si) (Ne) (Ni) For the sake of brevity, a full description of Carl Jung’s 8 cognitive functions will not be provided in this document. Since Jung initially introduced the concept of cognitive functions, the language used to describe them has evolved as has an understanding of their nature. For the purpose of this work, the cognitive function descriptions provided in [4] will be used throughout this work.
In 1923, Katherine Briggs and her daughter Isabel Briggs Meyers were exposed to the newly available English translation of “Psychological Types” [7], [8] At the time Katherine was in the process of developing her own personality theory motivated partially by a desire to understand the personality of her son-in-law. Upon reading Jung’s work they came to the conclusion that Jung’s theory was superior to their own and so decided to adopt and refine the Jungian model. Over the next 20 years the motherdaughter team obsessively observed and documented human nature with regards to type. They eventually created a variation on Jung’s system with an associated sorting instrument known as the Myers-Briggs Type Indicator (MBTI).
The evolution of Jung’s psychological type proposed by Myers and Briggs had a few important characteristics that are worth mentioning. First, Myers and Briggs observed that all people used all of the cognitive functions. The dissimilarity between different types of people was the preference with which they used the cognitive functions. In this model each person uses the cognitive functions in a hierarchical manner with a dominant cognitive function, followed by an auxiliary, tertiary and inferior cognitive functions. Classically, Meyer & Briggs focused on the first four cognitive functions. Furthermore, Myers & Briggs introduced very specific constraints on the hierarchical order the cognitive functions were allowed to assume. One of the constraints Meyers & Briggs introduced was that the hierarchy of cognitive functions had to alternate between introverted and extroverted orientations. It is worth noting that the writing of Carl Jung can be interpreted to indicate a different scheme for ordering the cognitive functions, and competing ordering systems are in existence. For the sake of clarity, the author tends to prefer the ordering system outlined in [4]. However, with regards to artificial personality synthesis, the framework outlined in this work is flexible enough to be adapted to any desired cognitive function ordering scheme.
Ultimately, based on their imposed constraints on cognitive function order, Meyers & Briggs identified 16 unique personality types. These personality types were given fourletter labels. The first letter is either E or I to indicate an introverted or extraverted orientation of the dominant cognitive function. The second letter is either S or N to indicate the dominant perception preference of sensation or intuition respectively, the third label is T or F to indicate the preferred judging cognitive function of either thinking or feeling, and the last letter is either J or P to indicate whether or not the preferred perception function has an introverted orientation (J) or an extroverted orientation (P). An example of a personality label from the use of this model would be ENTJ. This would indicate a personality who’s dominant function is extroverted thinking with introverted intuition as the auxiliary perceiving function. The constraints imposed by Meyers & Briggs on the cognitive function order would then further specify that the tertiary function is extroverted sensing and the inferior function is introverted feeling.
There is a very widely held misconception that each of the four letters indicates a dimension of personality, like that found in the Big-5 model. It cannot be stressed enough that the four letters used to provide a personality label are in no way representative of dimensions. It is more appropriate the think of each grouping of four letters as a label. The Jungian model is not based on the concept of dimensions in any way. It is based on the concept of cognitive functions and the hierarchical preference with which people with different types of personality use them. Another common misconception is that the Jungian/Meyers & Briggs model implies a binary distinction between -for instance- thinkers and feelers, or judgers and perceivers. The common criticism is that the model implies that a person solely uses only one class of cognitive function or the other. For example, that a person is either a thinker or a feeler. Once again this is not how the theory works. The theory indicates that all people have access to use all the cognitive functions. It is just that people have different orders of preference for different functions. Alternatively, Berens and Nardi [4] explain the preference in terms of energy expenditure. They describe the use of a given cognitive function as requiring the expenditure of more or less energy depending on a person’s personality type. Thinking of cognitive function use in terms of energy usage is a very convenient way to guide the selection of cognitive function to use in a given situation because it interfaces well with computational thinking on cost functions in optimization as well as with results in psychology that suggest that selfregulation relies on glucose/energy levels [9].
The proposed Jungian type-based framework for artificial personality synthesis is based on three key principles. 1. Algorithms that have been developed to emulate cognition (e.g. Principle Components Analysis [10], Artificial Neural Networks, Linear Classifiers, etc.) can be mapped to the cognitive functions that make up the building blocks of the Jungian personality model. These algorithms can be used to implement the cognitive functions. 2. Human personality is inherently serial in nature.
Human personality arises from the limitation that humans can only use one cognitive function at any given time. Or at least the human ability to use more than one function at a time appears to be severely limited. 3. The Jungian type-based personality framework allows for the possibility of an individual agent to use any of the 8 cognitive functions at any given time, however, personality emerges from a hierarchical preference for certain cognitive functions over others. The order and magnitude of preference can be selected based on models such as the Myer & Briggs cognitive function orders [4].
They could also possibly be learned from data, or could even be chosen arbitrarily. Initially the authors suggest using the Myers & Briggs cognitive function orders as guidance. is less common for the ENTJ machine to use extraverted sensing (Se) and it rarely if ever transitions to the introverted feeling (Fi) state. Personality is emulated by executing the algorithms that correspond to the current cognitive function. It is worth noting that alternatives to the Markov chain representation could be envisioned. For instance, instead of probabilities, the proclivity to transition to another cognitive function could be couched in terms of energy costs associated with the use of a cognitive function for a given personality type as discussed in [4]. Also, please note that only 4 cognitive functions are shown in Figure 1. For completeness all 8 cognitive functions could be included. Only the first 4 functions were used as is specified by the classic Myers and Briggs framework for the sake of clarity. Lower order functions were not included. In initial implementations of this framework it might be preferable to only use the first 2 highest preference cognitive functions for the sake of simplicity. The reason being that the first two cognitive functions are all that is needed to distinguish one personality type from the other 16. Furthermore Jungians tend to believe the first two are the most important.
In order to implement the Jungian type-based personality framework, it is necessary to map algorithms that emulate cognitive processes to the cognitive functions. Table 1. provides a listing of the 8 cognitive functions, a short description of each of the cognitive functions, and a list of possible algorithms that could be used to emulate the cognitive function. This list is by no means exhaustive, and revisions are expected as artificial personality research progresses. The descriptions provided in Table 1. borrow heavily from [4] in order to maintain some uniformity in descriptions. The introverted and extroverted feeling judging functions merit additional discussion. The feeling judging functions are often associated with emotional response. The author currently likes to think of emotional decision-making in general to be similar to an artificial neural network in the sense that an artificial neural network can often take many complicated inputs and learn relationships between the inputs that can be used to quickly make decisions, however it is not always clear how exactly those decisions are made. The author also prefers to think of the emotional cognitive functions as being partially the results of an embodied intelligence. Feeling judgements can be thought of as using the body as an analog computer to perform simulations and make decisions. Work by Nummenmaa [10] on mapping the sensation of emotions felt in the human body lends some support for this perspective.
As an example of the application of the framework, a machine endowed with an ENTJ personality using the Markov chain in Figure 1, we might employ the A* algorithm to perform the dominant extroverted thinking (Te) cognitive function to form a plan to move through an environment. After the plans are generated they might be analyzed in the introverted intuition (Ni) state using an appropriate simulation that corresponds to the environment and task of interest. If the simulation verifies that the plan is acceptable the plan may be executed by the extroverted Sensing function. In many cases the machine may totally ignore the introverted feeling state and proceed directly to execution, but when it does go into the introverted feeling state it may use a neural network trained to look out for the machine’s own well-being to decide whether or not to actually execute the task based on whether or not the machine “feels,” it will advance the machine’s self-interests. It is also worth noting that creating the ability to endow robots with personalities may also have application in human machine teaming applications. Much as diversity in teams of humans tends to lead to better results [4], endowing teams of robots with different personalities may also lead to more robust robotic teaming. For example, a team consisting of humans could be augmented with machines endowed with personalities different from the existing team members in order to enhance the overall team diversity. Teams consisting only of robots could be designed in such a way that individual team members are endowed with complimentary personalities, thus enhancing the overall robustness and performance of the machine team.
Since the MBTI was developed more than 10,000 companies,
2500 colleges/universities and 200 U.S. government agencies
have used the test [11]. It is estimated that more the 50
million people have used the instrument since 1962. Jungian
personality theory has had a great influence on corporate
America as well as popular culture. For instance, Carl Jung
popularized the terms introvert and extrovert. Despite the
popularity of Jungian personality theory in industry and
among laypeople, academia tends to discount it. One common
criticism against the MBTI is that it lacks test-retest
reliability [12]. The current perspective of the author is that
this criticism is probably valid. At this time the instrument
itself appears to have problems. The reason for the lack of
reliability with the current instruments may be that the
instrument is based on the analysis of a self-report inventory.
This type of instrument may be suitable if a Cartesian model
of personality is used, but the Jungian model is better
described as a dynamic system. Ultimately enhanced versions
of projective tests such as those suggested by Ottley [
A number of interesting results have come out of the
psychology and neuroscience communities that lends some
support for the idea of using the Jungian type model as
outlined in this work. The idea that humans can effectively
only use one cognitive function at a time gains support from
the results of Watson [
Personality from a Jungian perspective can be thought of as a zero sum game. Humans only have limited cognitive resources, and our personality is based on how we tend to choose to use those resources. This proposed framework is particularly attractive because it uses established algorithms as building blocks for personality. Because the building blocks are algorithms, and in some cases learning algorithms, the machine is ultimately able to learn and adapt to experiences. The personality framework provides a genotype so to speak but the ultimate phenotype of the machine depends on the experiences it encounters throughout its span of existence. This framework allows for great diversity in resulting perceived personality phenotype.
An interesting implication of the Jungian type-based
personality framework is that it might help in the
development of robust, high-performance human-machine
teams consisting of members with diverse personalities.
Observations on the personality diversity of teams and their
performance suggest that teams consisting of members with
diverse personalities tend to perform better [
Experience has shown that data science problems often benefit
from the use of a combination of many heterogeneous models.
The Netflix prize provides a good example of showing the
advantages associated with simply combining different
approaches [
Artificial personality synthesis is at a similar point in
development as the airplane was at the time of the Wright
Brothers. You do not need to understand how a bird flies in
order to build a transatlantic aircraft, and you do not need to
understand how the human mind works in order to build a
machine that has a recognizable personality. Finally, it is
interesting to note that while the Jungian perspective on
artificial personality synthesis presented here is very different
from the mainstream views on the topic that center on the use
of Big-5 theory, this paradigm is not any more controversial
than the Freudian paradigm advocated for by Marvin Minsky
[
This work was completed under the support of a Los Alamos National Laboratory Early Career Research Award #20140629ECR. The concepts presented in this work were refined with the help of Sebastián A. Zanlongo.
[1] A. Vinciarelli and G. Mohammadi, "A Survey of Persoanlity Computing," IEEE Transactions on Affective Computing, vol. 5, no. 3, pp. 273-291, June 2014. [2] C. G. Jung, Psychologische Typen (Psychological Types), 1 ed., Zurich: Rascher Verlag, 1921. [3] C. Jung, The Portable Jung, J. Cambell, Ed., New York: The
Viking Press, 1976. [4] L. V. Berens and D. Nardi, Understanding Yourself and Others: An Introduction to the Personality Type Code, Huntington Beach, CA: Telos Publications, August 1, 2004. [5] The Myers and Briggs Foundation, [Online]. Available: http://www.myersbriggs.org/my-mbti-personality-type/mbtibasics/sensing-or-intuition.htm. [Accessed 1 June 2015]. [6] The Myers & Briggs Foundation, [Online]. Available: http://www.myersbriggs.org/my-mbti-personality-type/mbtibasics/thinking-or-feeling.htm. [Accessed 1 June 2015]. [7] The Meyers & Briggs Foundation, [Online]. Available: http://www.myersbriggs.org/my-mbti-personality-type/mbtibasics/isabel-briggs-myers.htm. [Accessed 1 June 2015]. [8] Center for Applications of Psychological Types, [Online].
Available: http://www.capt.org/mbti-assessment/isabelmyers.htm. [Accessed 1 June 2015]. [9] M. Gailliot, R. Baumeister, C. DeWall, J. Maner, E. Plant, D.
Tice, L. Brewer and B. Schmeichel, "Self-control relies on glucose as a limited energy source: willpower is more than a metaphor," J Pers Soc Psychol, vol. 92, no. 2, pp. 325-336, February 2007. [10] K. Pearson, "On Lines and Planes of Closest Fit to Systems of Points in Space," Philosophical Magazine, vol. 2, no. 11, pp. 559-572, 1901. [11] L. Nummenmaa, E. Glerean, R. Hari and J. K. Hietanen, "Bodily maps of emotions," Proceedings of the National Academy of Sciences of the United States of America, vol. 111, no. 2, pp. 646-651, November 27 2013. [12] L. Cunningham, April 13, 2013. [Online]. Available: http://www.seattletimes.com/business/myers-briggspersonality-test-embraced-by-employers-not-allpsychologists/. [Accessed 1 June 2015].