A recurring theme in this column is that change is the only constant in media technology and now we’re entering yet another inflection point. For the past two or three years, the conversation has been dominated by generative artificial intelligence (GenAI) large-language models, synthetic media and the promise (and peril) of machines that can create. But a new concept is emerging to take center stage: Agentic AI.

This shift is more than just a buzzword swap. It represents new thinking in how we consider automation and interaction. Where GenAI focused on content generation, Agentic AI is about delegation and communication. While the technology is still maturing, the trajectory is clear: agents are coming.

What Is Agentic AI, Really?
Before we go further into the world of agentic systems, it’s worth stepping back to clarify what we mean when we talk about “AI.” AI is not a single technology—it’s a spectrum of capabilities, each suited to different kinds of problems.

Over the past few years, the spotlight has been on generative systems—models that can produce text, images, audio or video based on patterns learned from large datasets. Attention is shifting toward something more dynamic: Agentic AI. These are systems that don’t just respond—they act. They can pursue goals, make decisions and interact with other systems or agents on behalf of a user.

A key distinction between AI and traditional automation lies in determinism: Traditional automation excels in deterministic environments, where inputs and outputs are well-defined and predictable. Think of a transcoding pipeline or a playout automation system. These are engineered for consistency and reliability.

AI, by contrast, thrives in nondeterministic contexts—where inputs may be ambiguous, incomplete or constantly changing, and where outputs are not always binary or fixed. This makes AI especially useful in areas like content personalization, natural language interaction or adaptive media workflows, where flexibility and learning are more valuable than rigid rules.

As we move into the agentic era, this distinction becomes even more important. We’re building systems that can operate in the gray areas, where human judgment used to be the only option. This is a significant evolution in how we think about automation. These agents may be powered by generative models, but they go beyond them by incorporating memory (context), planning and the ability to interact with other systems or agents. In some cases, it may even act without direct prompting based on what it knows about your goals.

The key point is this: Agents are not just tools. They are actors in a system, capable of making decisions, forming strategies and interacting with other agents in ways that mirror human delegation.

A New Media Ecosystem
To understand how agentic systems might reshape the media landscape, it helps to visualize the ecosystem they could create. That’s where the Agentic Model for media comes in.

At the center of the model is “Agentic Discovery and Communication.” This is the core function that ties everything together: the ability of agents to find, filter, personalize and exchange content on behalf of their human or organizational counterparts. This is a foundational concept: the emergence of a general agent communications plane-A layer that sits “above” the internet as we know it today. This plane would allow agents to interact, negotiate and transact with one another directly, without requiring constant human mediation.

Some envision a future where this agentic layer becomes the dominant interface for digital interaction—potentially superseding the traditional web. In such a world, websites and apps may become secondary to the agents that navigate the digital world on our behalf.

Surrounding this core are four key roles:

• A Creator Agent might help manage rights, optimize distribution or assist in content creation or personalization;
• A Brand Agent could autonomously place ads for a brand, negotiate campaign terms or monitor performance;
• A Personal Curator Agent would act on behalf of the consumer, filtering content, managing preferences and even negotiating access or pricing; and
• An Influencer Agent represents any entity granted the authority to shape or guide the behavior of other agents. This could include institutions, communities, regulatory bodies or even parents wishing to influence the curator agents of their children.

What emerges from this model is a vision of a media ecosystem where agents mediate nearly every interaction. It’s a shift from a platform-centric media model to an agent-centric one, where the locus of control moves closer to the individual or organization being represented.

Personal Agents and Data Ownership
Among the most transformative elements of the Agentic Model is the Personal Curator Agent—a digital representative that acts on behalf of an individual consumer. This agent doesn’t just recommend content: it negotiates access, filters noise, adapts to evolving preferences and potentially even manages subscriptions or monetization decisions. It becomes, in effect, a media concierge—one that knows your tastes, your values and your boundaries.

The need for such a capability has never been more urgent. We are rapidly approaching—if not already living in—what some have called the “dead internet,” a digital landscape increasingly saturated with AI-generated content, synthetic engagement and algorithmically amplified noise. In this environment, the content signal-to-noise ratio is getting worse. We will need agents to sift through the junk and identify what truly matters.

For a personal agent to be effective, it must have access to a rich and continuous stream of behavioral, contextual and preference data. That data might come from viewing history, social interactions, biometric signals or even inferred emotional states. In today’s media environment, much of that data is collected and controlled by platforms. But in an agentic future, the balance of power could shift—from platforms to people.

The Long Road Ahead
The pace of technological innovation often outstrips the pace of business model evolution. We’ve seen this before—file-based workflows were technically feasible long before they were widely adopted. Cloud infrastructure was ready years before media companies trusted it with their core operations. Even streaming, now ubiquitous, took more than a decade to become mainstream.

The core technologies—autonomous agents, large-scale models, distributed orchestration—are already emerging. But the real constraint isn’t technical; it’s organizational, economic and cultural. Business models will need to adapt. Rights frameworks will need to evolve. Standards for agent behavior, identity, and trust will need to be developed and adopted. And perhaps most importantly, people will need time to adjust to the idea of delegating meaningful decisions to machines. It’s a decade-long transformation, at minimum.

For media professionals, the message is clear: don’t wait for the future to arrive—start preparing for it now. Begin experimenting with agentic workflows. Rethink how your content is discovered, curated and monetized. Invest in data quality, interoperability and flexible infrastructure. And most importantly, stay curious.

Over the years, there has been one thing I have consistently written about: change. How are things evolving, and what can we do to get ready for these changes? In this column, I want to share my thoughts on the future, based on four decades of experience in guiding clients through uncertain times. Recently, many of us attended the NAB Show, which reminded us that the future in the media industry is always uncertain.

Change is unpredictable. If you look at the NAB Show agenda from 10 or 15 years ago, you will notice that some big topics back then, like 3D television, did not have the impact we thought they would. On the flip side, today’s hot topics, like artificial intelligence, barely got a mention back then.

No one can predict the future perfectly. To get ready for different possibilities, it is crucial to have a flexible business model, a talented team, and adaptable technology. These factors are all connected; a change in one affects the others. During my career, I have seen a lot of unexpected changes.

Oddly, innovative ideas would pop up quickly and shift our thinking but fully implementing them often took much longer than expected. For instance, moving from tape to file-based workflows took decades after it started.

Think about how things have changed over the last 40 years:

1985
When I first got into the industry, tape formats were the critical topic. Betacam had come onto the scene, but everyone was wondering if they should wait for Betacam SP or M2. We had computers, but they were just for text and not part of our workflow. Tube cameras were the norm, and edit rooms looked like space launch control centers. Cable TV was growing like crazy.

A camera operator needed a big, shoulder-mounted camera with a battery belt and a tape deck. Using the camera meant serious training and daily upkeep. Editing required a room that cost $20,000 to $30,000 for basic cuts and up to a $1 million suite (in 1985 dollars) for full effects.

Back in 1985, distribution channels were limited. Without network backing, you would have had to go through public access TV or distribute VHS tapes yourself, which made getting your content out there tough.

1995
We were now getting into file-based systems. Computers could make videos! We developed nonlinear editing and many other computer-based technologies and began to slowly—very slowly—get rid of tape in our facilities. We had solid-state cameras. Cable had exploded and the internet was just entering our consciousness.

2005
The digital revolution was underway in more than one sense of that word. Digital TV! Internet-based video services. Standardized HDTV was here, and in the U.S. we were trying to figure out how to deal with the move from 4:3 to 16:9 aspect ratios. Phones did more than make calls and were everywhere. They had screens that were playing video. User-generated content (UGC) was arising in online services. Cat videos were everywhere.

By 2005, that cameraperson had an integrated camcorder that was solid state with no need for separate battery and VTR. Consumer camcorders were in the market for under $1,500 with superior performance to the images of 1985 and it was a more stable and easier-to-use device. We now could edit on a computer and an edit suite was $5,000 with full mix and effects capabilities.

Think about what a difference that is—in 20 years, we made professional-quality media production capabilities available to anyone who could pull together single-digit thousands of dollars, down from nearly $1 million. Many more people could make content, and they were.

You had many more options for distribution. There were hundreds or even thousands of TV channels and you also had internet-based distribution options including early social media and streaming services. You had a greater chance of being seen by audiences.

2015
Change continues; now we are talking 4K, even early 8K! Streaming is all the rage and we are now immersed in augmented reality (AR) and virtual reality (VR). ATSC 3.0 was on the drawing board. Targeted ads were driving new revenue and the competitive landscape was changing with new media entrants.

2025
Any number of trends are critical today. Of course, there is AI, as has been discussed extensively in this column. But not just that—our business is being severely disrupted by streaming wars and the rise of social media, especially by Generation Z. We have mobile devices that can shoot professional quality videos and even edit them.

Today, all the capability you need to make professional content is in your pocket for under $1,000. You can shoot and edit high-quality content with professional-grade mix and effects with the device you carry around with you every day: From around $2 million in today’s dollars to less than $1,000 in 40 years.

You also have a tremendous number of options to get your content out; self-publishing is available to anyone on video-sharing sites and social media and the audience is now there. You can go “viral” if you have the right content and monetize without the same level of difficulty you would encounter just 20 years before.

2035
Can you predict what 2035 will look like?

I cannot. But there are some clear trends here we should recognize to help us look forward. There is a megatrend in the above: From 1985 to 2025, content production and distribution were democratized dramatically, from the few people who could make and distribute professional content to almost everyone.

I have every reason to believe that these democratization trends will continue. We have evidence that they will already. When it comes to production the obvious continuation is with generative AI. And there will also be new forms of distribution with Agentic AI playing a role. More on that in a future column.

Be Ready for the Future
Regardless of what future comes, we can be prepared for anything; we can act now. Here are some key truisms I have learned in architecting systems and workflows while engaging in transformation efforts from getting off tape to turning on AI.

What is most important here is to focus on business. All the technology and operational decisions are downstream of what the business is trying to accomplish. Of course, engineers are creative, too, and should become even more embedded in business to provide innovative ideas.

From a technological perspective, there are a couple of key things that will always be good to engage in—not only for your needs today, but to be ready for the future. You need to have a highly flexible infrastructure on which the business runs and a systems architecture that is flexible and easy to change. What does that mean in the modern era? Well, certainly commodity hardware and virtualization where possible, and cloud is appropriate in many cases to provide flexibility.

From a systems standpoint, you need the ability to manage and orchestrate best-in-class capabilities and operations. I started talking about Service Oriented Architecture nearly 20 years ago. You need to decompose your business requirements to workflow steps that fulfill those requirements; you then need to look at how to create services that can be reorchestrated or changed without breaking everything else. The term “Service Oriented Architecture” is not dead and applies today just as much as it used to. It’s worth getting familiar with, as AI does not deprecate SOA.

And then there is our data layer, which often lacks a unified source of truth or consistent dwata. We need to continue to engage in the endless project of making our data layer better through master data management, governance, metadata cleansing and even metadata generation. We need to do everything we can to have a clean pool of data our systems can rely on. AI only really works with good data to train and execute with.

I hope all this gives you comfort that you will not be out of work soon. It has been an incredible 40 years. I believe that in the next 40 years, we will have at least as much change to work through and enjoy. We all have much to do between now and 2035. Let’s get to it!

Artificial intelligence, and its implications for the film and TV industry, were among the main topics of discussion at the SMPTE DC Chapter’s “Bits by the Bay” conference held in Chesapeake Beach, Maryland, from May 21-22.

The conference—organized by the late Peter Wharton, former chief strategy officer for TAG Video Systems—returned after a long hiatus, and the event is dedicated to his memory.

Rounding Up
In a presentation devoted to fractional film frame rates, James Snyder, formerly with the Library of Congress’ National Audio Visual Conservation Center (NAVCC) and now an industry consultant, discussed the evolution of frame rates in tandem with the development of film and television and the expectation that with the film industry transitioning to digital, that frame rates can be modified to address the streaming market.

The battle waged over interlace vs. progressive image standards during the transition from analog to DTV several decades ago illustrated the coming battles between broadcasters—most of whom favored interlace—and the computer industry, who favored progressive.

In a world where 24 fps in traditional film is really 23.98 frames per second (fps) for TV and digital video, and 60 fps is technically 59.94 fps, Snyder said it’s time to round up to the next number (an even integer) because advances in technology have made fractionals less relevant in today’s media distribution ecosystem.

As streaming and gaming have taken hold over the past several decades, progressive has won that battle, and streamers simply don’t care about frame rates, Snyder added.

“Interlace and fractional are two sides of the same coin—they're both problems we need to get rid of,” he said. “[Interlace is] the easiest to get rid of right now, because all displays are now natively progressive; we have no CRTs anymore and progressive is baked into our technical distribution now, or at least the end viewing. And frankly, getting rid of fractional is also baked into it.”

Snyder blamed the current U.S. broadcast TV standard.

“The only reason we can't do even integer right now is because ATSC 1.0 is in the way; ATSC 3.0 can accommodate it, the displays can accommodate it,” he said. “Every vendor I've talked to who has equipment that does production or post production can switch to even integer. The only thing that's standing in the way right now seems to be ATSC. 1.0 is a technical blocker.”

Snyder said he came to this conclusion in his work at the Library of Congress, where he worked in film digitization and restoration. Adherence to global media standards and the demands of storing and managing millions of hours of film and video content prompted him to think of new ways to approach the problems of cost and efficiency.

“We had to deal with the problems of interlace, we had to deal with the problems of fractional frame rates, and we had to deal with the technical issues that came with creating new versions for reuse, because people could license the content, and they would ask for various versions,” Snyder said. “And so I got to see firsthand the extra cost. And the reality is, there are increased costs just in producing new content, there are increased production costs when it comes to reusing old content.

“And so that's where a number of us got together and asked ourselves, ‘do we really need to be doing this?,’” Snyder added. “A genuine technical reason why we're still doing this, and the only one we can come up with is ATSC 1.0. If you take ATSC 1.0 out of the mix, ATSC 3.0 doesn't have that problem, and streaming certainly doesn't have that problem. Streaming doesn't care what your framework is. Your computer just adjusts it to present the content.”

Snyder added that this approach needs to be modified when considering live broadcasts.

“The one area that still needs some discussion and some work is, ‘how do you go from live feeds that are fractional to live feeds that are integer?’ We do need to talk about that. There are converters out there that do the work we can do… but they have a strange cadence to them. We need to figure out how to not have that strange cadence, and work it out in the software and the hardware.”

AI and Democratization of Media
John Footen, managing director with Deloitte, discussed the democratization of media production over the past four decades and how AI is impacting its future, pointing out that for as little as $500, today’s mobile device can produce professional content that cost thousands of dollars to produce in the 1980s. (Footen also pens TV Tech’s Media Matrix column.)

“You don't even have to have an education to do it, people are just doing it … and that's an overall trend of democratization in content production and distribution,“ he said. “Are we done democratizing media? Is it all democratized out? The answer is no.”

The ability for the average Joe to produce Disney-level content powered by AI is coming sooner than we thank, Footen added.

“That level of production is going to be theoretically possible for good or for bad, it's going to happen,” he said. “And it’s going to further democratize the content types that can be created by anyone. You don't have to be good at using a camera because you don't have to touch a camera to make an image, right? You don't have to get a lot of training and framing and all that stuff, because a machine will frame it for you.”

Footen elaborated on agentic AI, the concept of using AI “agents” to do specific tasks that could eventually replace the “app” environment of today.

“You can think of it interchangeably with the word 'apps' that you've been using today,” he said. “An agent is a system … it could be a simple system or it can be a complex system that performs a function. It actually does something like you would expect out of many kinds of applications.”

As these agents become more adept at each individual’s media tastes, this trend will lead to a breakdown of brands that still act as a sort of “walled garden.”

“A curator agent is an agent that works on your behalf to get your content for you today,” Footen said. Using the Plex video playback app as an example, Footen said the more information a user adds to it, the more intelligent it gets, the less important the origination.

“All you need to add to that is keeping track of who you are and what your preferences are, and giving it access to your data and giving it this AI capability to curate for your content,” he said. “Eventually, that curation disintermediates the media company, because why do I care what channel it's on, what app it's on… if it's just coming to you through my agent, my agent is my app. I don't care where the content came from.”

Content Validation
Matt Galek, Footen’s colleague at Deloitte, discussed the issue of authenticated content in the age of AI, elaborating on the latest developments around the Content Authenticity Initiative (CAI), a six-year-old project designed to to promote transparency and authenticity in digital content; and the C2PA, a consortium of tech companies tasked with developing the technical standards for content credentials, among other things.

C2PA is the “leading standard” to confirming and maintaining trust in digital content and is on the path to becoming an ISO standard this year, according to Galek, who pointed to the three pillars of technology that drive the standards developed by the CAI: metadata, fingerprinting and watermarking.

“All these together provide a solid foundation for authenticity,” Galek said.

While C2PA’s standards don’t deal with detection of AI-altered media, it starts with the assumption that most of it is now, Galek said. “C2PA kind of assumes that users are going to be changing a piece of media, and [in] some of the research I was looking at, most studies were saying that over 50% of people posting content to social media platforms have manipulated in some way, whether it's due to editing that content or adding a filter,” he said. “So that's a significant amount that has been altered in some way, and kind of unknowingly, you're consuming that media.”

Galek pointed to the importance of metadata in tracking a piece of content’s evolution through the complete production chain and how that vital information is getting lost when content migrates to various platforms. Adopting C2PA will help protect that information, he added.

“There's a ton of tech that’s used today where metadata gets stripped inherently out when media is transferred from one system to another,” he said. “For vendors adopting this standard, you will have that methodology to retain that metadata through an entire supply chain.”

Working groups with the C2PA are focused on areas like audio and video watermarking, but Galek said there has to be intra-industry cooperation.

“There needs to be an additional spotlight on the implementation side of things, to provide a forum for broadcast-based and other industries to collaborate,” he said. “We know how increasingly fragmented the broadcast vendor landscape is, so it’s that much more critical for us to collaborate together around content authenticity.”

Detecting Deep Fakes
Anyone who has dabbled in generative AI knows how better it is getting with time but altering video and images (aka “deep fakes”) has been done for decades, something Ed Grogan, a motion imagery researcher and former Defense Department official, discussed during the session “The Dark Side of AI.”

(Grogan discussed his work in video forensics with TV Tech in 2018).

“When we started this, it took a little Hollywood studio to really make good deep fakes, and they couldn't do much,” Grogan said. “We’re now at the point that some of these levels where that little Hollywood studio was are [now] being done by middle schoolers.”

Using a historic picture of Winston Churchill, Franklin Roosevelt and Josef Stalin at the Yalta conference in 1945—with Groucho Marx in place of the Soviet dictator—Grogan emphasized the importance of spotting discrepancies as early as possible, since a widely distributed AI-altered image could rapidly alter public perception. In other words, it could allow a lie to travel halfway around the world before the truth has its boots on, a line often attributed to Churchill himself.

“If you think it's fake, put it in your documentation, because someone 20 years from now might lose the reference material,” he said. “And if you don't say you think it’s fake and the audience didn’t have the historical reference, they might not know that really should've been Stalin and that’s the wrong Marx. So sometimes, because they don’t have the historical understanding, you can get something past somebody to tell your story.”

Although no one sees media companies losing their gatekeeper status anytime soon, Footen summarized the promise and perils of AI in removing media companies as the “middleman” for content consumption.

“If everyone can do what we're doing today, I think our industry is going to be disintermediated,” Footen concluded. “I'm not sure how much of a media industry we have left when everyone is a media company. Every person, every company, is a media company. Our job, historically, has been to create the content, to gather the audiences and to monetize that content. But what if we’re not needed to do that? That’s kind of happening already.”

The contemplation of artificial intelligence has a long history, arguably predating the invention of computers. Since the advent of modern computing, the hype cycle for AI has repeated itself numerous times. In 1970, Marvin Minsky, a pioneering figure in AI, was quoted in Life magazine saying, “In from three to eight years we will have a machine with the general intelligence of an average human being.”

This forecast did not materialize then and remains unfulfilled. Once again, the current AI hype cycle is approaching the "trough of disillusionment," but it is expected that we will soon reach a new "plateau of productivity" as the latest advancements are assimilated.

I have been somewhat taken aback by the rapidity with which the present cycle has moved past its peak, leading to more grounded expectations throughout the industry. Over the past couple of years, there has been extensive discussion about the hype and associated fears of AI.

In this column, I aim to delve deeper into the challenges inherent in AI technology itself. Many of these challenges exist independently of their application in media and can arise in any context. Although my focus will be on technological issues, it is crucial to acknowledge substantial non-technical concerns such as economic implications, rights and royalties, cultural transformations, and the legal and regulatory landscape surrounding the technology.

Within the scope of technological challenges, there are those where a resolution path is foreseeable and others where no clear solution currently exists. An example of a challenge with a visible path to resolution is electricity usage. We have multiple methods to generate electricity and can eventually develop the necessary infrastructure. Here, however, I will concentrate on a few challenges for which there is currently no evident way to resolve.

When talking about any form of technology, it's crucial to first define the term. In the context of AI, this definition is quite expansive, covering a range of underlying technologies. Commonly, AI now refers predominantly to Generative AI, particularly Large Language Models (LLMs) and associated technologies such as Generative Adversarial Networks (GANs) and others. This article will concentrate specifically on issues related to LLM technology.

Accuracy, Reliability and Quality
By now, we are all aware of some inherent issues in the predictive nature of LLM technology. One of the most prominent concerns is the tendency for these models to "hallucinate," producing results that contain objectively false, bizarre, or highly improbable information. While a certain degree of this can be beneficial, especially in creative tasks, it poses significant challenges in many situations.

Completely avoiding this issue is difficult. Architectures such as RAG (Retrieval Augmented Generation) aim to mitigate this by automatically supplementing the prompt with additional constraining data from traditional data systems like databases. Though this approach is promising, its inconsistent performance makes it difficult to rely on these tools for automated operations. For use cases that demand greater predictability and reliability, it is advisable to use historical automation technologies or trained human operators.

Artificial intelligence has demonstrated its proficiency in handling generic tasks, particularly those with abundant training data. Unfortunately, in creative fields, this often results in subpar content. Most data available to train AI content systems tends to be of low quality and lacks exceptional creativity.

Furthermore, due to the AI's inherent tendency to generate outputs that align with the "average" result, models typically produce quite unremarkable content. To date, AI has not consistently delivered high-quality, creative outputs, and achieving such results seems implausible without human intervention.

Model Collapse
Arguably, the greatest danger to the future of AI is AI itself. Numerous academic studies conducted over the past year have highlighted an apparent irony in our current approach. Essentially, the more effective AI becomes in serving our needs, the less beneficial it ultimately may become. This phenomenon is referred to as model collapse.

As illustrated below, LLMs (Large Language Models) are particularly compelling because they excel at predicting the next likely word, pixel, or other data points in their outputs. These predictions are highly accurate because they tend to average out the possible results. I previously mentioned this issue as a quality concern where the models generally generate "average" content.

A more intrinsic issue emerges from this statistical behavior. If we deploy AI in the real world and the volume of data produced by AI systems increases, the model becomes increasingly skewed by this averaging effect. Paradoxically, the more we use AI effectively, the more challenging it becomes to train it for future utility, leading us towards even more homogenized content.

Even with first-generation models today, you can observe a kind of uniformity in system outputs. As an experiment, try entering 12 different prompts about an elephant or another favorite animal into your preferred image generation platform. Then, use an image search engine to find real-world animal pictures. You'll notice a distinct similarity in the AI-generated images compared to the natural diversity found in actual photos.

The images below illustrate how rapidly these issues arise. After merely five generations of training on datasets containing AI images, the models start producing remarkably similar images. The greater the proportion of AI-generated data within the dataset, the more pronounced this problem becomes.

Introducing AI-generated data into the training set may lead to various other issues, especially if a broader variability is permitted to prevent uniformity. Rare events, like visual artifacts, could become more prevalent and rapidly degrade the dataset in unusual ways, as illustrated by the images below.

I compare the model collapse issue to the problems our industry has faced with analog generational loss. Just as we experienced increased dropouts and image fuzziness with successive generations of tapes, model collapse behaves similarly in its impact.

There are various techniques under discussion to prevent model collapse, but they all present their own challenges. One commonly suggested method is to train exclusively on data that hasn't been generated or influenced by AI. However, if AI effectively serves its purpose, such data will become increasingly scarce. This situation presents a significant irony not lost on this particular human!

Biased Training Data
One of the principal challenges with large language models (LLMs) lies in the inherent bias present in the datasets used for their training. These datasets are frequently sourced from the internet or private data collections. Considering that over 99.999 percent of worldwide information and experiences are neither online nor digitized, these sources introduce a significant bias in training.

The sheer volume of "data" consumed by an individual human daily far exceeds the amount available on the internet. And most of this is neither particularly compelling nor worth preserving and will never be in a form a computer can absorb.

Moreover, the simplest and most apparent realities are seldom directly published on websites, blogs, or other online platforms. Consequently, data is inherently biased towards what we find interesting enough to publish.

A pertinent example is the prevalence of smiling faces in image searches for people. Individuals typically post and store images where they are smiling, which skews the dataset. This bias results in generated images frequently depicting smiling individuals, which does not reflect real-life diversity.

Conclusion
I am inherently optimistic. Despite the challenges I've mentioned, I am enthusiastic about the innovative solutions being tested to address them. It is crucial, however, for media technologists to gain a deeper understanding of the fundamental workings of these technologies to effectively implement them within their organizations.

In my next article, I will explore the essential skillsets we need to develop over the next few years to ensure successful adoption.

The Olympics are more than just a sporting event—they are a showcase of human stories that inspire and captivate us. As media technology evolves, so does our ability to share these stories with audiences around the world. Artificial Intelligence is one of the technologies that is transforming how we experience sports media, especially at the Olympic level. AI can help us tell the story of each game or match, the historical significance of the event, and the ways we can keep fans engaged even after the event is over.

From Past to Present
Ever since the modern Olympics began in 1896 (Athens), people have been fascinated by the games. And with each new era, the latest technological innovations were used to bring the games to viewers across the globe. The first coverage of the event was through newsreels in theaters. Radio started with live coverage in 1924 (Paris), 100 years ago. You might be surprised to learn that television coverage began in 1936 (Berlin), with a kind of closed-circuit system that showed the games in public places near the stadium. It wasn’t until the1960 games (Rome) that the broadcast reached other countries.

The Olympic Broadcasting Services (OBS) was established in 2001 with the core mission to act as the host broadcaster for the games, delivering the sights and sounds to viewers all over the world. It started fulfilling that mission with the 2008 games (Beijing) and has been responsible for the main infrastructure and media related services for every Olympics since. OBS has been remarkable in its constant innovation with each event, maintaining quality and reliability during a time when media technology has changed a lot.

AI is part of this change. Depending on which specific AI technology we are talking about, we can say that AI was first used by OBS at the 2018 games (PyeongChang) where it was used for content tagging, recommendations, and language translation. It was also used in several other areas of the games, such as time recording systems and biometric analysis of athlete performance.

When in Paris…
According to OBS, the 2024 Games will be the most advanced yet in terms of technology with the two week event. The games will be fully produced natively in UHD HDR, along with immersive 5.1.4 sound using more than 1,000 camera systems and 3,600 microphones.

OBS will produce more than 11,000 hours of content and process more than 3,000 UHD and HD feeds within the International Broadcast Center. More than 80 different distribution feeds will be managed. The IBC facility covers about 40,000 square meters, a 13% reduction from the 2020 Tokyo Games held in 2020. A total of 36 different venue broadcast compounds will be supported.

Amazingly, OBS does this with only about 160 full-time employees. The core group expands to more than 8,000 people from more than 110 different countries during the games itself. This is an incredible organizational and technological achievement that only gets more sophisticated with each Games.

Of course, AI is also playing a bigger role at this scale of event production and distribution. AI will be used for auto-clipping of content and descriptive metadata tagging. The technology will be used to provide transcriptions and translations of interviews for journalists and to help them find content. It will also be used to provide captions/subtitles in real time for live coverage. Data gathered by various biometric and other sensors deployed through the event will be processed and provide unprecedented information to viewers, often in real time.

Perhaps most impressively, it will be used to create automated highlights for potential distribution to several different platforms, including those using vertical formats. These will be generated on demand at any level of interest from a county to a sport to an individual athlete and can be tailored to mood or many other factors that producers may want to consider.

2026 (Milan) and Beyond
AI technologies of all kinds will continue to play a role in allowing media companies to produce major live events with richer and more sophisticated viewing experiences. Even in the areas where AI is already playing a role, it is easy to see the potential for an even greater role. For example, clips and highlights packages could be generated for smaller audiences or even an individual viewer in a reasonably cost-effective way.

These highlight-generation technologies could include even more relevant stats and help producers and viewers find even more interesting “gems” hidden in the content. It can also help production teams fill time in between gaps in the action with some real-time context setting, or information about the sport, scores or other key story lines. It can bring real-time information with context directly to presenters and allow them to provide even more texture for audiences.

It can also help production crews keep an eye on what’s happening outside the venue and have a more 360-degree view of the “story of the games.” Audiences always want more and we’re reaching the limits of what’s feasible for a human production crew to do, so working AI into production and creative workflows will be crucial to bring better experiences. It will bring otherwise hidden stories to those who want to see them. AI is unlikely to ever be used to alter the “reality” a fan sees; instead it should serve as an enabling function to enhance our production of the content.

Fear of the potential impact of mal-intended generative AI creating fake or distorted content is an important concern that is being taken seriously by the IOC (International Olympic Committee), OBS, and broadcasters around the world. OBS has been explicit in its commitment to not tampering with the video.

One technology I would anticipate playing a role in coming games in this regard is C2PA (Coalition for Content Provenance and Authenticity), an open technical standard providing publishers, creators, and consumers the ability to trace the origin of different types of media. This will be critical in many areas beyond sport.

Between the Games
We can start thinking now about how AI will help with the story between the games. It might not be so obvious, but 17 days of competition isn’t enough to tell the lifetime story of how an athlete got there. When the flame goes out, preparation for the next games begins right away. AI tools that are looking beyond the content broadcasters generate and into social platforms and the broader “conversation” can help audiences experience the whole journey to the games and beyond.

The legacy for competitors, for cities, for fans who were present at these epic moments and watched on TV far away can be joined up using the careful application of AI to continue to activate fans throughout the cycle and maintain interest.

At its core, the Olympic games are a human event—an AI cannot compete in the games and can never tell the powerful stories in a way that will bond in a powerful, emotional way with viewers. AI’s roles are becoming clearer, however. In addition to providing capability for increased scale and sophistication, it is time to see AI as a technology to bring more of the “truth” to viewers. AI can assist us in bringing out powerful stories by helping us find and present more relevant and informational content than ever. l

John Footen is a managing director who leads Deloitte Consulting LLP’s media technology and operations practice. He can be reached at usmediamatrix@deloitte.com.

"Matrix” is a word with many different meanings, but they all share the idea of connecting different things into one. In our industry, we often use it to describe the switching equipment that connects signals to various destinations, like an SDI (Serial Digital Interface) “matrix.” 

I decided to name this column “The Media Matrix” because I will discuss how many new technologies are linked to media. Using technologies as varied as cloud computing, analytics and AI (Artificial Intelligence), media technologists are constantly changing our companies, the way that creators make content, and how audiences enjoy it. 

I will share this space with industry leader Karl Paulsen, where we will talk about the future of tech in media. We are ready to dive into any hot tech trend that pops up and share our insights. 

AI at NAB Show
Like the 2023 show, AI was clearly the buzzword at the NAB Show in April and we are going to zoom in on that first. There are many aspects of our business that will be impacted. From preproduction to distribution, every step of the media workflow has the potential to be transformed as AI evolves.

One of the first things that struck me in Las Vegas was how few booths had AI plastered all over them. Sure, there was a lot of buzz about AI at the show, but I thought every booth would have something akin to a flashing neon sign.

Maybe it’s because AI is not new to our industry, and we’ve learned to tone down the hype just a bit. I think technologists are looking for real examples they can apply to their operations now for practical use. There are some areas in media where AI is solid and ready to go—you can see for yourself how subtitling and language translation and even voice generation are working well. You can use AI to analyze content and add metadata to make it easier to find and use. You can use AI in post to improve images or create mattes. These and other use cases are worth exploring and implementing now.

At the conference sessions I managed to attend or speak at, there was also a lot of talk about the future possibilities of AI. While people often worry about fake content, in one session there were discussions about how AI can help bring more facts to journalists or directly to viewers in news and sports content genres. There were also discussions about how AI can help personalize content in digital distribution.

What I found most interesting were the number of seasoned technologists who were saying how we’ve seen this kind of disruption many times in media and how the “doom and gloom” predictions about how such disruptions impact creativity or jobs are often wrong.

AI’s Impact
In the next few months, we’ll have a lot of interesting topics to cover here:

Defining AI: What is AI? What are the different kinds of AI? How does AI relate to machine learning? What is the future of AI? What are the limits of AI? How do various types of AI work under the hood? Is the future the big models that get all the attention now, or smaller, more focused approaches to AI?

Workflows and Integration: What parts of the workflow in a media company can benefit from AI and how? How do we make AI work well with other technologies in our workflows? What standards are there for interacting with AI and what standards should media develop?

People and Management: How will AI affect jobs in the media and entertainment industry? What skills do technical folks need to learn to be ready for more AI in their work? How do we deal with the human side of change?

Infrastructure and Architecture: How do we set up our computing and storage layers to handle the heavy AI load in our environment? How do we design our data layer to improve the quality of data used to feed and train models? How do software application layers talk to AI models? Do we have enough power, cooling, and space to run all this stuff?

Cost and Financial: How much does AI really cost and is it worth it in different use cases? Are there ways to optimize the real-world costs of AI? How can we monitor the technology to give us efficiency insights?

Creativity: Can AI take over various craft or creative functions? How “creative” can we expect AI to be? What new content types or genres will AI enable in the media space?

Rights and Legal: What are the intellectual property issues with AI? How do we make sure that AI outputs won’t put our business at risk? What technological controls can we use to help protect the business?

Viewer Response: Do viewers really want a lot of highly personalized content? What about genres like sports where the shared experience is part of the value? As we collect more data about viewers how can we protect their privacy and ensure appropriate controls?

These questions are not all purely technical, but technology issues have been merging with business issues since media adopted IT practices. Today’s broadcast engineers are always asked to evaluate their technology investments through a business lens.

We won’t just talk about AI here. We have some slightly older tech trends still evolving like cloud and blockchain and many more. Interestingly, all these trends intersect with each other. For example, the C2PA (Coalition for Content Provenance and Authenticity) specification is a new approach for content provenance that uses watermarks and blockchain technologies to deal with critical issues like the risk of fake AI content in our ecosystem.

But let’s be honest. We are at the start of the hype cycle for AI, and it’s important to be well-informed on the subject for the many discussions that are coming and so we’ll focus there first.

Writing this column is something I’m very excited about. I’ve been in the media technology field for more than three decades, and it just keeps getting better and more fascinating.