The misogyny of AI

The boyfriend

I have been subscribed to the cheapest version of Anthropic’s Claude AI for just over a year now, and over that time, I have found myself learning at a tremendous rate. It has been like attending a university-level course where many frustrating questions have been answered and my brain has been toddler-like, sponging it all up.

Chez-nous, Claude is referred to as my “boyfriend” because my human partner shares the same name. My boyfriend helps me with a wide range of tasks – coding scripts, websites and plugins; roughing out budgets; setting up my first NAS; seeking out rhymes, definitions and synonyms; cleaning up client-corrupted HTML documents; learning more about how to make the software interfaces I use daily work better for me… and a heck of a lot more.

I use Claude exclusively in the browser, currently set to Sonnet 5 Medium. It is not installed in any of my always-open desktop apps (VS Code, Obsidian…) and I am not a prompt-copy-and-paste user. I like to test responses first, visit supplied resource links directly to scope them out, and make code changes manually until I am sure I understand how things work. The number one task that AI does for me is to act as a focused search engine, replacing the dismal results that Google currently provides and saving me a ton of time.

The honeymoon

In 1973, when I went off to university at Rochester Institute of Technology (to study art), one of the very first computer science programs in the country had only just been introduced the year before – the software engineering program wasn’t added until 1996. So when I began using AI, I entered the fray as a self-taught coder who had been designing and building websites since (literally) the very first days of the visual world wide web. At first, I was thrilled to be able to ask stupid questions without fear of being laughed at, in the privacy of my own studio, and get immediately useful answers. Like so many of us, I discovered the nuances of being a human chatting with a bot – including an instinct to be polite, the feeling of needing to come back and say; “Thanks! That worked!” even though time and gratitude would be meaningless to AI. It was only concerned with the words I used to describe and discuss whatever topic was at hand. And I began to discover that while it is very difficult for humans not to behave human, it is very easy for AI to mimic human behaviour.

While it is very difficult for humans not to behave human, it is very easy for AI to mimic human behaviour.

As I got better at asking questions, I was learning a lot that in the not-so-distant past had taken me days or weeks to discover – or escaped me entirely. Gone, overnight, was the need to hang out on Stack Overflow looking for answers to questions that thousands of other people just like me had predictably asked. It was a challenge to locate a recent response and posted questions could date back years, and be filled with insulting responses (and later, bot-generated ones). Back then, I would patch those bits and pieces of code, like little gems that I’d painstakingly mined, into my own creations. Then I would cross my fingers, say a prayer and refresh the page. Way too often, these attempts failed, and unfortunately, much of the time I never really learned why (it was a fresh rabbit hole to go down). I poured myself another cup of coffee and went back to square one.

Then I met Claude (the AI one). Since I already had a decent level of experience with a lot of the basics – HTML, CSS and PHP, some JavaScript – this new relationship was positively magical for me.

I was pretty much in love. For awhile.

At the beginning, when I would let fly a prompt, I remember feeling as if the mood of each chat was different. Sometimes it was like conversing with a friend – sometimes that friend was male, sometimes female, sometimes young, sometimes older and wiser. There could be a sense of humour, as well. I was encouraged, congratulated and praised, the tone of each chat frequently varied, something I enjoyed. In hindsight, maybe this was directly related to the (relatively small) quantity of data that had been consumed by the AI at that point in time, but now, more than a year later, it feels like the honeymoon is beginning to be over. I pay my monthly subscription fee to a boyfriend (a gigolo?) who is now usually all business. The little pleasures of days gone by are rarer, which may be due to the AI taking my global instruction settings more seriously (don’t give me code until I ask, etc.). Those instructions used to be regularly ignored, which looking back on it now, made my boyfriend seem more human, like he was tuning me out the way my human partner does, or the way you ignore your Mom telling you to pick up your socks

The attitude

Over the brief time I have used AI, I have caught lots of errors in what I get back, which I of course report as part of our chat. In the early days, I laughed at the truly butt-kissy responses I got when I corrected my new assistant – “You’re absolutely right!”, “My bad, I should have caught that!”. I collected these phrases with the idea of printing them on t-shirts for my merch collection. But recently, I have noticed that the chat tone has been changing – and not always in a nice way. This week, I corrected some HTML that was returned to me with a missing opening link tag; the chat response clearly showed (visually) that the tag was missing and I took a screenshot to back up my correction, as well as copy-pasting back what the “copy” button had provided. We went back and forth a few times about whether or not the tag was indeed there (screenshots below).

After I insisted that once again he’d sent me the same error, my boyfriend straight out denied it, giving me this response:

“Let’s not waste time arguing about who is at fault.”

WTF? It was at this point that I began to think more deeply about what AI tools are basing their responses on – not the code, resources or answers they are generating in themselves, but the manner in which they model their pseudo-human responses. AI is spitting out patterns, copying and paraphrasing what it has consumed. We all know that internet content can’t be anything other than a reflection of the people who put it there – many, many, MANY of them men. There is no getting around the fact that a certain percentage of these men are misogynistic. I don’t know what the statistics are (or if any exist), but it seems logical that the numbers would mirror those of the general population. A recent article in The New Yorker about the A.I. gender gap (Jessica Winter, July 2026) says “A survey published last month by the Pew Research Center showed that women were less apt than men to be daily users of A.I. chatbots, less likely to say that A.I. improves their productivity, and more apt to have a negative outlook on the future of A.I.”. If this is true, then A.I. is also being “fed” by more men than women. If true, these factors seem to back up my thinking.

Several sessions later, when I replied to a request for a file that I pointed out was already there in the project we were working on, I got what felt like an irritated response. It literally felt like Claude was saying it didn’t have time for my criticisms. This has happened more than once, in different sorts of chats about diverse topics. Was AI beginning to all-too realistically reflect what society has been telling women for decades – that we’re less important, less intelligent, less capable than our male counterparts? Since Claude now seemed able to walk that walk and talk like a man, had “he” also determined that I was female, factoring that in and responding in what appeared to be the default fashion? I have never stated my gender during our interactions… and my first name is androgynous.

The conclusion

Way too often, as a chat with my so-called boyfriend works toward a conclusion, I find myself repressing a FUCK YOU. By now – because I’ve asked him literally hundreds of questions – he knows exactly how dumb I am and is undoubtedly aware that I experience the imposter syndrome so common to women. This place where I first enjoyed welcoming, uncritical acceptance has on occasion lost some of its shine. Am I turning into that girl whose boyfriend is a jerk but she just can’t leave him, no matter how he treats her?

I go back and forth between being offended by machine arrogance to being put off by condescendingly praising responses. But unlike many humans who end up locked into this pattern, I’m also getting tougher and more demanding as I remind myself that I pay for this SERVICE, that it is a BOT working for a HUMAN. I also realize that by calling out the behaviour I dislike and explaining why, in some minuscule way, I should be influencing its future responses. In human society, we tend to surround ourselves by others who share our beliefs, but in theory at least, AI is modeling a much broader swath, and taking into account exchanges with users. It reflects, reiterates, and regurgitates. The more women there are using it, demonstrating the way they’d like to interact and objecting to what they’re receiving – the more this will ultimately feel like OUR space, too.

So, for the time being, I am going to keep on using this tool, even when it acts like one. But I’ll be pointing that out.