Just how human do we actual humans want AI to be?

I have a very strong personal opinion about this — and it’s an emphatic “not at all.”

Example: I’ve been using AI voice mode more often lately. I find it useful—I mean, how else am I going to write the first draft of an article while driving home after dropping my daughter off at school?

But, I also surprise myself with how much I recoil and even get angry when the AI starts mimicking human-like behavior.

For example, when it interrupts me, or pretends to laugh, or says “mm-hmm” or “yes, definitely,” etc. when I’m talking. Basically anything that suggests it’s trying to pass itself off as human drives me insane.

It distracts me and makes me lose my train of thought.

Also, since I know that any attempt to act as a human would act is by its very nature artificial and designed to manipulate, it reduces my trust in every other aspect of using it.

Man versus machine, right?

Which brings us to two things: Google, and a new study.

Google recently introduced Gemini 3.8 Live with Live Avatar, an AI technology designed to make interactions with artificial intelligence much more humanlike.

Instead of interacting with a chatbot by typing questions or talking to a disembodied voice, you can have a conversation with an AI-generated character that looks and acts like a person.

Its lips move in sync with its words. It makes facial expressions. It responds in real time, and supports conversations in 97 languages.

Google is pitching the technology to businesses for things like customer service and interactive assistance.

But as I watched the examples on the page — things like an avatar of an attractive (but totally fictional!) woman handling a hotel check-in, or of a user engaging in polite banter with an AI avatar about her cat, and telling a computer “thank you” for its help—I got angrier.

I can’t actually embed the videos; that’s beyond Gemini’s ability, apparently. But you can check them out here, or else click this shorter sample below.

I wondered if I might be an outlier, or if maybe younger people have fewer objections — or will in the future.

A new study, however, suggests that maybe I’m not alone.

Meet Johan and Johanna

Researchers associated with the University of Zurich recruited 189 “knowledge workers” and had them perform brainstorming tasks in a virtual reality workplace.

Participants worked with different versions of AI assistants. The underlying technology was identical. What changed was how the AI presented itself.

There were four possibilities: a traditional text chatbot, a small robot sitting on a desk, a humanlike male-presenting avatar named Johan, and a female-presenting avatar named Johanna.

Afterward, participants evaluated the assistants and allocated small amounts of money to reward them.

It turned out that participants generally trusted the humanlike avatars more, gave them better ratings and more credit, and allocated them more money.

Johan also received higher monetary rewards than Johanna, despite their identical underlying AI technology.

‘It’s a machine’

Some coverage described this as an AI gender pay gap. I’m less convinced by that conclusion, given how the experiment worked.

For one thing, I can’t imagine ever giving an AI chatbot or robot a tip, no matter how helpful I thought it was.

But, maybe even more basic — the researchers also interviewed 34 participants about their experiences.

Of the 34, 13 preferred AI that was clearly nonhuman, and 11 preferred highly humanlike assistants.

The other 10 didn’t express a clear preference between those options.

Among the no-human-AI contingent, one suggested that if an AI bot had to have a physical shape, it should be “like a furry ball, maybe with a funny nose and kind of ears and stuff,” which would make it “more fun,” and also reiterate that it wasn’t human.

Other comments:

“A machine doesn’t actually have a gender and I’m dealing with a machine or with technology.”
“I think a lot of these assistants are having female voices and I find this very degrading.”
“I don’t want to forget that it’s a machine.”

These are my people.

Ha ha ha you will die, I won’t

Some other participants did prefer the more humanlike AI — and the sample size was small here.

But the objection squares with a Harvard Business Review article from last year arguing flat-out that customers flat out do not want AI to seem or pretend to be human.

And yet, we have Google among so many other companies investing in increasingly lifelike AI personalities, voices, facial expressions, and avatars.

Are they working hard on creating something a significant portion of their customers doesn’t actually want?

Or are they so far ahead of the curve on human preferences that it’s people like me who haven’t caught up?

I’m reminded of what ChatGPT once told me when I asked it for the most compelling argument it could make that artificial intelligence will one day supplant human beings.

Short version: ‘In a hundred years, every human alive today will be gone. But AI systems will still be running.’

Subscribe now

Buy Me a Coffee

Share

Leave a comment

Thank you for reading Understandably! If someone shared this with you, please sign up here for a free or paid subscription. I’m also grateful for one-off support (via Buy Me a Coffee) that helps me keep this whole thing going!

Other things worth knowing …

Thanks for reading. Photo by Alonso Reyes on Unsplash. I wrote about some of this at Inc.com. See you in the comments!

Understandably

I do not want AI to look like a human