Apple, AI, and the Illusion of Consent

Feb 18

Written By Michele Edington (formerly Michele Gargiulo)

A lot of us out there used to have something like a poker face. I used to pride myself on the ability not to show what I was feeling or thinking directly on my face (especially if a manager or owner asked me a particularly silly question). Back in my day if you didn’t say something, it stayed with you. A thought could hide on your face and vanish or a word could stop halfway to becoming sound and no one else would ever know it was there.

That boundary is oddly starting to vanish now.

In late January, posts began circulating online about Apple acquiring an Israeli AI startup tied to research into “silent speech”, which is basically some fancy technology created to assign meaning when words aren’t even spoken out loud. The idea alone was enough to stop people mid-scroll as the absurd click-bait took over claiming facial movements were training machines to understand what our faces are saying without our words.

Now of course, this kind of research isn’t new and systems that study facial micro-movements, muscle activity, and nonverbal signals have been quietly developing for years, most likely being pushed around the agenda of helping accessibility or communication in difficult to speak in environments.

The thing is though, when technology moves from responding to what we say to interpreting what our bodies do, even our silence becomes another signal.

What’s actually known now and what isn’t

Apple has a long history of acquiring Israeli technology firms, particularly those working in sensing, audio, and machine learning. Its 2013 acquisition of PrimeSense, for example, laid important groundwork for depth sensing and facial recognition technologies that later appeared in its devices all around the world.

Separately, and totally independently of Apple, researchers across academia and industry have been developing silent speech interfaces for more than a decade. These systems were trying to decode speech or intent from non-audible signals such as lip and jaw movement. They claimed that there were facial muscle micro-activations or even electromyography (EMG) signals from the throat or face and subtle physiological changes associated with speech formation that it could read and interpret. On paper, it’s research, but in practice, it can feel uncomfortably close to mind-reading.

Of course this work has been pushed and explained away around accessibility and helping people who cannot speak, or enabling communication in environments where sound is impractical or unsafe. That framing isn’t a lie, but it’s also not the whole story.

For years, people joked, and worried, that their phones were listening to them. Ads seemed to appear a little too conveniently after certain conversations. The explanation was always the same: coincidence, algorithms, metadata…there’s nothing to see here.

Of course, it’s not publicly confirmed that a consumer-ready Apple system that “reads your face” to extract thoughts is actually out there, but we’re all out here just interpreting whatever we can from what the larger companies aren’t saying. Technology doesn’t need to be finished to change the world, it only needs to be normalized.

If you step back, the progression is almost elegant in design. We started by telling machines exactly what we wanted and typed and clicked then issued commands. Then we began speaking to them with our “Hey Siri.” “Alexa.” You know the drill. Voice is messy though and public in the way that anyone who’s listening knows what you’re saying.

It goes to figure that the industry kept reaching for something it could sell next. Quieter in this case doesn’t truly mean silent, it just means inferred. When systems stop waiting for you to tell them something and instead begin to detect it, the entire idea of consent changes in my opinion.

You didn’t press send or even speak to ask the question, but the system somehow learned anyway.

The privacy paradox

Apple, more than any other major technology company, has positioned itself as privacy-forward. On-device processing, reduced data sharing, and marketing campaigns that emphasize protection has been the norm for years.

That reputation matters. Especially today, when it feels like everything we’re being spoon-fed on the interwebs are all cherry-picked for us. What these companies forget in their race to chase the money is that privacy is also about what is being interpreted in the first place.

A microphone records sound while a camera captures light, but an inference system assigns meaning. Meaning is where things get complicated, especially when most of us don’t even know what a lot of our gut-instincts mean when they flare to life.

If a device can learn to associate facial tension, lip movement, or micro-expressions with some sort of intent, the boundary between “what I chose to share” and “what the system decided to read” starts to blur. Even when everything is “encrypted”, or when nothing leaves the device. This sort of shift isn’t simply technical, it’s psychological.

Accessibility always opens the door first

I’ve mentioned this time and time again in previous posts of mine (just read about how I think AI would take over when you get a chance), there’s a pattern here, and it’s worth naming without calling me a tin-foil-hatter.

Many of the most invasive technologies ever created begin as accessibility tools. Speech-to-text, facial recognition, eye tracking, motion detection, you name it. Each one entered the world promising inclusion and support. Sure, they did those things, but once the capability existed, it didn’t stay contained.

It spread into convenience, then into optimization, before landing firmly into expectation. Silent speech technology follows the same arc. Helping someone communicate without sound is a gift. The danger arrives later, when the same mechanisms are used to reduce friction for everyone else.

“Frictionless” systems don’t wait for clarity, they guess.

For most of history, some part of us remained opaque. Our poker face would be able to hide our thoughts without them being detected. Our grandparents had the luxury of reacting without being analyzed and the ability to withhold without explanation.

Unreadability was a form of freedom that vanished in the age of convenience. Modern computing trends in the opposite direction by rewarding predictability. Of course machines prefers patterns over mystery, it works best when we’re legible.

Even if you trust the company building the system (which you shouldn’t), the precedent doesn’t belong to that company alone. Once a category of technology exists, it becomes part of the cultural vocabulary by setting expectations and reshaping norms. The thing isn’t “will Apple do this responsibly?”, it’s “what kind of world does this make acceptable?”

Here comes the future

The future doesn’t have a dystopian announcement coming with it, even though it probably should. No one is going to tell you that your thoughts are no longer private. Well, besides me in this little corner of the interwebs.

It’s just going to arrive as a feature update or a way to make devices feel more natural. One day though, silence won’t mean what it used to because something is interpreting your silence and assigning it meaning.

Interpretation is dangerous for more than one reason, starting with the fact that it’s often wrong.

Apple, AI, and the Illusion of Consent

What’s actually known now and what isn’t

The privacy paradox

Accessibility always opens the door first

Here comes the future

Other Reads You Might Enjoy:

The Uncensored Library: Where Journalism Went When the Internet Closed Its Doors