The Future of Radio in an AI World
Updated
Radio has already survived several predictions of its death: the television, the cassette tape, the CD, the MP3, the streaming service. Each time, the technology that was supposed to replace it instead changed the landscape around it — and radio found its footing in the new terrain.
Now the question is a different kind. Artificial intelligence is not just a new delivery mechanism; it's a potential alternative to the editorial act that defines radio. If AI can programme a station, read the weather, voice a presenter, and generate music — what does that leave for humans? And what does it leave for radio?
This is an honest attempt to answer that. Not with scaremongering, and not with tech-sceptic dismissal, but with a clear look at what AI does well, where human curation still holds ground, and what radio might look like in ten years.
What AI Already Does in Radio
It's worth being clear about current reality: AI is already doing meaningful work in audio and broadcasting, and it's been doing it longer than most listeners realise.
Automated programming — Music scheduling software has used algorithmic decision-making for decades, selecting tracks based on tempo, energy, recency, and format rules without human intervention on each individual choice. This is a primitive form of AI programming that has long been part of commercial radio infrastructure.
Voice generation — Text-to-speech technology has improved to the point where generated voices are now often indistinguishable from human recordings at short listening distances. Some stations already use AI-generated voices for news reads, weather, and traffic updates. The economic motivation is clear: recorded voice is cheaper than live presentation.
Music recommendation — Streaming services have spent years developing AI systems that recommend music with notable precision. The technology is genuinely good: given enough listener data, these systems make accurate predictions about what any individual listener is likely to enjoy.
Content generation — Large language models can now produce coherent broadcast copy, programme descriptions, and even scripted presenter material. AI tools can assist writers, generate station idents, and produce advertising copy.
Music generation — AI systems can now produce music that sounds credibly human across several genres. The quality continues to improve.
None of this is hypothetical. These capabilities exist and are in use. The question isn't whether AI will enter broadcasting — it already has. The question is what it changes.
What AI Does Genuinely Well
This deserves a direct answer, without deflection.
AI recommendation systems are good at personalisation at scale. No human curation team can generate different programme streams for millions of individual listeners, calibrated to each person's time of day, mood signal, and listening history. Algorithmic systems do this reasonably well and will do it better. That's a real capability.
AI voices are improving fast. For factual, functional content — news, weather, sports scores — the difference between a human voice and a well-trained AI voice is narrowing. Some listeners may not care which is which if both are accurate and clear.
AI-assisted music discovery — not generation, but analysis — can surface connections between pieces of music that human programmers might miss: tracks with similar harmonic character, tempo, or emotional register across very large catalogues.
AI can also reduce the cost of entry for small and independent stations. Tools that once required expensive production staff — editing, mixing, voice recording, scheduling — are becoming cheaper and more accessible. A community station with no budget can now produce a more polished output than was possible five years ago.
These are genuine contributions. Acknowledging them is necessary to think clearly about what the future actually holds, rather than a future we prefer to imagine.
Where Human Curation Still Holds Ground
The case for human curation is not that AI is bad. It's that some things matter in listening that AI is not well-positioned to supply, at least for now.
Editorial conviction — A human music director makes choices they believe in. They play a track because they think it's important, even if it's unfamiliar, even if it risks audience attrition. That willingness to bet on aesthetic judgment — to lead rather than follow — is structurally different from a system optimised to predict and confirm preferences. See The Case for Human Curation for the full argument.
Cultural embeddedness — Community radio, college radio, and specialist genre stations work because their programmers are part of the culture they're broadcasting about. A presenter who has followed a local music scene for a decade brings something that a model trained on streaming metadata doesn't have: participation, relationship, and the kind of understanding that comes from being there. Community and college radio is the clearest expression of this.
The social dimension — Radio at its most vital is a social medium. A live presenter is aware of external events — a local crisis, a cultural moment, the weather outside — in a way that creates real connection with listeners experiencing the same thing. AI can simulate this, but simulation of social presence is not the same as social presence.
Serendipity by design — The human programmer who puts something unexpected into a sequence is taking a deliberate aesthetic risk. The AI recommendation system is minimising risk: it optimises for preference satisfaction. These are structurally opposed editorial stances. Discovery — genuine discovery, of music you didn't know you wanted — tends to come from the risk-taking stance.
Taste development — The most influential figures in radio history didn't just reflect audience taste; they helped form it. DJs, music directors, and programme controllers who championed artists before they were popular contributed to the cultural landscape as active agents, not just data-driven responders. AI systems optimised on existing data cannot play this role.
The Likely Shape of the Next Decade
Rather than a binary outcome — AI wins, radio survives — the more credible picture is differentiation.
Fully automated, AI-programmed stations will proliferate. Some will be very good at their jobs: consistent, targeted, cheap to run, and predictable in a way that suits background listening. For many listeners, this will be entirely sufficient.
At the same time, the value of human curation may increase as a differentiator precisely because it becomes rarer. When the default is algorithmic, the stations that sound unmistakably like a person made editorial choices will stand out. There's a parallel with other media: when digital photography became ubiquitous, film photography didn't disappear — it became valued more consciously.
The formats best suited to survive and thrive are those where human judgment is most obviously the point: specialist genre stations programmed by genuine experts, community stations rooted in specific places and cultures, and themed collections assembled editorially rather than algorithmically. These won't outperform AI on scale or personalisation. They'll offer something AI can't: the feeling that a person who cares about music chose this, and chose it for you.
Live presentation — actual humans on air in real time — may become more prized as AI voice becomes commonplace. The knowledge that a voice is genuinely human, responsive to the moment, capable of surprise and error and genuine enthusiasm, carries a different weight when the alternative is clearly available and clearly cheaper.
What Doesn't Change
Through every technological transition radio has faced — and there have been many — one thing has remained constant: someone decides what plays. That editorial act is what makes radio radio rather than a music delivery pipe.
The editorial act may be performed by a human, or assisted by AI, or substantially automated. But the choice of what a station stands for — what it will and won't play, who it's for, what it sounds like at 3am versus 9am — is still an act of will and judgment. That won't be automated out of existence; it will just be made by different people, or by people using different tools.
Why radio still matters addresses the underlying resilience of the format. The future of radio is not a question of technology. It's a question of editorial conviction — whether there are enough people who care enough about what they broadcast to make something worth listening to. The history of radio suggests there always have been, regardless of what the technology looked like.
72FM and the Human-Curated Approach
72FM's themed collections and how we curate reflect a deliberate editorial stance: that the most interesting listening experience comes from human editorial decisions about which stations belong together, what moods they serve, and what character they share. This isn't positioned against AI; it's positioned around what we think human curation does that algorithmic aggregation doesn't.
In an AI world, that position is likely to become more clearly defined, not less.
Frequently Asked Questions
Will AI replace radio DJs? AI can already replicate some of what a DJ does — announcing tracks, providing links between music, generating patter. What it can't easily replicate is genuine editorial conviction, cultural embeddedness, and the live social responsiveness that makes the best radio presenters feel like a genuine human presence. Whether that matters to listeners will vary. Specialist and community radio is most insulated from AI replacement; high-volume commercial formats are more vulnerable to automation.
Can AI really create good radio programmes? AI can create programmatically coherent radio — consistent tempo and energy, smooth transitions, appropriate content for format rules. Whether that constitutes a "good" programme depends on what you value. If you want a consistent, predictable listening experience calibrated to your preferences, AI-assisted programming can deliver this. If you want surprise, editorial courage, or the feeling of a human intelligence at work, that's harder to replicate.
Is AI-generated music a threat to radio? AI-generated music will create a much larger volume of music to programme from, which may reduce barriers to starting stations (no licensing costs for original AI music) but may also dilute the sense that you're hearing something real. How listeners respond to knowing that music was AI-generated is a genuinely open question.
Will internet radio still exist in 10 years? Almost certainly. Internet radio's core advantages — global reach, genre specificity, low operating costs, freedom from terrestrial broadcast constraints — are not undermined by AI; if anything, AI tools reduce the cost of running a station further. The question is what kind of internet radio survives and what kind of radio listeners prefer.
How is 72FM thinking about AI? 72FM's current approach is centred on human themed browsing — collections assembled by people, not generated algorithmically. That approach reflects a belief that what listeners value in radio is editorial conviction, and that conviction is a human quality. As AI tools evolve, the interesting question is how they can assist human editorial judgment without replacing it.
Summary
AI is genuinely capable, and it will change radio in ways that are real rather than speculative. Automated programming, AI voices, generated music, and algorithmic curation are already present in the broadcasting landscape and will become more prominent.
What they haven't replaced, and what looks difficult to replace, is editorial conviction — the choice to play something because you believe it's worth hearing, not because a model predicted it would be accepted. That conviction is the root of what makes radio feel like radio rather than a stream. It doesn't require expensive technology. It requires people who care.
Radio has always been most interesting when someone was willing to bet on that. The future probably isn't different.