Believe Half of What You See and None of What You Hear

Sandra Dunham

This quote, or some paraphrasing of it, has been attributed to Edgar Allan Poe, Benjamin Franklin, and others. (See our commentary last month about checking quotes.) Whoever the original author and whatever the original statement, it is far more true in 2024 than it was in earlier times. Today, the addition of computer- or AI-driven content makes it even more difficult to know whether you are witnessing or hearing something real, or something computer generated.

We’ve long known that real photos can be altered and fake ones created. CFIC provided some hints about how to verify the authenticity of photographs in the May, 2022, edition of Critical Links. 

More recently, there has been a great deal of media attention focused on the ability of voice generative AI to create realistic sounding human speech, “not just in terms of speech clarity but also in inflection, emotion and tone variability.” And the terrifying part of this is that AI requires just 3 seconds of a voice recording to replicate that voice.

There are many benefits to AI voice technology. For people who are unable to speak, it offers the ability to create realistic, human-sounding speech from text. For example, Val Kilmer, an actor who lost his voice due to throat cancer, was able to act in the film Top Gun: Maverick using AI-generated speech. AI-generated speech also produces low-cost, reliable audio content improving on text-to-speech technology. If you’ve ever had to re-record your voicemail greeting several times because you paused, hmm’d or otherwise messed up — think about the ease of typing your message and having it recorded, sounding like a perfect you. This technology has also improved on translating between languages and auto-captioning.

However, as with most technology there are many threats to guard against. Imagine the grandparent scam (an email from your grandchild begging for help) when it comes with the voice of your loved one begging for help!

Also frightening is the ability of AI to use real voices of celebrities and politicians to create disinformation. This was used by Elon Musk in the lead up to the U.S. election when he posted a “parody” of Kamala Harris stating that Biden was senile and she would be a good “diversity hire.”   

So, what are we to do about audio content. Besides believing none of what you hear, you can try:

1. Listen for a lack of emotion in the speech (although AI is increasingly becoming better at this).
2. Listen for slurred, unnatural speech — especially when you are listening to unusual words that don’t come up often.
3. Listen for unnatural background noises, especially from speakers you expect to use professional recording equipment.

However, advances in technology mean that even these cues are becoming less available.

For more information on protecting yourself from various deep fake scams, including AI audio, check out some tips from the Better Business Bureau.