A Comment on Consistency

Consistency defines things. It marks boundaries; it delimits where a thing ends and a different one begins.

Think about a door. In spatial terms, a door ends is delimited by its edges, where the gap between itself and the adjacent wall begins. But within the door, other things can be identified, e.g., the knob. Though part of the larger concept/object [door], it receives a differenciated treatment due to the internal consistency it possesses with respect to the rest of the door.

This concept of consistency is complimented by that of the negative nature of signs. Things, in so far as they come to act as signs, are largely defined to the extent to which they differ from all the other things they are not. Humans appear to tend to identify these features in their environment and make interesting things with them. Language, seen from a structural perspective, is perhaps the prime example of this assertion. Phonemes and morphemes are characterized each one as such insofar as they display certain distinctive features that distinguish them from the rest. The same is true for lexemes and full words.

In the world of ideas, logic looks for consistency in reasoning. An argument is said to be consistent if it does not entail a contradiction. Since Modernity, natural sciences have followed this approach and have extensively applied logic-based classifications to natural phenomena (e.g. taxonomy, the periodic table).

In the sphere of social affairs, consistency serves to define individuals. As with any other thing, we recognize our family and friends by their consistent appearance. Moreover, we tend to draw patterns in the behavior of people to characterize them, thus creating consistency. A smart person is not someone who once made a clever comment, but someone who consistently makes clever comments and consistently shows mental agility.

In discourse, consistency, also called coherence, delimits sentences, paragraphs, and whole texts in terms of the meaning they convey. In school, we learn to identify the main ideas of paragraphs, which usually correspond to headings and titles, when available. Failure to keep coherence when writing a text results in a very painful reading experience.

These last two faces of consistency merge weirdly in the world of politics and rhetoric. Drawing from Goffman, we can see interactions as a scenes, where everyone plays a character with a lifetime script. One has to be consistent with what has been previously said and with one’s self-image in order to succeed. Trying to change one’s line will most likely result in criticism for being inconsistent or at least in a lot of discussion. Politicians are professionals in the management of this kind of consistency and thus come more readily to mind when talking about this topic; however, everyone who is part of a community has to deal with politics and take care of one’s self image in one way or another.

Consistency, in any of its forms, is, for the most part, a good thing to have. Nevertheless, in the real world, fuzzy as it is, different forms of consistency from time to time will clash with each other, suggesting a certain irreducibility in their nature. Something that results particularly interesting to me on this regard is that, whenever politics is involved, more basic forms or consistency are often neglected in favor of sociopolitical and discoursive consistency. If I were to say today that I like a certain flavor of ice cream, while tomorrow I changed my mind, and the day after I went back to my original preference, I would be taken as a joke. In the same way, in a professional environment, if a manager decided to start a project just to cancel it the next week and to restart it the week after, he would most certainly come under fire from both his superiors and subordinates.

A similar inconsistency in the decisions taken by a head of state would very likely cause a political crisis. Indeed, in many cases when such inconsistency exists, it is due to either mental instability or a lack of competence. But why do other types of consistency are irrelevant in these cases? What if the inconsistency of the manager are caused by sudden changes in the market and a consistent principle of doing what is best for the economy of the company? What if my change in taste is due to some hidden consistent principle that I am aware of? It could be argued that we are only ready to take certain consistent behaviors as such. If I eat carrots every day, I will be known for my habit of eating carrots. If I eat carrots every day in odd-numbered months and tomatoes in even-numbered months, I will be known for my consistent change of habits. If I do the same in odd-numbered years and change between carrots and tomatoes every two months during even-numbered years, despite it being a consistent change in the consistency of change of habits, figuring out the pattern would require a good deal of observation; in a real-world situation, it would simply be considered weird and inconsistent.

Reflecting on this issue is of special importance for the development of AI, which is currently one of the hottest fields in science. What type of consistency should AI systems give preference to? Think about an image classifier AI that encounters a transexual person. Should it be categorized according to what it sees or what the social consensus says? Now, consider a dystopian future in which machines are now able to take high-impact political decisions. One day, in a country with an aging population, a query is made to the machine in terms of what the life-preserving policies should be. The machine possesses flawless logic reasoning and knows more about the world than all the encyclopedias ever written combines. After processing the problem, the deliberation is printed out: Penalize abortion, create policies to deter suicide in young citizens, and promote euthanasia for citizens over 80 years old. These policies, logically consistent and designed to solve the demographic issue of the country, would certainly cause a lot of controversy and outrage.

What does this problem entail? On one side, it suggests, as hinted above, the existence of irreducible types of consistency. Logical consistency follows a priori rules, while social and political types of consistency are a posteriori. No one in his right mind would dare to challenge the consistency of an instance of Modus Ponens. Yet, determining whether assisted suicide is unethical depends on social conventions.

On the other side, we find that consistency is a necessary but not sufficient condition for the formation of identities in social contexts. The case of color terms in natural languages provides a good example of this point. It is well known that the repertoire of color terms varies among languages. Regardless of whether we adhere to linguistic relativity or not, we can agree that whenever a color is shown to any person who is not visually impaired, a single color will be perceived, and not two or more. The consistency perceived in this case is understood as a particular object or token. The differences in the availability of color terms in human languages come when a difference is introduced. A gem is described as green or blue in a language inasmuch as it contrasts with other existing colors and it resembles other tokens belonging to the same class.

Thus, on an initial level, consistency leads to the recognition of tokens. Contrasting two tokens gives place to identity. Resembling tokens can then be grouped by virtue of a higher-level consistency, leading to the formation of classes. I will argue that the construction of classes is a private affair and does not require the intervention of sociocultural factors. However, labeling — seen as the delimitation of boundaries among identities — does appear to rely on social constructs. Thus, in order to be able to communicate, private categories are overridden by social classifications, which are often arbitrary. This leads to the question as to whether these socially created labels can in fact alter the way things are perceived — put in our terms, whether defining an entity beforehand leads to the perception of a different consistency. And along those lines, we can ask ourselves, what point is consistency in the eye of the beholder? And what can we say about social consistency if one is constantly changing masks depending on the situation and the environment, as Goffman suggested? Is personality then a social construction of a collection of social constructions?

The more we know about flowers, the more types of flowers we can recognize. This is not true for colors, however. Sociopolitical issues resemble the case of flowers. Digging into social issues has led to important changes in the way society is understood. New identities are now recognized. Freedom of choice in all respects is now the norm. Aren’t we allowing identity precede consistency in these cases? What are the benefits and dangers of doing so? Does this mean we will have to tune the parameters of the policy-making machine every time a new Zeitgeist comes in? Will the image classifier identify me as a different person depending on the role I am playing at the time?

AI thus faces a major challenge with major repercusions. Drawing from its logical nature — representative of modern era, it will have to find consistency in the current postmodern world, where, consistency at times appears to be highly volatile.

“In this world, bonds are dissembled into successive encounters, identities into successively worn masks, life- history into a series of episodes whose sole lasting importance is their equally ephemeric memory. Nothing can be known for sure, and anything which is known can be known in a different way — one way of knowing is as good, or as bad (and certainly as volatile and precarious) as any other. Betting is now the rule where certainty was once sought, while taking risks replaces the stubborn pursuit of goals.”
(Zygmunt Bauman: The Making and Unmaking of Strangers)