On Meaning and the Culture of AI

July 30, 2017

In the recent New Yorker article “What the Enron E-Mails Say about Us”, Nathan Heller notes the following:

The tendency to weave stories where evidence is missing is the human brain’s sustaining feature, precipitating heroic actions, senseless love, and mindless hate.

Over the last decade, Heller explains, the corpus of Enron emails released to the public have become the data for a number of analyses about business etiquette and communication styles. Heller concludes that while these scientists and researchers may use these textual data to uncover something fixed and predictable (e.g., who uses “Dear” to open a message), the humanities—and, I would add here, most humanists—are more interested in meaning and how the body of Enron emails take on a particular meaning within their broader context. For instance, an email letting employees know about the company’s policies on proper corporate conduct sent ten weeks prior to the CEO’s resignation takes on a sense of irony for us. Heller suggests no model a data scientist creates would be able to “read” this sense of irony in such an email.

I’m not interested here in whether or not machine learning algorithms could be developed to ascertain these broader meanings that apparently only the human mind is capable of weaving a story around. Rather, I imagine such a model could be developed through the use of neural networks that would make meaning out of the data. The model could weave a story meaningful to the “black box” of the neural network, which is not able to be completely deciphered by the designers of the model.1 The inner meaning—the intermediate stages of information being processed by the system and the mechanism that are manipulating and gauging that information for usefulness—is like the brain to B. F. Skinner and other behaviorist psychologists: largely responsible for a response to some stimulus but how it processes the stimulus exactly cannot be fully known.

As a psychological anthropologist who is interested in how people make sense of themselves and the world around them, I am especially interested in narrative. Narrative is one way in which individuals take the cultural material around them (e.g., beliefs, customs, myths, tropes, stereotypes, etc.) and use it to construct stories about themselves and their societies. As such, narratives and narrative styles vary as cultures vary. By studying individual narratives, we can glean something about the cultural milieu in which one lives. Conversely, given a particular cultural context, we can often predict what kinds of tropes and narrative devices individuals will use to talk about their own life histories.2

An underlying presumption in the study of narrative is that narrative style—like culture for anthropologists—is relative, not normative. That is, it is not the job of the researcher to determine how well or how poorly a given person’s narrative of himself or herself makes sense or fits the way it’s supposed to be structured based on the researcher’s notion of a well-structured narrative. Rather, it is the job of the researcher to understand how the narrator is structuring his or her life history or narrative. With several samples of narrators from similar backgrounds, we can perhaps begin to see patterns emerge: common plot devices or turns of phrase or story arcs that are used the narrators. These patterns, then, can be used to understand how a particular culture or subculture makes sense of their world such as what it means to be a person or to have agency, a sense of being able to act and affect the world through those actions.

If an AI system is able to construct meaning out of the data it is fed that is meaningful to it but which may not have much meaning or explanatory power to the system’s designer, how is that similar to people’s personal narratives? If the system is drawing on heuristics only it fully “understands,” can we say that this AI system has created its own culture? And what would an ethnography of such an AI culture look like? I do not have answers to any of these questions. Instead, I’m simply raising them here in the hopes that myself or others will address them more in the future.

  1. https://www.youtube.com/watch?v=BFdMrDOx_CM [return]
  2. See, for example, McAdams, Dan P. (2006). The Redemptive Self: Stories Americans Live By. New York: Oxford University Press. [return]