BOCHUM, Germany — AI technology presents numerous ethical and regulatory challenges. The media industry, in particular, is grappling with complex problems surrounding the multifaceted ethical and legal implications of generative AI-produced content, complicated by the unprecedented speed of progress in the field — beyond the ability of the regulatory bodies to keep up.
In February, OpenAI announced Sora, a generative AI that can produce high-quality video clips from text prompts. It followed that in March with Voice Engine, which can convincingly recreate an individual’s speech patterns from a 15-second clip. Even though neither product has been released on the open market and the company is working with the development community to establish guardrails for their usage, their very existence has thrown up a whole raft of issues.
These include, for example, the ethical questions in replacing human speech. In radio production alone there are multiple use cases including reading news reports in the middle of the night, automating links between music tracks by reading out appropriate metadata fields and targeting advertising to the individual streaming listener level. Some we can consider AI enhancements, but some, such as automatically covering the graveyard shift, would no doubt replace a human.
Widespread disquiet
While such a use benefits a small radio station that can do more with less, there has been plenty of enthusiasm from corporate CEOs for reducing headcount, with all the attendant ethical considerations that follow. And that is just the start of the conundrums. If on-air talent is replicated using AI, are the proper contracts in place? Who owns the voice? Was permission given for its use, and in what circumstances? If the talent leaves the station, do they take their IP with them, or are those rights the station’s in perpetuity? If so, what is fair compensation? Were the AI “clones” trained on closed or open data from the internet? If it is the latter, their output could be a breach of the law.
These are all profoundly difficult questions with far-reaching impacts. There is also widespread disquiet about bad actors using this technology. Individual countries are erecting legal frameworks to cover these situations, and more, but consensus at an international level is lacking.
Numerous initiatives incorporate watermarking technologies into either human-generated or AI content so that people can tell whether the content is AI-generated. To date, a universal solution has been elusive.
The Artificial Intelligence Act
One of the difficulties regulators and developers face is the speed of progress and the seismic nature of some of its implications. Even the broadcast industry, a sector used to rapid change, is left in the wake of constant developments. Governments are slower to act. The EU is leading the regulatory charge in much the same way it did with introducing the GDPR to proscribe limits on the use of consumer data. The Artificial Intelligence Act was adopted on March 13, and became a law on Aug. 2.
It subdivides AI use into high, limited and minimal risk. Use in the creative and media industries will likely fall into the latter two. The law will impose conditions for media surrounding the ethical use of intellectual property and the transparency of the models creating them.
Where AI is originating content, transparency will prove essential.
When the European Council passes the law, enforcement will start in a window from six to 24 months afterward to allow time for industry compliance.
This is, of course, almost a geological era compared to the pace of AI development. Meta’s Llama2-70B Large Language Model took one month to train when it was released in July 2023. Estimates are that it could be done in a single day using the latest chips. One of the key developments this year will be on-device AI that will allow users to access generative AI models securely without going to the cloud.
Transparency is essential
So how do broadcasters navigate these choppy and ever-changing waters? One of the lessons learned from the introduction of the GDPR is the vital importance of trust. This must permeate everything, from the interactions with the end users to the collaborative networks that the modern broadcasters and their vendors build into the broader broadcasting ecosystem. When we are using generative AI or when it is being used within our products by a plugin, for example, that usage needs to be flagged. It must be clear to the listener that the voice from their radio at 3 a.m. is not a real human being, and, equally, metadata flagging generative AI content must follow that content throughout a newsroom system.
This doesn’t have to be for every AI use case. Many day-to-day tasks have already pivoted towards AI, from cleaning up the dialog on a noisy recording to translating text. Still, where AI is originating content, such transparency will prove essential.
Both regulation and public perception of the use of generative AI are currently behind the development curve. By being open and transparent with their use of it and by ensuring that their usage follows delineated ethical guidelines, broadcasters can ensure they are ahead of the changes still to come.
The author is vice president of consulting services for CGI.
More intelligence, less artificial
Benztown introduces AI Listener Voice Generator
The Innovators: AI voice technology for broadcast: A professional audio perspective
