GOING FOR A SONG OP-ED: AI and the future of creativity — can artists’ histories travel with their work?

“The microwave did not kill cooking,” digital director at Universal Music Group Nic Burger said, and the room laughed.

Burger offered an apt analogy: microwaves didn’t make pots and pans obsolete, they simply lowered the threshold for cooking, a distinction whose appeal was immediately clear to me. I can’t cook, yet the two microwaves in my kitchen allow me to reach the result without acquiring the skill. Burger thought AI might do something similar for creative work.

I was at the Creative Cnergy Policy and Investment Symposium in Sandton, where bank towers and corporate headquarters make economic power unusually visible. That concentration of wealth has earned it the familiar description of “Africa’s richest square mile”. Inside, panellists sat close enough that exchanges felt worked out together as creatives and policymakers pushed past conference courtesies.

Dr Karabo Maiyane, head of philosophy at Nelson Mandela University and researcher with the South African Cultural Observatory, placed AI within a longer history of technologies that had disrupted creative life. 

Printing multiplied text beyond the hand-copied page, while photography made it possible to produce an image without drawing or painting it. Maiyane’s distinction was that AI is no longer simply helping creative work travel – it’s participating in the making itself.

Burger made that possibility personal: after 28 years finding and developing artists for record labels, he joked that he cannot sing or write, yet he could sing badly into an AI tool and ask to sound like Celine Dion. The result, he said, could return something recognisably hers. What once required years of training and an unmistakable voice can now be summoned by someone who possesses neither.

Taking it to a larger scale, Stephen Hollis, an intellectual property lawyer and partner at Adams & Adams, used the example of Deezer, the French streaming platform.

Deezer reported almost 75,000 fully AI-generated tracks daily, about 44% of all uploads. Hollis reduced that number to one artist uploading an album while tens of thousands of machine-generated songs arrived beside it.

Still thinking about Celine, I was reminded of an earlier conversation with Mbali Njomane, a South African social innovator working at the intersection of culture and AI literacy. We had discussed African users meeting AI systems designed and trained in the West, and whether we, as Africans, would remain consumers or shape the languages and uses they recognise.

“We’ve always been consumers of technology, but we’ve never been framed as co-creators,” she said, shifting the question from access to agency. Then she reframed the notion of authorship: “What we call inspiration is data.” It made me wonder how much of what I call originality began as someone else’s work.

I recognise that in my own writing. The cadence I think of as mine was learnt from sentences I read years before I understood them. A musician absorbs another’s phrasing until influence becomes instinct. Humans have always created from material that was never entirely theirs.

That thought came back to me in Sandton. Humans are trained too, so borrowing cannot be the whole objection.

The intricacy of authorship, however, is not a novel dilemma.

In 1939, Solomon Linda recorded Mbube with The Evening Birds in Johannesburg. The song evolved into Wimoweh and later The Lion Sleeps Tonight, becoming globally famous, though Linda’s connection to its immense value took decades of investigative reporting and legal battles to recover.  

What Linda’s story points to is the distinction between creation and the work’s entry into the machinery of distribution, ownership and payment – a circumstance that becomes more complex with AI now in the picture.

Pfanani Lishivha, chief executive of the South African Music Performance Rights Association (Sampra), illustrated this concern of authorship. One generative AI song, he said, could have “up to 50 original songs making that song”. He took the imagined track apart: A guitar from one recording, bass from another, a recognisable voice and somebody else’s drums.

I stopped hearing one song. Instead, I heard a roomful of earlier musicians folded into one new AI-generated work, and I wondered whether Lishivha’s imagined track might already resemble one of those 75,000, arriving on a platform with more human histories inside it than a listener could ever know were there.

Sampra is building what Lishivha called “attribution technology” to identify human performances inside AI-generated music. One technology could assemble the track in minutes, and another had to move backwards through it, finding who remained and where payment should go.

“Because we can’t pay the machine,” Lishivha said.

Burger had offered three conditions: Consent, credit and compensation. Lishivha showed what happens to those principles at AI scale. As he listed guitar, bass, voice and drums, I realised how much human labour I assume sits behind a song that sounds fully formed. A polished recording once implied studio hours and trained ears. Now, AI changes something more fundamental.

AI can reproduce the evidence of influence without undergoing influence itself. A generated song can carry the phrasing, timbre or tonal habits of an earlier artistic language without the years of listening and making through which influence becomes part of an artist’s own practice.

That’s where credits begin to matter differently. A credit can look like administrative residue after the work is over. When one generated work can carry traces of many earlier creators, it becomes a record of relationship, preserving whose judgement shaped the work and giving the listener a route back into the culture it emerged from.

Back in Sandton, Mbali’s argument about participation turned into a more challenging question: If African languages and cultural contexts remain marginal in the data shaping these systems, Western assumptions can acquire the appearance of neutrality.

Greater African participation can change what those systems recognise, but participation alone does not decide who owns what follows or who is paid when culture travels. It becomes an old imbalance wearing new technology.

Tomorrow, another 75,000 machine-generated tracks may reach Deezer, and somewhere in South Africa, a songwriter who has never had access to studio time may make a first song, perhaps sounding uncannily like Celine Dion.

Then they press upload.

The song leaves their hands and joins thousands of others. AI can widen who gets to make culture, even as it makes the human histories inside each work easier to lose.

The song may travel further than ever, but the question is whether the people inside it can travel with it. DM