From Turntables and mixers to AI-based Radio: Some Hypotheses on the Near Future of Radio Media

ChatGPT

In December 2022, the world discovered Artificial Intelligence (AI) based on large language models (LLMs) with ChatGPT. Globally, the radio industry immediately saw countless attempts to create programs or even stations entirely managed by AI. Not exactly high-impact projects for the moment, or at least that's how it seems.

Probably, the integration of AI will be slow and profound. It will go hand in hand with the evolution of the capabilities of various LLM models and will require reimagining radio, and not just asking the various AIs to create radio as it has always been done.

Claude 3

Always remembering that the evolution of LLMs is incredibly rapid, to the point that what was true even just a few months ago is no longer true now: anyone with doubts should read, for example, what the potential is in fields such as "Graduate level reasoning" or "Reasoning over text" of the latest arrival, the Claude 3 model from Anthropic.

The Studios

Historically, before the classic console with the famous three turntables, radio corresponded to a large auditorium with musicians performing live: recorders had not yet been substantially invented, and each station relied on live broadcasting with its own autonomous production, alternating written newscasts with live performances, as can also be deduced from old Radio Schedules.

Console and Screens

This was followed by the long era of consoles with turntable virtuosos and then the current one, with studios equipped with many computer screens where music and interventions come from some hard disk, perhaps even hosted "in the cloud". And where the presenters are often guided (sometimes caged) by clocks, in accordance with management strategies.

Now and Then

The screens will probably remain in future studios as well, although it is now quite certain that the window interfaces (adopted by the various automation software) will be supplanted by text interactions, written or spoken: it's useless to struggle to find the infinite options of the various programs in the menu structure when you can simply ask the AI to do what you want.

From this point of view, the concept of LAM, Large Action Model, introduced by Jesse Lyu at the Rabbit R1 presentation seems particularly convincing.

AI-Powered device

Our hypothesis, however, is that in AI-powered radio, the center of gravity of radio stations will shift from the current pair of "speaker + automation software" to a model of "Creator + AI + one-to-one interaction".

Creator, not speaker

We wrote "creator" and not speaker. We mean to say that the human who speaks on the radio will probably be very different from the current figure of the host-logged-by-clock-and-format.

To differentiate themselves from AI, the host will have to be a creator, a person with great analytical, inventive, and - precisely - creative abilities: all qualities necessary to be preferred over an automated broadcast, as well explained in a recent article on Newslinet.

WIP

We mentioned changes in future radio studios. At 22HBG, we are thinking of an innovative device/service, provisionally called WIP (Work In Progress).

One of its features is the automatic recognition of the type of content on air, whether it's music, advertising, or talk radio programs. This capability allows for the insertion of advertising announcements in a non-invasive manner at the most opportune moments, maximizing their impact without compromising the listener's experience.

WIP can also act directly on geographically differentiated programming, taking into account location data, weather information, and individual listener preferences. In this way, users can receive real-time traffic updates, notifications of local events, and relevant news correctly interspersed with the normal flow of the broadcast.

Understanding the "mood" of listeners

WIP constantly analyzes data from various sources, including social media, listener feedback via WhatsApp, SMS, and messages.

Through the usual functions of LLMs (see an example of "system" and "user" prompts here above), WIP is able to signal similar viewpoints (in the style of "trending news") and/or provide a reasoned summary of the most interesting messages received regarding a topic covered in the broadcast. This analysis can allow those responsible to dynamically modify the programming, in order to address the topics of greatest interest on air. Or simply to make listeners aware of the prevailing "mood".

Personalized interaction

WIP incorporates numerous chatbots capable of directly dialoguing with individual listeners, through a WhatsApp-style interface directly in the radio app or aggregators such as FM-World.

Thanks to these bots, listeners can obtain additional information or participate in surveys during broadcasts. Obvious implications for what is today defined as "engagement".

And the listeners?

We believe that listeners are ready for the transition towards listening devices inspired by the Rabbit R1. The "OnAir" project of 22HBG foresees for the future the launch of an innovative "multimedia object" equipped with a touchscreen and voice control containing an AI (in the cloud, like the current Peperoni AI, or even locally, a possibility made feasible by recent SLM, Small Language Models).
A gadget capable of reproducing high-quality sound through integrated speakers, hidden from view by a premium matte finish.

But things don't stop there: within the FM-World system, this device will be in direct communication with the station's software, receiving data and metadata from it and providing it in turn to the broadcaster.

Privacy & Big Data

In accordance not only with European legislation but also with common sense, the device provides encryption of all uploaded user data, so that it can be transmitted to the station anonymously, but analytically complete. The station will still be able to create a historical "big data repository", allowing for analysis and segmentation that will become increasingly valuable over time.

Radio 3.0

Infinite are the fields of application of this true Radio 3.0, where - in line with Marc Andreessen's well-known vision "Software is eating the world" - stations will be able to find their differentiating factor in the conception and realization of personalized software.

From Watt to Python

From the era when whoever had more Watts in the antenna prevailed, to the one where those who are able to have good ideas and realize them in Python will win. And with the support of AI, it won't even be too difficult (M.H.B. for FM-World)