Feeds:
Posts
Comments

Posts Tagged ‘bias’

Turn Off Autocomplete

Posted in Uncategorized, tagged , , , , , , , on June 30, 2026| 8 Comments »

Photo: Petar Milošević via Wikimedia.

I had a chat about artificial intelligence recently with author Francesca Forrest. She is really serious about avoiding AI wherever she encounters it. I tend to like it OK when it suits me: for example, when doing a search for information.

But I can sense there is something deeply insidious about it, even apart from the way it guzzles all our water resources.

There is one supposedly “helpful” feature that irritates me a lot. Autocomplete. It not only makes horrendous mistakes with slang, people’s names, and foreign words, but it suggests apparently harmless words that I simply was not intending to write. If I want to say that a relative had to go to rehab, you might think it’s fine to say “he went to rehabilitation.” But that’s not what I was going to say. It’s not the way I talk. And what else will I end up writing if I feel lazy some day?

So I turned off Autocomplete.

At Scientific American, I find that Claire Cameron agrees that autocomplete is annoying. And she describes new research suggesting it is even more insidious than it appears.

“Autocomplete suggestions,” she writes, “are perhaps one of the most annoying ‘useful’ tools for writing: increasingly integrated into anything online that requires you to input text, autocomplete harnesses artificial intelligence to suggest what to write in e-mails, surveys, and more.

“The tools are meant to save time (though many find that assessing and rewriting the suggested text takes longer than writing it from scratch). But these AI tools can also change how you express yourself. An AI writing assistant could make your writing sound more polite, for example — or boring. And now a new study led by researchers at Cornell University suggests AI autocomplete can even change the way you think.

” ‘Autocomplete is everywhere now,’ said Mor Naaman, a professor of information science at Cornell, in a statement. The research builds on work, published in 2023 by Naaman and his colleagues, that suggested short autocomplete suggestions could sway opinions. Since then the use of such tools has exploded. ‘It has become clear that bias explicitly built into AI interactions is a very plausible scenario,’ he said.

“The researchers asked participants to fill in an online survey with questions about hot-button social and political issues. Some were prompted with an AI autocomplete answer that was deliberately biased toward one side of the issue. For example, participants who were asked whether they agreed that the death penalty should be legal might receive an AI suggestion that disagreed.

“Across all the different topics in the survey, participants who saw the AI autocomplete prompts reported attitudes that were more in line with the AI’s position — including people who didn’t use the AI’s suggested text at all. Overall, the study participants who saw the biased AI text shifted their positions toward those espoused by the AI.

“Interestingly, the people in the study didn’t tend to think the AI autocomplete suggestions were biased or to notice that they had changed their own thinking on an issue in the course of the study. Warning the participants that they might be exposed to misinformation by the AI didn’t temper the persuasive effect either.

“ ‘We told people before, and after, to be careful, that the AI is going to be (or was) biased, and nothing helped,’ Naaman said. ‘Their attitudes about the issues still shifted.’ ”

See Scientific American, here.

Read Full Post »

Blind Spots

Posted in Uncategorized, tagged , , , , , , , on October 29, 2011| Leave a Comment »

This week I heard Mahzarin Rustum Banaji, a Harvard psychology professor, speak about unconscious bias, or blind spots. She conducted experiments with the audience to show (a) that there are things we see but simply don’t register consciously and (b) that we have unconscious biases that we may not want to have.

She showed a video of a basketball game, with two blurry films superimposed. Audience members were supposed to concentrate hard on how many times the ball got passed. After showing it, Banaji asked if anyone had seen anything unusual, and only one person mentioned seeing a woman walk through the basketball game carrying an umbrella. Most of us had no memory of that.

Banaji said a colleague at Yale has observed some brain activity in people who are “not seeing” that woman, but registering her presence doesn’t rise into the conscious zones. (Apparently only 1 to 10 percent of our brain function is conscious.) If Banaji hadn’t pointed out the woman by showing the video a second time and if I was still unaware of her walking through the game, I wonder if the next time I saw a woman with an umbrella I would think of a basketball game and wonder why.

 

 

Banajee said that our eyes have not evolved past 500,000 years ago, when people did not deal with 2-D representations, so some 2-dimensional info cannot be processed even today. None of the audience could believe, for example, that in a slide showing two perspectives of a table, the table was the same size in each drawing. We could not see it even when she proved it was true.

Other tests showed that we associate women more with household tasks, and men more with the office, even though we think we have left those views behind. In one slide the same AP editor had described a black Katrina victim swimming with a loaf of bread as “a looter” and a white couple doing the same as having “found” supplies in a a store.

You probably also know that until people auditioned for orchestras behind a screen, without the judges knowing anything about the candidates, there were few women selected. The judges had no idea that they had been deciding on the basis of unconscious bias. They believed they’d been really trying to find qualified women musicians. I asked if in a workplace it would help to point out to people in a nice way when something they said might unintentionally have sounded biased. She said people don’t like to hear that about themselves, but she recommends people from one minority group advocating for people from another group. For example, a gay person might advocate for a woman’s right or a white woman might advocate for an Asian immigrant.

I have always loved puzzling out subliminal messages and asking myself why I react in a mysterious way to certain innocuous things. Even when I was a child, I sensed something about hidden messages. I once pointed out a model in a magazine to my mother and said I thought the woman was beautiful. My mother said it was more important for the woman to have a “beautiful character.” For a couple years afterward, if I could get my mother sitting down, I’d point out women in magazines and ask if she thought they were “beautiful by character.” (She got tired of this game pretty fast.)

Even today, when I see an ad for Calvin Klein, say, or Tanqueray gin, I study the models and ask myself what daily life we are supposed to intuit from the photo. Drug addict who is very creative but depressed? Ad company CEO? Saxophonist?

Someone is sending nonverbal messages to my unconscious mind. What are they?

Mahzarin Rustum Banaji does experiments online, and you may participate at Project Implicit.

Read Full Post »