Voice will be the new keyboard (and that is not good news for everyone)

Until recently, talking to a voice assistant was a small humiliation: broken sentences, “sorry, I didn’t get that”, waiting. Today’s models are something else. Full-duplex ones, like GPT-Live, which now also powers Lumen’s assistant, listen and speak at the same time: you can interrupt them, they let you finish, they understand an “mmm” or a hesitation.

It is a change of interface, not just of technology. And interfaces change who uses a tool.

Who gains

People who don’t like typing. Many can explain a problem perfectly out loud and freeze in front of a text box. With voice, the “prompt” becomes a conversation.
People with busy hands. Driving, on a building site, cooking, walking between clients.
Older people and people with disabilities. For those who struggle with screens and keyboards, a spoken interface can be the missing door.
Learners. Repeating a topic out loud with a patient tutor, or practising a foreign language, works better than reading.

How usage changes

We ask differently out loud: more context, less precision. The model has to ask clarifying questions, summarise instead of listing, and leave details to written text. That is why Lumen’s voice assistant answers briefly and puts links in the chat: voice is great for understanding, text is still better for checking.

The risks, plainly

The illusion of a person. A warm voice, natural pauses and a laugh at the right moment make us forget there is a statistical system on the other side. That is why the AI Act requires people to know when they are talking to an AI, unless it is obvious. Every serious voice assistant should say so at the start and whenever asked.

Cloned-voice scams. If machines talk like us, a voice stops being proof of identity. The call from a “son in trouble” or a “director asking for an urgent transfer” is already a common scam: I cover it in the guide on deepfakes and scams. A family password and one simple rule (never pay on a phone request without calling back) are worth more than any software.

The privacy of listening. A microphone that listens is a microphone that can listen too much. Always ask: when is it on? Where does the audio go? Is it stored? On Lumen, for example, the microphone only turns on when you press the button, audio goes to OpenAI to answer and the site does not record it.

The hidden cost. Real-time voice costs more than text and the meter runs during silences too. A company that wants it on its own website should set time and spending limits.

My position

I think in a few years we will talk to AI more than we type to it, especially outside the office. It is good news for accessibility and for people who are left out today. But it needs three habits: always know whether you are talking to a machine, never trust a voice as proof of identity, and ask for the text when an answer matters.

In one line

Voice makes AI feel more human to use. Which is exactly why we must remind ourselves, out loud, that it isn’t.

Sources

Informational content, current as of the date shown. Prices, models and rules change often: always check official sources. Written with the support of AI tools and reviewed by the author.