Twitter Thinks My Gender Is [EICAR String] (Or “Server-Side Checks Matter”)
soatok.blog
soatok.blog
I suspect this plan might even work. I happen to maintain the libmspack library[2] in RHEL, it's a library for extracting Microsoft CABinet files. Part of the source contains regression tests. Some of those tests are flagged by AV software because they contain past exploits for the library: the idea of these regression tests is to ensure that they don't reintroduce the vulnerability by accident.
Well, the libmspack source RPM causes all kinds of problems, both internally at Red Hat and at customer sites. It is constantly being flagged by AV software. Every release I have to manually waive the package through our AV checks, and we've had feedback from customers that we're "shipping a virus" which had to be handled very diplomatically.
AV software is crazy, but let's make it work for us :-)
Nobody detects EICAR when it’s surrounded by other data.
Strings like
<script type="text/javascript" src="http://web.nba1001.net:8888/tj/tongji.js"></script>
But unless they've tested the backend against a 64 gigabyte gender...
Stereotypes exist because they model people pretty well.
If you enter that your gender is unicorn you probably just get slightly more generic ads.
Here’s a scary thought: if you walk the street everyone who sees you will assume your gender.
It's also just really bizarre and rude. Imagine referring to a cis woman with a beard as an 'it', or 'he'. For whatever reason, people go out of their way to commit this kind of absurd faux-pas with trans people.
I do that sometimes, unintentionally (regardless of the person in question having a beard or not). It's just that gender in language is most of the time redundant and it only introduces noise into conversation. For most topics, the gender of the persons involved is irrelevant. But many languages force you to choose one, which just creates confusion if this piece of information is crucial or if it's just this bloat that the language forced you to include in your sentence.
Oh, and genders are much more prevalent in my native language. English is actually better than most languages in this respect. But let's go a step further and just get rid of it. I've heard the Swedes chose to move to "person/thing" pronouns, as opposed to "male/female/thing".
In my native language, verbs change their forms based on subject's gender. And then when you have sentences like "if I were in your shoes, I would verb", my brain tells me to assume the gender of the other speaker when I say verb. I mean, it's only logical: if I were in your shoes, I would be of your gender. But the language works otherwise. It's just stupid.
As to people's reactions to my occasional misnamings, most people don't care at all, some are amused. Because, really, this part of language is ridiculous. And making an issue out of it is immature.
While most people will eventually be recognized - at least if they fit within the binary somewhere and medically transition - a few will not, and almost nobody will on the long way there.
To insist it is not a big deal because you can't relate, that is immature.
It's an interesting tangent here: English has been very slowly converging on singular `they` since Shakespearean Middle English. It's interesting that singular `you` (which started after singular `they`) beat singular `they` to the modern English party, but it somewhat makes sense that removing language formality distinctions was quickly more useful in short term language evolution than removing English's few remaining gender-required assumptions.
That isn't true for any stringent definition of well. Plus you don't have to be a "did you assume my gender" person to be peeved that someone's poorly trained ML system is making false assumptions about you and then labeling you with those assumptions. Even if it were an "are you an innie or outie" field it would feel intrusive to have an automated system guess and label you off of "stereotypes".
That IS the internet these days...
While some ads could be targeted at anybody, I've recently seen ads that presume I'm
- Jewish - Muslim - a woman - a millionaire - Black - have HIV - am schizophrenic - am diabetic - am bipolar - am a musician
...none of which are things I have reported about myself or verifiable facts.
I am a lot more annoyed by people who say "stereotypes exist because they model people pretty well" than by the AI systems going haywire.
Genetic testing does a great job determining sex objectively.
What if what we "feel" is wrong and our fabrication of identity has led us to a point where a machine can determine our "truest" gender based on our behaviours much better than an individual can self-ascribe one? Rendering it (to borrow from econ) a preference that is expressed and not always revealed, and nullifying any socially inferred stereotyped.
At best (for those that care) this at least can give us a gender spectrum, and a means of objectively assessing someone, which has upsides and downsides of course.
This doesn’t even take into account social sex determination, or gender identification, which are psychological (and physiological!) in scope & involve a much more complicated biological system.
Most vendors now simply a guess 128 floats about you as a vector which encodes your gender, your love of stars, your favourite foods, and how much you're itching to buy a deck chair...
The vendors themselves don't really know what's in the vector, apart from the fact it helps them predict ads. They couldn't explain it if they tried - nobody could - it's all generated by stochastic ML methods.
This isn't true. These systems are not unknowable. It's just generally not worthwhile to figure out what these encode, but it is doable.