Why Ben Affleck is Probably Wrong

Glossing Over the Likely Effect of GenAI on Hollywood

Ben Affleck being emphatic

When a lot of people ask me the same question, it’s usually a good indication it’s worth writing about. Over the last week, many have asked what I thought of Ben Affleck’s comments at the recent CNBC Delivering Alpha conference about the effect of AI on Hollywood.

Many pundits have opined about the implications of AI for Hollywood over the last year or two. Affleck’s comments struck a chord. Maybe that's because we’re at that point in the AI hype cycle where the public is primed for a skeptical take. Maybe it’s because he’s Ben Affleck. Or maybe because he’s one of the few Hollywood luminaries who's been willing to venture a specific view about AI, not just vague platitudes. But, whatever the reasons, they did.

Bilawal Sidhu, who posted a clip of Affleck’s conversation on X/Twitter, wrote that it was “[f]inally a grounded take on AI & filmmaking from a Hollywood A-lister.” Jori Lallo posted that he “[d]idn’t expect Ben Affleck to have the most articulate and realistic explanation where video models and Hollywood is going.” Affleck’s comments were also widely covered in the press, like here, here and here.

Below, I distill what he said, where I think he’s right and why his overall conclusion is probably wrong.

Tl;dr:

• Affleck essentially dismisses GenAI as a sustaining innovation in Hollywood, something that will enable Hollywood to create “more shows” for the “same spend” and could represent a “new revenue stream.”

• His logic is that AI can’t create art or replace actors, but will reduce costs by ~30% by eliminating “laborious” tasks like VFX. It will also enable fan creation and personalization, for which IP owners should be able to charge.

• I agree with many of his points but think his overall conclusion is wrong. GenAI is more likely to be a disruptive innovation than a sustaining one.

• The notion that “AI can’t make art”—and the implication that therefore anything made with AI can’t compete with Hollywood—is a red herring. Maybe AI can’t autonomously make art, but that doesn’t mean it isn’t an artistic tool.

• GenAI may not displace actors in many genres, but many of the most successful TV show and film genres make little use of (undoctored) actors, like animation, sci-fi, fantasy and horror.

• Moreover, GenAI doesn’t need to replace actors to dramatically lower costs. Above the line costs are less than 20% of most projects. Animation is the canary in the coal mine, where there is growing evidence GenAI will bring down costs by ~90%.

• Most important, Hollywood is not competing in a closed system, so the risk of lower entry barriers is not just “more shows.” It is already under assault from creator content and GenAI may accelerate that disruption.

• In other words, the risk is less that Hollywood will replace people with AI, it is more that AI will hasten the displacement of Hollywood.

• Affleck deserves credit as one of the few in Hollywood willing to go on record about a controversial topic. But Hollywood is already a place that resists change. Downplaying the potential effect of GenAI won’t help.

---

---

What Affleck Said

Here’s a link to the entire 33-minute discussion between CNBC’s David Faber, Affleck and Redbird Capital founder Gerry Cardinale. Below is the relevant clip (courtesy Sidhu). It’s worth the four minutes.

Although he doesn’t use this phrase, Affleck’s argument is essentially that GenAI is a sustaining innovation for Hollywood, something that will make it marginally more efficient. His rationale is that:

• Movies “will be one of the last things…to be replaced” by GenAI.

• That’s because AI can’t replicate the “taste” that emerges from having actors together in a room together or write Shakespeare because AI is a “craftsman, at best” not an artist.

• It will disintermediate the more “laborious” and “costly” parts of the business, lower barriers to entry and allow more voices to be heard.

• It will “hammer” the VFX business, because maybe you won’t need “1,000 people to render something.”

• According to “macroeconomics…what should happen, with the same demand and same spend, they should just make more shows.”

• It will enable personalized versions of content and some types of fan fiction.

• All told, it may eventually “take 30% off the cost of production,” increase the amount of production and should be a new revenue stream (he hopes that AI licensing will replace DVD, the decline of which took 15-20% out of of the business).

In isolation, I agree with a lot of what he said.

• Can GenAI replace emotive actors? Not yet and potentially never.

• Will it reduce costs, lower barriers to entry and empower more voices? I certainly think so. This has been the central point of everything I’ve written about the topic over the past two years, including Forget Peak TV, Here Comes Infinite TV, AI Use Cases in Hollywood, How Will the “Disruption” Of Hollywood Play Out? and GenAI in Hollywood: Threat or Opportunity?, among others.

• Will it “hammer” the VFX business? It’s easy to pick on the VFX business, which is often considered outside the Hollywood mainstream and has long struggled financially. But he’s right that it will be one of the first areas affected. As I explained in AI Use Cases in Hollywood, the most likely place that GenAI will show up first in traditional production workflows is in pre- and post-production. This includes automating some VFX work (like digital makeup from Vanity AI or de-aging by Metaphysic); localization (so-called “subbing and dubbing,” using tools like Eleven Labs, Respeecher and Flawless); and using Midjourney or DALL-E for concept art. Soon, it will become more common to use Runway Gen-3 or Sora for previsualization, too. As I also discussed in Fear and Loathing (and Hype and Reality) in Los Angeles, it will take awhile before major studios will use GenAI extensively in “final pixel” or “final frame,” the frames that show up on screen. They’re very leery of blowback from talent and, perhaps even more so, unresolved legal issues.

• Will it enable personalized versions of content and fan fiction? Yes, I agree with this too, as I explained in IP as Platform and GenAI Video as a New Form.

Glossing Over the Risks

Still, I think his overall conclusion is wrong. I believe there are several oversimplifications in his thinking that collectively cause him to dramatically understate the likely impact of AI on Hollywood. It is more likely to be a disruptive innovation than a sustaining one.

Let’s break it down.

GenAI Can’t Make Art, But Art Can be Made With GenAI

Affleck’s logic for why GenAI can’t compete with Hollywood is that it “can’t make art.” This is a very narrow view of GenAI. Implicit in this statement is either a misunderstanding or intentionally gross oversimplification: that GenAI is synonymous with zero-shot prompting. In other words, that using GenAI equates to putting in a prompt and a getting a movie.

Using GenAI is not synonymous with delegating all creative decisions.

There’s no consensus definition for art, but let’s use what Rick Rubin wrote in The Creative Act “…the attraction of art is the humanity held in it.” Under this definition, can GenAI make art autonomously, with virtually no human involvement? No, by this definition it cannot. But this doesn’t mean that GenAI can’t be an artistic tool. Put differently, just because GenAI can automate many creative decisions doesn’t mean it must.

Just because GenAI can’t autonomously create art doesn’t mean it isn’t a democratizing artistic tool.

You can easily imagine a continuum. Let’s take filmmaking as an example. At one extreme is the zero-shot prompt, like “Produce a 90-minute love story set in Rome in 430 BCE.” At the other, a person writes a screenplay (perhaps aided by AI); maybe performs mocapless motion capture on simple acting (again, facilitated by AI); creates and uploads reference images or video to an AI video model; runs many (many, many) generations on the model, prompting for specific camera angles and lighting; creates a soundtrack (also perhaps aided by AI); uses voice actors or maybe synthetic voices and a lip synch/portrait animation tool; and finally edits it all together. At this latter end, there are countless human creative decisions.

GenAI is controversial and confusing because it is able to automate so many creative decisions, but there are many creative tools that automate at least some decisions.

GenAI is controversial and confusing because it is able to automate so many creative decisions, but many other creative tools automate at least some decisions. Stating that “GenAI can’t make art” is akin to arguing that an automatic camera can’t produce art because it automatically calibrates shutter speed, aperture and ISO; CGI can’t be used to create art because it creates landscapes using procedural generation or determines how objects move using physics engines; 3D software can’t make art because auto-rigging determines how characters move; or drum machines or auto-tune can’t make art because they make decisions about rhythm and pitch.

Watch the two videos below, each made by one person. The first is “Dominant,” by Nik Kleverov, which is about his father’s life as a nonconformist artist in Soviet Russia, made with Sora. The second is “Battalion,” by Dave Clark, about the 320th Barrage Balloon Battalion, the only unit to storm the beach on D-Day that was composed entirely of African-American soldiers. He made it using a whole bunch of tools (Kling, Runway, Luma, Minimax, Midjourney, Magnific, Eleven Labs, Comfyui, Live Portrait, Stable Diffusion and ChatGPT). There is very clearly enormous creative decision making and intentionality in both of these pieces. When you watch these, do you still believe that AI can’t—or even more of a stretch, will never—make “art”?

Movies Often Don’t Require Emotive Actors

Affleck states that GenAI can’t replace human actors or replicate the “taste” that emerges when two or more actors interact on screen. As I have written before, I think this is probably right for genres that require emotive human actors, like drama, comedy, (some) action/adventure, suspense, etc.

But a lot of the most popular genres rely sparingly on humans—or at least emotive human faces—if at all. Obviously animation doesn’t require human faces. Sci-fi, fantasy and horror use humans that are heavily doctored with makeup and CGI or often entirely CGI figures. Figure 1 shows the top 10 movies at the box office year-to-date. Half are animated films or sci-fi that make scant or no use of undoctored human faces.

Figure 1. Half of the Top Films this Year Make Little or No Use of Human Faces

Source: Box Office Mojo, Author.

Potential Cost Savings Could Significantly Exceed 30%

Toward the very end of the discussion (after the end of the clip above), Affleck estimates that AI will “eventually take 30% off the cost of production.” I think that this, too, reflects a narrow perspective on the use of GenAI. The cost savings could be much larger, even if one assumes that actors are still needed in many cases.

If GenAI is just plugged into existing production processes, then 30% is probably a decent estimate. But if the entire film production process is rethought from the ground up, I think it is far too conservative.

Figure 2 shows my estimate for the breakdown of the typical blockbuster film budget between above the line, below the line, post and other costs.¹ This is probably also a decent proxy for premium scripted TV. The above the line costs, which are less than 20% of the total, include the director/showrunner, producer, top writers and top actors. The other 80%+ is, well, everything else: photography, sound, facilities, the art department (set and production design), hair/makeup, costumes, equipment rental, electrical, location shoots (transportation, permits, rental), physical effects, catering, administration, security, and all the post-production work around VFX, editing, etc.

Over time, all non-above the line costs could converge with the cost of compute.

Even if you believe, as I do, that to make anything compelling you will still need substantial human oversight of key creative decisions (a director/showrunner and writers) and, in many cases, human actors, almost everything else is up for grabs. Actors could show up, in street clothes and limited makeup, shoot for a short period on a simple soundstage, and everything else could be done synthetically. Today, for a $200 million film, the non-above the line costs—80% of the production budget—run $1.5-$2 million per minute. Eventually, this cost could converge on the cost of compute.

Figure 2. Estimated “Topsheet” Breakdown of Film Production Budget

Source: Author estimates.

This may seem extreme, but animation, which requires no actors or sets, is a canary in the coal mine. Last year, DreamWorks co-founder Jeffrey Katzenberg famously predicted that GenAI would reduce the costs to produce a “world-class” animated film by 90%.

There’s growing evidence that Katzenberg’s prediction that animated production costs would decline by 90% was right.

There’s growing evidence that he was right. I recently wrote about Invisible Universe, an independent animation studio that has moved to a GenAI workflow. Videos that previously took 10-12 business days now take several hours and production costs have declined by over 90%. Another independent animation studio, Toonstar, has also stated that its production costs are 90% less than traditional production processes. Or consider Where the Robots Grow, by AiMation Studios. At 87 minutes long, it bills itself as the first feature-length AI-enabled animated film. It reportedly took a team of nine people only three months, at a cost of $8,000 per minute, or ~$700,000 for the entire film. That’s 99% less than the average Pixar film. Is it Pixar quality? No, but with the right talent and the continuing advancement of technology, there is no reason it couldn’t be.

Lower Entry Barriers are Likely to be Much More Disruptive than More Seasons of House of the Dragon

These oversimplifications—AI can’t make art, you can’t make movies or TV shows unless you have humans emoting, if you just use GenAI to automate VFX you might reduce production costs by 30%—compound. Collectively, I think they result in him substantially understating the implications of AI for Hollywood.

As Inigo Montoya said in The Princess Bride, “Let me explain. No, there is too much. Let me sum up.” Rather than rehash everything I’ve written over the last two years about this, I’ll try to sum up.

In general, the most destabilizing force in any business is falling barriers to entry, because it enables new competitors and, sometimes, the disruption process. For Hollywood, the compounding problem is that the market is relatively saturated. As I’ve discussed before, time spent with media is no longer growing.

For Hollywood, the most fundamental question is to what degree GenAI will lower barriers to entry and therefore change the competitive dynamic for a finite amount of time and attention.

So, for the traditional TV and film business, aka Hollywood, the most fundamental question is to what degree GenAI will lower barriers to entry and therefore change the competitive dynamic for this finite amount of time and attention. I think it has the potential to change it dramatically, for several reasons:

GenAI threatens to exacerbate the low end disruption by creator content that is already underway. As noted above, Affleck said that “what should happen, with the same demand and same spend [and lower unit costs], they should just make more shows.” His example is that maybe there will be two seasons of House of the Dragon per year, not one. The flaw in this logic is it implies that Hollywood operates in a closed system, competing only with itself. It does not.

Hollywood doesn’t operate in a closed system, competing only with itself.

Hollywood is already being disrupted from below by creator content. As shown in Figure 3, I estimate that social video (primarily YouTube, TikTok and Reels) is now ~25% of all video consumed in the U.S. As shown in Figure 4, YouTube now represents 11% of all streaming to TVs in the U.S. (this includes no mobile or PC consumption), about equal to all the streaming services of the traditional media companies—Disney+, Hulu, Tubi, Peacock, Max, Paramount+, and Pluto—combined. And it keeps growing, up from 7% just two years ago.

Figure 3. Social Video is ~1/4 of Total Video Consumption in the U.S.…

Source: Maverix Insights MIDG data, Nielsen, Author analysis.

Figure 4. …As YouTube Time Spent Now Equals All Traditional Media Streaming, Combined

Source: Nielsen, Author analysis.

A visual representation of this disruption is shown in Figure 5, corresponding to Clay Christensen’s model of disruptive innovation. The idea is that creator content is working its way up the performance curve, picking off progressively more demanding customers. It has already picked off kids and unscripted. The most popular kids show in the world is CoCoMelon, which is on YouTube, and the most popular unscripted show in the world is Mr. Beast, also on YouTube. New GenAI video tools—Runway Gen-3, Flux, Sora, Kling, Minimax, LTX, Veo, Adobe Firefly, etc., etc.—promise to enable creators to continue moving up this performance curve, reducing the distinction in production values between video created within Hollywood and video created outside it. In other words, the risk is less that Hollywood will replace people with AI, it is more that AI will empower creators to hasten the displacement of Hollywood.

Figure 5. Creator Content is Already Disrupting Hollywood from Below

Source: Author analysis.

Internet scale means that infinitesimal proportions of this new content need to be good. Most creator content is not good. But the scale of creator content is vast. Many times I’ve cited my estimate that Hollywood put out 15,000 hours of new TV and film content last year, while creators uploaded 300 million hours of video to YouTube. That means that very little of this needs to compete for consumer time and attention—not exhibit comparable production values, but compete for consumer time—to upset the supply/demand dynamic. Consider that 0.01% of 300 million is 30,000 hours, twice Hollywood’s annual output.

The consumer definition of quality is changing. This, too, I’ve written about at length. The key is to think of quality not as a stated value judgement, but as revealed preference; quality is reflected by what people choose. Long-time executives and practitioners in a field tend to get anchored to a definition of quality. But consumers’ definition of quality tends to be much more fluid. That shift is clearly underway today in video.

Hollywood content will maintain an edge in production values for a long time, but those production values are of diminishing importance to consumers.

While many Hollywood executives equate quality with production values, consumers now value other attributes, like authenticity, relatability, intimacy, social relevance (whether to a small community or to broad cultural fluency), digestibility, indie, underground, niche, low friction, etc. So, even if one believes, as I do, that Hollywood content will maintain an edge in production values for a long time, that edge is of diminishing importance to consumers.

Hollywood is Better Off Confronting the Challenge than Dismissing It

Affleck deserves credit as one of the few Hollywood luminaries willing to go on record about such a controversial topic. Plus, as someone who has ascended to the upper echelons of Hollywood and whose fortunes are tied to it, he could be excused for a little wishful thinking.

But Hollywood is already fighting for finite consumer time and attention, under assault by creator content and losing ground every day. And while AI can’t “make art” whole cloth, it doesn’t need to autonomously create art to dramatically lower the barriers to entry for the practically infinite creator class and accelerate this disruption.

Hollywood is not a place that is renowned for embracing change. Glossing over the potential implications of GenAI won’t help.

1. More about these definitions can be found in AI Use Cases in Hollywood.