Podfly AI - Actual Intelligence
What We’ve Learned About Using Machines to Serve Creativity
It was what now feels like not so long ago that our Creative Director began drafting our company's position on the use of artificial intelligence in our production workflows. Notionally, we would take a stance on the incoming technology, and state for the record how, when, or even if, we would use it. Fast forward to the time of this writing, and the technology I speak of has likely advanced beyond my current understanding of it.
I'm no Luddite. I currently have Perplexity, Notebook LM, ChatGPT, Gemini, and Claude on all my devices, all of which are paid versions. Moments in their implementation have been both indispensable and laughably dispensable. I'm a creative by nature, and like many of my kind, L.L.M.s are a valuable tool for summarizing, rephrasing, and conducting research. More specifically, there are two compelling implementations in my workflow: 1) getting unstuck, and 2) affirming that talent has value.
Getting Unstuck
I've had countless situations where A.I. tools have towed me out of the procrastination ditch. I take an RFP, a transcript of a 30-minute presentation, another of a sales call, various emails, links to relevant web content and materials, and dump them into an application. With a few careful prompts, I have a bulleted summary of first and next steps toward a thoughtful and comprehensive proposal and response.
When I can't find the words to reply to an email, have the time to read a 67-page scope of work before an upcoming meeting, or simply need a second opinion or fresh perspective on something, LLMs are very much like the overconfident and underpaid virtual assistant that I need. Not to do the work, but to bounce the work off of so I can clear the clutter and focus on the task at hand.
In our company, this is a powerful internal-facing option for administrative tasks, operational activities, and day-to-day requirements that come with running a small team with 50 clients. But that emphasis is on the administration of our small business.
We are a creative agency. Although we utilize AI to assist with certain aspects of office work, we never use it to deliver creative work.
Talent Has Value
Mediocrity is not only the end product of AI creative, it is the programmed intent. The swaths of material ingested by these models aim to determine the most likely output that a human might create. Be that the following string of words in a sentence, the shape of clouds in a given sky, or the amount of noise reduction on an audio file that improves the sound without distracting the listener. It reminds me of some of the government accounts we've worked on. What is the most generic and least offensive way to accurately convey our point of view? Therein lies the vanilla-flavoured contraction in governance and AI-generated creativity. There is no "point of view".
To argue what makes art art isn't my intent. Our role as creative beings isn't dissimilar to the LLM at the outset. We contain a corpus of knowledge accumulated through our learning and experience that informs what we can output. We are given access to wider ranges of datasets to fold into that knowledge and influence the working process. And that is where we split.
Two Key Areas Matter to Me
The first being that we are fallible in what LLM programmers wrongly label as hallucinations. As a program, the LLM generates something that sounds right to fill in either an information gap or follow the natural linguistic path towards something that might pass as accurate. We, too, "hallucinate". Only ours is a mislabelling of overconfidence, fallacious thinking, or hubris peppered with occasional dumb luck. That's what makes it creative. It comes from our particular, often inaccurate, point of view.
The second is that the best creatives don't ulitmately give a shit what you think. We don't need to sound right — we need to sound good. And good is what we feel, not what something objectively is. Period.
We Are A Dirty Mirror
The ugliness of our bias shapes the beauty of our creations. I've spent 50 years preparing myself for this moment, pecking away at a keyboard or speaking into a microphone, interpreting the world I've ingested, which will likely differ tomorrow and tomorrow and tomorrow. What tools I've collected internally are the experiences and knowledge unique to me - it's my database, my corpus. The sequence of next logical steps that my programming assembles is dirtied by the mirror I use to review, interpret, and reflect the life around me. It's that point of view, sullied with my messy worldview, that an AI can approximate, but never duplicate.
It's A Feeling
Right now, the feeling of AI (to me) is gross. The ick factor of an AI-generated email (em dashes and all) or a soulless summary of a podcast episode feels more likely to please the programmer than the user. That is for now. I am reminded that this is the worst it will ever be, and it's pretty darned incredible.
That's often the case with new technology. Engineers and programmers rarely build products that can predict how the artist will use them. It's through observation of its use in real-world applications that the directionality of a product is revealed.
No engineer would have imagined that a guitar amplifier intended to make an instrument loud and clear would eventually be more desirable when it becomes overloaded, saturated, and distorts. Entire genres of music exist almost entirely on that unintended consequence. We have yet to discover the artful flaws hidden in what is made to be clean code.
What now?
There's little debate that we stand on the edge of a massive shift in human existence. This time is as big or bigger than any era or advancement thus far. To be here to witness it is indeed a fantastic thing.
For our little corner of the world, AI and podcasting are only starting to intersect in valuable ways. The post-production aspects of creating content are time-consuming and, in many cases, too expensive. When all we need is a balancing of audio levels, cleanup, and polish, AI is the way to go. The "magic wand" button in our photo app is what 99% of us need to get the picture to look great and share with our friends, family, or audience. Podcast audio is no different.
However, there is that 1%. The editor's sense of timing, matched with the producer's direction and the writer’s point of view, creates something worthy of our full attention. We make those shows. They are people-made, full of flavour and “je ne sais quoi”.
Not all photos need professional editing. Not all bread requires kneading by hand. Not all podcasts need to be produced by experienced, educated humans. However, the standout moments when a photo, bread, or podcast is intended to change you with its point of view or make you feel something new, then I strongly recommend that you find a person who has spent their whole life preparing for the moment they make it for you.
There are eyes for both Karen’s Insta and Annie Leibovitz. There are palettes for both Wonder Bread and the corner bakery. There are ears for Derek’s Trucking Podcast LIVE in his shed and Radiolab. I ask you to consider what AI computes that the majority of people will like before you start making yours.
For more thoughts and perspectives on AI, and all matters podcasting, join us at our company blog over at: Podfly.net/Insights