The AI 'speech' deception: piped dreams and stolen voices

Increasingly AI systems are talking to us in human voices. What are we speaking to when we answer back?

In the second of his series of video essays, 'The Wind and the Song', Lewis Heriz explores the deception of AI voices.

We now live in a world where any voice you hear (and the words they speak) may be algorithmically generated.

AI voices appear in videos, training courses, podcasts and adverts. But what are we really listening to when we hear them? And if we talk back what do we think we’re speaking to?

In The Wind and the Song Pt II: Pipe Dreams, I wanted to ask what is lost in human experience if we talk not to humans but to databases made manifest as ‘voices’?

To say that a machine ‘understands’ because it has all the text it needs to replicate the forms of communication we use, is surely to misunderstand what it is to be alive in the world?

Making part two of this video essay series posed a particular set of challenges, since A.I. technologies have advanced so fast since The Wind and the Song Part One. For instance, the cloning tool I used for my own voice (in part one) sounds, comparatively speaking, unconvincing, and that was created less than a year ago.

A.I. models like NotebookLM can now quickly generate a half-hour, virtually presented ‘podcast’ that’s indistinguishable from any other run-of-the-mill, human-presented U.S. show. So, I had to include those virtual voices as a marker of where we are in this new phenomenological frontier.

As I make these video essays about AI unfolding around us, I am in a battle between an old form and a new, emerging one: one which feels like it is desperately attempting to be as average as possible, in order to convince me it has worth.

And the effect of that on me, as a human, is to try to prove that is precisely NOT what being a human is — that there is no such thing as an ‘average’ individual.

This is why, despite it being a serious and important subject, I have tried to inject little moments of humour and the unexpected — in order to distinguish it from the exponential and unrelenting rise of media that’s created with little or no input from any human agents.

I promise this intro text was written entirely by my own hand.

Your (fully human) responses to this conversation and any other Fully Human output are always treasured. We have an inbox should you want to get in touch: info@fullyhuman.org.uk