Tim's journal

Language affects thought, and part of why science isn't objective is that communicating scientific knowledge relies on language, which is always imprecise and governed by politics and culture.

In "The Egg and the Sperm", Emily Martin wrote about how the language used to describe human reproduction distorted the truth. Scientists, mostly cis men, were biased towards seeing sperm as active penetrators of the passive egg. In fact, as Martin detailed, eggs do a lot of active work to reject weak sperm and entice strong sperm. (Of course, even the metaphor of "weak" or "strong" sperm reflects socially mediated beliefs.)

Another example from reproduction is the misunderstanding of the biological function of menstruation that also arose from sociopolitical biases about gender. In a 2012 journal article, Emera, Romero and Wagner posited that the function of menstruation has been misunderstood due to sexist beliefs that bodies coded as female are intrinsically nurturing: the endometrial lining was previously construed as the uterus creating a nurturing environment for a potential embryo, where in fact, it might be more accurate to view it as a hostile environment that only the strongest embryos can survive (there's that "strong/weak" political language again.) I'm not qualified to assess on the accuracy of Emera et al.'s idea, but I am qualified to observe that assessing its validity has been so far hindered by the misapplication of gender stereotypes to biology.

Yet another example is that of same-sex sexual behavior in non-human animals; Bruce Bagemihl's book Biological Exuberance details the history of (again, mostly heterosexual cis male) scientists getting itgrievously wrong about the nature and function of sexual behavior. It would be funny if it wasn't so harmful. Just one example is the publication of a paper, in 1981, entitled "Abnormal Sexual Behavior of Confined Female Hemichienus auritus syriacus [Long-eared Hedgehogs]". It's not objective, rational, or scientific to label hedgehog sex as "abnormal" -- rather, it reflects social and political biases. And in that case (and many similar cases), politics kept scientists from understanding animal behavior.

In all of these cases, bad metaphors kept us from seeing the truth. We used these metaphors not because they helped us understand reality, but because they were lazily borrowed from the society as it was at the time and its prejudices. This is why scientific research can never be fully understood outside the context of the people who produced it and the culture they lived in.

Master/Slave: a Case Study

In computer science and electrical engineering, the term "master/slave" has been used in a variety of loosely related ways. A representative example is that of distributed databases: if you want to implement a database system that can scale up to handling a lot of queries, it might occur to you to put many servers around the world that have copies of the same data, instead of relying on just one server (which could fail, or could become slow if a lot of people start querying it all at once) in one physical location. But then how do you make sure that the data on all of the servers are consistent? Imagine two different whiteboards, one in the computer science building at Berkeley and one in the computer science building at MIT: there's no reason to assume that whatever is written on the two whiteboards is going to be the same unless people adopt a mechanism for communicating with each other so that one whiteboard gets updated every time the other does. In the context of databases, one mechanism for consistency is the "master/slave" paradigm: one copy of the database gets designated as the authoritative one, and all the other copies -- "slaves" -- continuously ask the master for updates that they apply to themselves (or alternately, the master publishes changes to the slaves -- that's an implementation detail).

A lot of the historical background behind the use of "master/slave" in a technical context already got covered by Ron Eglash in his 2007 article "Broken Metaphor: The Master-Slave Analogy in Technical Literature". Unfortunately, you won't be able to read the article (easily) unless you have access to JSTOR. Eglash examined early uses of "master/slave" terminology carefully and pointed out that "master/slave" entered common use in engineering long after the abolition of slavery in the US. Thus it can't be defended as "a product of its time." He also points out that "master/slave" is also an inaccurate metaphor in many of the technical contexts where it's used: for example, for a system with multiple hard drives where the "master" and "slave" drives merely occupy different places in the boot sequence, rather than having a control or power relationship.

But I think the most interesting point Eglash makes is about the difference between power as embodied in mechanical systems versus electrical systems:

A second issue, closely related, is the difference that electrical signals make. Consider what it meant to drive a car before power steering. You wrestled with the wheel; the vehicle did not slavishly carry out your whims, and steering was more like a negotiation between manager and employee. Hence the appropriateness of terms such as "servo-motor" (coined in 1872) and "servomechanism" (1930s): both suggest "servant," someone subordinate but also in some sense autonomous. These precybernetic systems, often mechanically linked, did not highlight the division of control and power. But electrical systems did. Engineers found that by using an electromagnetic relay or vacuum tube, a powerful mechanical apparatus could be slaved to a tiny electronic signal. Here we have a much sharper disjunction between the informational and material domains. And with the introduction of the transistor in the 1950s and the integrated circuit in the 1960s, the split became even more stark.

This coupling of immense material power with a relatively feeble informational signal became a fundamental aspect of control mechanisms and automation at all scales...

In light of Eglash's observation, it's worth looking harder at why some engineers are so attached to the "master/slave" terminology, aside from fear of change. The "immense material power" of an electronic signal can't be observed directly. Do engineers in a white-male-dominated field like talking about their systems in terms of masters and slaves because they need to feel like they're somebody's master? Does it make them feel powerful? Given that engineering has become increasingly hostile to people who aren't white and male as it has become more dependent on leveraging smaller and smaller amounts of (physical) power to do more and more, I think it's worth asking what work metaphors like "master/slave" do to make white male engineers feel like they're doing a man's job.

Bad Metaphors

"Master/slave" both serves a psychological function and reflects authoritarian politics, even if the person using that term is not an authoritarian. No one needs to consciously be an authoritarian, though, for authoritarianism to distort our thinking. Language derived from societies organized around a few people controlling many others will affect how systems get designed.

A master/slave system has a single point of failure: what if the master fails? Then there's no longer any mechanism for the slaves to keep each other consistent. There are better solutions, which constitute an open research topic in distributed systems -- discussing them is beyond the scope of this blog post, but I just want to point out that the authoritarian imagination behind both societies organized around slavery (we still live in one of those societies, by the way, given the degree to which the economy depends on the prison industry and on labor performed by prisoners) impoverishes our thinking about systems design. It turns out that single points of failure are bad news for both computer systems, and societies.

I conjecture that the master-slave metaphor encourages us to design systems that have single points of failure, and that the metaphor is so compelling because of its relationship with the continued legacy of slavery. I don't claim to be certain. People who design decentralized, peer-to-peer systems may not be any more likely to have egalitarian politics, for all I know. So I'm asking a question, rather than answering one: do fascists, or people who haven't examined their latent fascism, build fragile systems?

Names are important. Lazy evaluation, for example, wasn't too popular when it was only known by the name of "cons should not allocate." So master/slave is worth abandoning not just because the words "master" and "slave" evoke trauma for Black Americans, but also because flawed thinking about societies and flawed thinking about technology are mutually self-reinforcing.

Good metaphors have the power to help us think better, just as bad ones can limit our imagination. Let's be aware of what shapes our imagination. It's not "only words" -- it's all words, and people who write software should understand that as well as anyone. Metaphors are powerful. Let's try to be aware of how they affect us, and not suppose that the power relationship between people and words only goes one way.

---

Do you like this post? Support me on Patreon and help me write more like it.