This is the third in an occasional series of “moonshot ideas” from Masculinity Research that propose large high-level solutions to the “problem” of masculinity. This article was first published at The Good Men Project.
When we think about Artificial Intelligence (AI) and gender we tend to think about how it is often given a feminine personality in popular culture. Two classic examples of this would be the robot Ava in the movie Ex Machina and the operating system Samantha in the movie Her. AI-enabled robots are often cast in a sexualized female role, which of course feeds into the perceived male fantasy of on-tap and emotionally uncomplicated sex.
None of this should come as a surprise given that technology and computing tends to be a male-dominated domain whose history has morphed from the politically innocent nerd culture of previous decades into the politically contentious discussion surrounding gamergate. But what can we learn about gender if we thought a bit deeper about the possibilities of AI?
GIGO: Gender In, Gender Out
Computer programmers have a phrase “garbage in, garbage out” (GIGO), which suggests that if you enter poor quality data into a program you will get a poor quality outcome. Let’s co-opt this acronym and re-cast it as “gender in, gender out.” This new form of GIGO suggests that whatever patterns of gendered behavior that are present in the people who create the first AI will be transmitted into the AI.
Those people who are clever enough to be AI developers are absolutely aware of how they imprint psychological aspects into the potential AI. They might choose to replicate certain gendered traits in order to make the AI appear more human, which in turn could facilitate our ability to interact with it. Alternatively, the developers may seek to erase gender from the AI. However, the ability to succeed in such a task would require a level of awareness about how all individuals are conditioned by gender (both consciously and unconsciously) that is probably beyond the abilities of most AI developers.
In order to mitigate this conditioning it would be necessary to perform what might be called a “gender audit” on the vocabulary programmed into the AI. In the most basic sense, this would require, for example, the AI to know that “pink” is at once a color, a signifier for femininity, and a slur on male homosexuality. Or we might want the AI to know that an axe is used to cut down trees, but is also a signifier for frontier masculinity, and perhaps also a prop for a more metrosexual hipster masculinity. There is a gendered field of meaning around a good deal of our common vocabulary, and this would have to be teased out: a large task, for sure, but a simple one relative to the overall architecture of the AI.
If the gendered nature of vocabulary could be instilled in the AI then it would have the ability to pursue its learning processes and develop its identity by countering the gendered nature of what it experiences. In order to create some form of comparison, you could run two instances of the same initial AI: one with the filter for gender and one without. Both AIs could then be exposed to the same stimuli and we could see how they develop differently.
There are various implications to this exercise that are largely derived from whether or not any identifiable form of gender becomes apparent in the AI running the gender filter. For example, if the AI forms a clearly-identifiable gender despite running an anti-gender filter, it would greatly support the argument of biological essentialists (inasmuch as a human-created intelligence is functionally similar to a biological human, and we can navigate the sex/gender distinction). However, if the AI forms no clearly-identifiable gender, it would greatly support the argument of social constructionists. In other words, running such an experiment would give us a greater understanding of nature vs nurture when it comes to gender identity: if we see traces of gender, the experiment supports nature; if we do not, it supports nurture.
There are also interesting thoughts to be had about the nature of gender and embodiment. Even before the successful creation of an AI we might well see a successful uploading of a human mind to a computer. What influence might the lack of embodiment and biological markers have on this newly accommodated and presumably continually learning instance of a mind?
You might be inclined to ask, What can we learn about human gender from a robot? But we’re not testing a robot, per se: we’re testing an intelligence over which we have some control of language and cultural conditioning (a control that is absent with humans). Try not to think of this as robots and humans, rather different intelligences. Furthermore, both intelligences are human, to a certain degree, inasmuch as one is regular human in a human body, and the other designed by humans.
Further still, if we are to believe the predictions of transhumanism, at some point such AIs will merge with other technologies and even us as people to trouble the very nature of human identity. The above discussion has been anchored in a contemporary understanding of “human,” “AI,” and “gender.” If our meaning of these words begins to evolve, or even collapse, that discussion becomes redundant and arcane as a wholly different variety of gendered and non-gendered identities are manifest. In short, we will have entered a phase of ultra-diversity, super-performance, and uber-queerness.