AI used to "resurrect" voices of deceased pilots

In the latest sign of an increasingly artificial intelligence-saturated era, the United States National Transportation Safety Board (NTSB) temporarily closed access to its docket system after discovering that the voices of pilots who died in a UPS Airlines aircraft crash last year had been reconstructed using AI and circulated on the internet.

Under federal law, the NTSB is prohibited from including cockpit audio recordings in its docket system, which contains extensive investigative data and has historically been open to the public, according to reports from Tech Crunch on Friday, 22 May.

A docket is an official filing system or database used to store and publish investigative documents.

However, the aviation accident docket included spectrogram files from the cockpit voice recorder. A spectrogram uses a mathematical process to convert audio signals, including low and high frequencies, into an image.

Scott Manley, a popular YouTuber whose channel covers physics, astronomy, and video games, wrote on X that the data contained in these images could potentially be used to reconstruct audio from megabytes of encoded data within them.

And that is precisely what occurred. People used the spectrograms, along with publicly available transcripts, to create an approximated audio reconstruction of the cockpit voice recording from UPS Flight 2976 in Louisville, Kentucky, according to the NTSB.

They employed AI tools such as Codex, based on social media posts.

On Friday, 22 May, the agency reopened public access to its docket system, but maintained restrictions on 42 investigations for further review — including the investigation relating to Flight 2976.