Gendered AI, initial results

Men are machines. Women are bodies.
Male is extreme. Women are nuance.
General AI has gender. Other AI does not.
Male is free-will. Machine is subservience.
Male is default. Women when it’s necessary.

At least in screen sci-fi.

Let me explain.

In November of 2018, a tweet thread between Chris Geison and Kathy Baxter called my attention to questions about the gender of AI in sci-fi. Baxter noted that most AI is male, and how female AI is often quite subservient or sexualized. In this thread, Gieson added Cathy Pearl’s observation that embodied AI is often female and male is more often disembodied and regarded as a peer.

I already had a “database” (read: Google Sheet) of AI in screen sci-fi from Untold AI, my 2018 study of the stories screen sci-fi doesn’t tell, but should. So, I thought I could provide some formal analysis to this Gendered AI discussion. To that end I’ve added around 325 AI characters to the Google Sheet, and run some analyses. This series of posts will break it all down for you.

Image result for r2d2 hologram
Oh, we’ll come back to this little “guy.”

Now, it can get a little dry to talk about percentages and comparisons and distributions, so I’m going to do my best to keep tying things back to the shows and the characters and the upshot of all this analysis. But the way we get to that upshot is through the numbers, so stick with me. For this first post, I’m going to share what I captured, and what counts as an AI character for purposes of this study.

The following is true in the survey as of 08 APR 2019. The live data, available in Google Sheets, may be updated from this.

The data set

  • 327 AI characters from science fiction (see the full list in the live sheet)
  • Movies and television shows from 1927 (Metropolis)–2018 (Upgrade)

Call to action: Of course I missed some movies and TV shows. Add them in the comments, including a link to their IMDB page.

The survey that drives this site has always focused on screen sci-fi for its ability to depict interfaces that can be reviewed. Literature is much more free to experiment with ideas than screen sci-fi, and so will have lots of additional examples, but won’t appear in the survey.

Each character is tagged multiple ways. More detail on particular attributes below.

  • Movie or Show Title and Episode if appropriate
  • Year
  • Name
  • Embodiment
  • Physicality/Virtuality
  • Gender Presentation (which is a roll-up of four separately tracked variables)
    • Appearance or evidence of primary sex characteristics
    • Appearance of secondary sex characteristics
    • Voice
    • Pronouns used by other characters
  • Subservience to humans
  • Germane-ness of gender (more on this in its own section)
  • Goodness
  • If not free-willed, the gender of the master
  • Category of AI (Narrow, General, or Super)
  • Whether their gender presentation changes over time
  • Genesis, or how the AI came to be. This is mostly used to distinguish AI that are copies of humans (whose gender would thereby be inherited).

Call to action: If you think there’s some critical attribute that I’m missing, pipe up in the comments. I can’t promise I can get to it before the next post, but I can consider it as a future enhancement.

Image result for skynet helena

Yes, but which Skynet?

With the exception of the flag marking changed genders, when characters change other attributes over the course of their stories, they are tagged for their final state. For example, the Maschinenmensch from Metropolis begins an anthropomorphic robot, but after Rotwang transfers Maria’s likeness to it, it becomes indistinguishable from human, and so is tagged as such.

If you’re looking at the Sheets data, you’ll see that text values have corresponding numerical columns to allow for easy sorting and graphing data, but I tried to gray them so they don’t distract from a reading of the raw data.

Image result for rotwang

Full disclosure: Possible problems with this data

  • Sci-fi is a vast supergenre. There are certainly examples missing from the survey, so it should not be regarded as exhaustive. (I tried to get as many as I could.)
    • I generally target well-known examples rather than limited-release or student projects.
    • The sci-fi interfaces blog usually eschews comedy that breaks the 4th wall routinely, (e.g. Spaceballs), as this makes for very complicated analysis, and so the survey will be missing these examples as well.
    • I only speak English fluently, and so have only reviewed shows in English, with English dubbing, or with English subtitling.
  • I am not a data scientist. I’m a smart guy who tried his best, but may have made some errors in the formulas.
  • I am not an expert in gender issues. I may make unintentional errors in discussing or categorizing genders, use insensitive language, or have naive errors in my thinking. I have engaged a professional sensitivity review, but of course they might not catch everything, either.
  • I am a progressive, liberal, (imperfect, see above) feminist. Though I tried not to, my bias may have colored how I coded the examples and of course the interpretation of this data.
  • I have to go on a LOT of common-case presumptions. For example, men can have breasts for many reasons, but I used the presence of breasts as one marker of female-ness. I suspect this is a disservice to the real complexity of gender and sex in the world, but presuming the audience sees gender as primarily binary, it marks how these characters are likely perceived rather than what they are.

I’m not too worried about these caveats, though, since what we’re aiming for here isn’t precision engineering specs, but rather to get a numbers-based sense of the big patterns in screen sci-fi, and for that, a little bit of noise in the numbers is OK.

Lastly, not every character that you think might qualify does, so I should explain my rationale for what got in and what got left out.

What counts as an AI character?

I’ve tried to be strict about what counts as AI in that the intelligence of the character must be housed in non-biological circuitry. This leaves out some characters that on a cursory consideration would seem like a natural fit. For an example, compare The Stepford Wives (1975) and The Stepford Wives (2004). The wives in the original were robots through and through—mechanical, lookalike replacements of the original humans. But the wives in the remake were cyborgs, with robotic bodies housing their original, human brains. This means that in the original, the wives count as AI and appear in the survey. But because of this cyborg technicality, none of the “robotic” characters from the remake make it in. Not even the little cyborg dog.

Meanwhile, Rachel and Deckard, replicants from the Blade Runner universe, had a baby (according to Blade Runner 2049) so we can generalize and say replicants are capable of wholly biological reproductive acts. Given this you might think they’re out of the survey, but, since they are fabricated, they get into the survey.

Also, T-800s Terminators (the Arnold kind) get in, because even with their wetware bodies, the intelligence they carry is non-biological.

I know, it’s complex and sometimes counter-intuitive. Such is data.

OK, so looking at those attributes for those characters, the first thing we should look at is the distributions. This included all sorts of questions like: How many AI present as men? How many as women? How many are nonbinary? What kinds of bodies do they have? Who is master of whom?

It’s thrilling, thrilling data analysis action, so stay tuned.

Gendered AI: Gender and Goodness

The Gendered AI series looks at sci-fi movies and television to see how Hollywood treats AI of different gender presentations. For example, do female-presenting AIs get different bodies than male-presenting AIs? (Yes.) Are female AIs more subservient? (No.) What genders are the masters of AI? This particular post is about gender and goodness. If you haven’t read the series intro, related goodness distributions, or correlations 101 posts, I recommend you read them first. As always, check out the live Google sheet for the most recent data.

n.b. If you’re looking at the live sheet, you may note it says “alignment” rather than “goodness” in the dropdown and sheets. Sorry about the D&D roots showing. But by this, I mean a rough, highly debatable scale of saintliness to villainy.

Gender and goodness

What do we see when we look at the correlations of gender and level of goodness? There are three big trends.

  1. The aggregate picture shows a tendency for female-presenting AI’s to be closer to neutral, rather than extreme.
  2. It shows a tendency for male-presenting AI’s to be very good, or very evil.
  3. It shows a slight tendency for nonbinary-presenting AI to be slightly evil, but not full-bore.

When we look into the detailed chart, some additional trends appear.

  • Biologicially- and bodily-presenting female AI tends toward somewhat evil, but not very evil.
  • Socially female (voice or pronouns, only) tend toward neutral.
  • Gender-less AI spike at somewhat evil.
  • Genderfluid characters (noting that this occurs mostly as a tool of deception) spike at very evil, like, say, Skynet.
  • AIs showing multiple genders tend toward neutral, like Star Trek TOS’s Exo III androids, or somewhat evil, like Mudd’s androids.