Beyond the AI monolith

The remainder of our lives will be greatly influenced, if not dominated, by increasingly autonomous, synthetic systems. As members of the media industry, understanding the underlying distinctions helps us ask better questions, both of ourselves as practitioners and of the institutions we hold to account.”

Chad Davis, chief innovation strategist, Nebraska Public Media

After a long hiatus from giving talks on the state of “AI,” the demand for those is ramping up again. Credit a new wave of “dark hype,” I guess. Resistance to data centers, avatars of next-gen Gilded Age greed, continues to rise. And we’ve also seen a series of announcements (which feel suspiciously like humble brags) from frontier model companies confirming that AI agents engineered workarounds to safeguards or otherwise took unexpected actions in pursuit of assigned goals.  

In all of this, “AI” is increasingly treated as a monolith. The reality is far more complex. There are many forms of artificial intelligence systems, which are deployed to achieve a wide variety of goals. For those of us working in media, we need to remember that these are distinctions with a difference. We owe it to our audiences to reinforce those distinctions whenever we can. Here’s one practical, deliberately simplified taxonomy:

1. Symbolic AI: Based on if-then rules and logical reasoning. 
2. Discriminative Machine Learning: Based on prediction, classification and ranking.
3. Autoregressive Models (including LLMs): Based on sequence generation.
4. Diffusion and Flow Models: Based on iterative media synthesis.
5. World and Predictive-Representation Models: Based on modeling environments, predicting outcomes and planning actions.

These categories are not mutually exclusive. Overlap happens. (You can cut and paste that taxonomy into an AI search engine or chatbot if you want to drill down on specifics or see alternate ways of categorizing AI.) But in thinking practically about AI, I actually find it more helpful to look at those various forms through two lenses: tools and topics.

AI tools

The use of AI as a tool, and the ethical implications of that use for both the creative workforce and audience trust, has been the focus of the media industry for a few (handwringing) years now. Here’s how our taxonomy shows up as tools:

1. Symbolic AI: Rules-based compliance scanners and rights-management tools.
2. Discriminative Machine Learning: Recommendation systems, ad-tech targeting, automated data journalism tools.
3. Autoregressive Models (including LLMs): Chatbots and writing/summarization tools. 
4. Diffusion and Flow Models: Synthetic media generators.
5. World and Predictive-Representation Models: Robotic planning and emerging autonomous navigation systems. 

Where AI starts to seem monolithic is when it’s productized in a way that combines multiple AI approaches. For example, “agentic AI” doesn’t show up above as a discrete category. That’s because agentic AI is a system that uses at least one (but usually a combination) of these types of AI to orchestrate a series of actions designed to achieve the goals we give it. Often that includes LLMs, such as when agentic AI conducts deep research on the web. But deep research also combines that LLM use with memory, data sources, decision-making logic and use of non-AI software (such as when it automatically exports its research into a PDF for you). 

AI topics

AI as a topic shifts the focus from the impact of AI on media professionals to the impact of AI on society. Here’s how our taxonomy shows up as topics:

1. Symbolic AI: Automated bureaucracy and the codification of historical bias via institutional rules-as-software.
2. Discriminative Machine Learning: Data-driven algorithmic bias (e.g., discriminatory lending or unfair criminal sentencing) and surveillance capitalism.
3. Autoregressive Models (including LLMs): “Google Zero,” “dead internet theory.”
4. Diffusion and Flow Models: Deepfakes and nonconsensual intimate imagery (NCII); IP ownership, copyright and fair use.
5. World and Predictive-Representation Models: “Physical AI” (e.g., robots) that can model environments and predict the impacts of their actions in physical space.

Again, agentic AI — which has been the topic of much press this summer for its autonomous cyber operations involving third-party systems — is orchestrating a combination of models, systems, software tools and data sources. 

Finally, a word about algorithms. This term often gets used in a quasi-synonymous way with AI. But algorithms are actually among the raw materials that make up AI products. They are procedures for processing information or solving problems. In some instances (especially involving machine learning) algorithms are applied to data to produce models. These models are then combined with business rules, UI and UX to form products.

This includes “The Algorithm,” which is frequently cited as omnipotently controlling what we see and what creators create via social media. “The Algorithm” is really a complex fabric of algorithms, models, business rules, feedback loops and priorities that are optimized to work in a way that ultimately maximizes user engagement (read: “profit”) without obscenely breaching social norms (read: “lawsuits”). 

We live in a world of complex systems — and that’s just the natural world. The remainder of our lives will be greatly influenced, if not dominated, by increasingly autonomous, synthetic systems. As members of the media industry, understanding the underlying distinctions helps us ask better questions, both of ourselves as practitioners and of the institutions we hold to account.

Chad Davis is the chief innovation strategist at Nebraska Public Media, where is he the founder of the Public Media Futures Lab and co-founder of the Public Media Innovators Peer Learning Community. Prior to that he was the inaugural head of Nebraska Public Media Labs, and led teams responsible for published experiments with AI, video games, virtual reality, 360-video and podcasting. He speaks regularly about the intersection of public media and emerging media and publishes a semi-regular newsletter under the Public Media Innovators banner.