ETHICAL SOURCING OF VOICES

All files used for the training and creation of a Large Language Model (LLM) or AI tool used to create Synthetic voices  should be fairly sourced, with the active consent and knowledge of the person the voice files are sourced from.

If the source data is not given with consent, the AI-voice is not ethical.

CONSENT

Creating a synthetic clone of someone's voice, or using their voice data to train an AI Large Language Model (LLM) should only be done with informed and explicit consent. Assumed agreement should never be accepted.

Some AI-voice businesses claim to be "ethical AI" because the people whose voices they clone have fully consented. However, the Large Language Models these systems rely on are generally trained on non-consenting voice data.

 

CONTROL

Any person who consents to their voice being cloned using Ai technology should be able to, if they so wish, have control over how, where, and what content their voice is used for. Anyone who's voice-clone or avatar is being used by another party, that person should have the right to a detailed log of prompts and outputs created.

COMPENSATION

A Voice Actor should receive an agreeable level of compensation for their talent and potential loss of future work by creating a digital clone of their own voice.

DATA SECURITY

Protection of data sets and source voice used to create a digital clone should be of absolute importance. Third parties must protect a voice actor's property, their voice. Anyone whose voice data is stored by a third party should be safe in knowing that access to their biometric data is within their control and that appropriate security measures are being taken.

TRANSPARENCY & LABELLING

AI-generated audio, video, imagery and text-based content should be disclosed audibly and visually to ensure the public knows when the content they are interacting with is not human-created.

Watermarking in the metadata of AI-generated content should reveal the identity of the person/company who generated it, and access to the source data (eg. human voice) should also be readily available to ensure an ethical production process was used.

The Australian Government released Australia's Artificial Intelligence Ethics Principles in 2024. These were created to help -

• Achieve safer, more reliable and fairer outcomes for all Australians
• Reduce the risk of negative impact on those affected by AI applications
• Businesses and governments to practice the highest ethical standards when designing, developing and implementing AI.

One key principle the government has outlined is that of Transparency & Explainability. "There should be transparency and responsible disclosure so people can understand when they are being significantly impacted by AI, and can find out when an AI system is engaging with them."

This is not happening as AI-generated voices, video, and written content is not being disclosed as standard practice. This is why AAVA is calling for mandatory labelling of AI-generated content.