Author: David Newhoff

David is an author, communications professional, and copyright advocate. After more than 20 years providing creative services and consulting in corporate communications, he shifted his attention to law and policy, beginning with advocacy of copyright and the value of creative professionals to America’s economy, core principles, and culture.

Sci-Fi Film Written by AI is Still Fundamentally Human

Image by Pond 5
Image by Pond 5

Back in June, ArsTechnica hosted the online debut of a short film called Sunspring. Directed by Oscar Sharp and featuring the actors Elizabeth Gray, Humphrey Ker, and Thomas Middleitch, the film was made for the Sci-Fi London film festival according to guidelines for the 48-Hour Film Challenge, and it placed in the top ten out of hundreds of entries.  What is most distinctive about Sunspring, though, is its screenwriter Benjamin. No last name.  At least not one he’s told anyone yet.  You see, Benjamin is an AI.

Writer Analee Newitz describes Sunspring as the product of Sharp’s own fascination with artificial intelligence, which led to his friendship and collaboration at NYU with researcher Ross Goodwin. Listed in the film’s credits as Writer of Writer, Goodwin is the chief architect of the AI—an LSTM recurrent neural network—that would eventually name itself Benjamin. “To train Benjamin, Goodwin fed the AI with a corpus of dozens of sci-fi screenplays he found online—mostly movies from the 1980s and 90s. Benjamin dissected them down to the letter, learning to predict which letters tended to follow each other and from there which words and phrases tended to occur together,” writes Newitz.  The whole process itself is very interesting, and I recommend reading her article to learn more.

The finished film is definitely engaging, though I would not personally subscribe to the descriptions hilarious and intense as stated in Newitz’s headline. But to each his own, and headlines are headlines.  What Sunspring emphasizes for me, of course, is not a contemplation of machine intelligence but the significance of human interpretation. Benjamin’s absurdist script is a list of non-sequiturs, both in dialogue and stage direction, making the film project an experiment that almost asks the question, “Can we make a watchable movie based on the screenplay of a madman?”  The answer is of course you can.  Because cinema is very much an interpretive medium—both for makers and viewers. We can’t help but interpret; it’s what humans do.

The distinction between Sunspring and the oeuvre of human-crafted, experimental, non-narrative cinema—sometimes comprising stream-of-consciousness writing akin to Benjamin’s composition of algorithmic probability—is subtle to the point of nitpicky. Sunspring is odd, yes, but barely so if one is familiar with a film like Daisies or Hallelujah the Hills or the works of David Lynch.  The difference, of course, is that Sunspring’s absurdity—at least at the script stage—is accidental while these other works are not. Having said that, though, artists do make instinctive choices all the time that defy literal analysis, and audiences make poignant meaning from of these expressions that were never intended or even considered by their authors.

Sunspring’s script is humorously absurdist, though presumably not in a manner of which its author could possibly be aware.  The experience of watching the finished product shares strands of comedic DNA with the same mechanism that makes the Bad Lip Reading series work—because it’s funny when a real person or a character says something absurd in an earnest manner.  When BLR has Mitt Romney on the 2012 campaign trail say to a supporter “Thank you for the bench,” the same comedy chromosomes are at work as when Sunspring’s Humphrey Ker says, “We’re going to see the money.”  Benjamin has no idea why these things are funny, but they are funny in a non-literal way that is indisputably human.

Sunspring may represent a baby step toward the expectation that an AI will inevitably write a traditional, narrative screenplay for a major motion picture.  As I wrote in a very early post, a comparison between human-only, formulaic script development and machine-made or assisted, formulaic script development may prove to be indistinguishable.  Instead of leading down that path, however, Sunspring reminds us that cinema is often most compelling when convention and formula are broken.  And giving the responsibility to an AI of writing the blueprint for a film is certainly one way to achieve broken conventions—not unlike the artist who might experiment with narcotics to break down barriers to his or her subconscious.  Naturally, the more an AI resembles or reflects us, the more we assume its destiny is to replace us.  This is always the two-part conversation, right? There’s the gadget question that asks what an AI can accomplish, but there’s also the existential question that asks at what point we can say the AI has an identity, which is really a reflexive inquiry about our own existence.

So, here’s a hall-of-mirrors thought exercise:  might a more advanced AI than Benjamin have written a very different screenplay for the film The Enigma Code about the life and work of Alan Turing?  Personally, I like certain things about that film but was ultimately disappointed because I felt the work neglected an opportunity to explore the narrative in which the father of AI—the inventor of the Turing Test to determine the “identity” of the machine—was a man who literally had to pretend to be someone he was not.

So, if Benjamin’s great-grandson were the co-writer of a biopic about Alan Turing, might “he” bring a unique empathy for Turing’s duality given the AI’s own centaur-like existence?  And if so, wouldn’t we have to call that writing?  I think we would. On the other hand, absent the capacity for empathy or the existential question, the script is just barely structured words on a page that, as in Sunspring, only humans can interpret has having any meaning at all.

The Librarian of Congress is not the nation’s copyright authority.

Library of Congress Main Hall Washington DC, USA Photo by Marco Rubino Pond5
Library of Congress Main Hall Washington DC, USA
Photo by Marco Rubino Pond5

Just because the surgeon general serves at the pleasure of the president, that doesn’t mean we think the president is, therefore, the more qualified expert in medicine.  We want a president to have views on domestic healthcare in general but not to have opinions about actual medical practice.  That would be scary.

Yet, solely on the grounds that the U.S. Copyright Office operates within the Library of Congress, many of the usual copyright antagonists—Public Knowledge, EFF, et al—are lately promoting the idea that the Register of Copyrights is meant to follow the Librarian’s lead with regard to copyright policy.  This rhetoric has been buzzing a little louder since the nomination of Dr. Carla Hayden to the position of Librarian, who copyright skeptics view as an ideological ally, and who will be sworn in on September 14th. Although Dr. Hayden is exceptionally qualified for this position and, like all librarians, has her own views about the role of copyright, the office she is about to hold has at no time in history performed the duty of the nation’s authority on copyright law and policy.

For the first 107 years after the 1790 Copyright Act was passed, there was no U.S. Copyright Office at all. Copyright law itself was a patchwork of state and federal statutes; the courts were still parsing the meaning of the IP clause and looking to England for precedent; and authors’ works were registered by depositing copies with the federal courts in each state.  Meanwhile, the Library of Congress was exactly that—a department located in the front of the U.S. Capitol, staffed by fewer than ten people, and containing a book collection of some 80,000 volumes by the early 1860s.

In 1864, President Lincoln appointed Ainsworth Spofford to the position of Librarian of Congress, and it was Spofford who had the first grand ambitions for what the Library could become.  His vision included growing a vast collection of works for the nation and building a monumental facility where the collection would be housed.  In order to expand the collection, Spofford lobbied to have the copyright registration and deposit functions centralized at the Library, a procedural change that became policy with the second general revision of the copyright law in 1870.  Moving deposits to the Library worked to so great effect that by 1874, according to the LOC website, there were more than 70,000 books piled on the floors of the office.

So, the procedural change in deposits grew the collection rather quickly, but Spofford lacked the funding (and the room) to fully handle the workload he’d asked for. An 1893 New York Times story refers to an eight-week delay in issuing copyright certificates as well as a significant backlog of filings and correspondence pertaining to copyright.

New York Times - January 1893
New York Times – January 1893

Most of Spofford’s energy by then was focused on building the new facility itself, but when the doors opened in 1897, figures as notable as Melvil Dewey were already advocating a new role for the Library that went beyond merely amassing the largest repository in the world. Dewey’s view was that the Library of Congress should become “a center to which the libraries of the whole country can turn for inspiration, guidance, and practical help.”

So, in July of 1897, Spofford stepped into an assistant role, and the Library was reorganized as an institution that would fulfill this new mission of which Dr. Hayden will now be the new custodian and leader.  Later that same month, the Copyright Office was established as a separate entity within the Library; and a figure with specific expertise in copyright law, Thorvald Solberg, was appointed as the first Register.  Solberg was directly involved in drafting the 1909 Copyright Act and in directing American policy with regard to international copyright and trade. Every Register since Solberg has fulfilled this advisory role on copyright policy.

New York Times - August 1893
New York Times – August 1893

Spofford’s legacy of deposits to the Library via the copyright registration process has been maintained, and this keeps the public collection growing as intended. At the same time, the Librarian of Congress has often provided comments and views on various issues to the Copyright Office but historically defers to the Register with regard to specific policy recommendations.  Just like every other agency in the country, the USCO has grown along with copyright law’s increased complexity as  protectable works have scaled in volume, variety, and means of distribution.

In 1897, when this division of labor was first established, the medium of motion pictures was in its infancy, and the first Pianola (player piano) had just been invented two years prior. Copyright law was likewise still relatively primitive and about to collide with the mechanical age when the 1909 Act would truly begin to reflect to the modern world we know today.   In contrast to the 1890s, when the first registered “movie” at the LOC was 45 frames long, the 21st century USCO has a separate Visual Arts Division that operates under one of three Associate Registers.

The complexity of contemporary copyright touches multiple sectors of the American economy, representing more than 5.5 million jobs and an estimated 6.5% of GDP. As such, the Copyright Office is called upon to provide counsel and guidance to courts, Congress, other federal agencies, and the general public on a wide range of topics that implicate copyright law. Although the Register of Copyrights is organizationally subordinate to the Librarian, it would be a huge mistake to confuse this with a functional subordination.  The two offices perform very different and very necessary roles. The parties who are recently seeking to dismiss the Copyright Office as having neither authority nor expertise to perform its mandated duties are standing on historical quicksand.  Nobody should throw them a rope.

Motion Picture Unions Opposed to FCC Set-Top Proposals

Photo by ponsulak Pond 5.
Photo by ponsulak Pond 5.

As noted several times on this blog, it takes a lot of highly skilled people to make a film or TV show, and these people generally do not own any copyright interest in the works they help make or any equity in the production companies.  Motion pictures and television production is a project-to-project business. Crew members and performers are not full-time employees of either studios or production companies and may go months between jobs; they’ve been “gig-economy” workers since long before anyone thought that term was a new thing.

Just like most labor in the United States, today’s motion picture craftspeople are the beneficiaries of hard-fought rights—many negotiated decades ago—to share in the financial rewards of successful products they do not own.  Films and TV shows are mostly made by middle-class, freelance workers whose average, annual incomes comprise not only day rates and overtime, but also residuals and health and pension benefits.  These terms are negotiated and managed for most crew and performers by the unions DGA, SAG-AFTRA, and IATSE. These unions are opposed to the FCC “Set-Top Box” proposal because, as it stands, the proposal would break the licensing structure on which their compensation packages are based.

To be a sustainable workforce, performers and crew members generally need to remain on the net-positive side of several averages—average number of days worked, average number of shows worked that succeed, and average number of shows that succeed in the market overall.  By taking the macro view of the ways in which these workers are compensated over time, it’s very much a rising-tide-raises-all-boats paradigm.  The successful show that Props Master A works on feeds the health and pension program of Make-up artist B, who might work a show that doesn’t make it. The spread of investment across multiple shows keeps the pool of skilled labor generally sustained among the various gigs and periods of unemployment between gigs.

Photo by sokolodv Pond 5.
Photo by sokolodv Pond 5.

It should be clear to anyone that if you disrupt the means by which labor is compensated, labor has a habit of not working.  And as stated in previous posts on this subject, the FCC proposal cannot be called a consumer-focused plan to introduce better and cheaper TV options if the plan simultaneously kills the means of production. This is exactly what the proposal can do by enabling companies like Google to create a parallel, commercial video service without licensing any of the programming.

It’s Not About Boxes

FCC Chairman Wheeler emphasizes the amount consumers spend renting set-top boxes from cable companies, and then “digital rights” groups like EFF and Public Knowledge echo the sentiment that this proposal is about innovative technology (just like the VCR) that will give consumers more flexibility in viewing options for programming that we’re already paying for. And it will unleash us, they say, from the rental boxes owned by the cable companies.  But the rental fees thing is a smokescreen for what’s really going on because the box rental part of our cable bill isn’t the biggest line item; there’s nothing in the proposal that would technically lower the cable bill; we’re free right now to go buy boxes and not rent from the cable providers; and the new licensing market is already providing consumers with viewing options way “beyond the box.”

The big talking point that is most likely to confuse consumers is that the new box Google wants to sell us would only make programming available for which “we are already paying” via the Multichannel Video Programming Distributors (e.g. cable companies).  This is the central reasoning why supporters of the proposal claim that it does not implicate a copyright infringement, and it’s the kind of talking point that will sound reasonable to consumers.

But this reference to our subscription fees completely misrepresents how the producers—and therefore all the labor represented by the unions—get paid for the programs they make.  Our subscription fees to MVPDs do not pay to produce multi-million-dollar TV shows; they never could.  The license fees paid by the MVPDs to the producers are what pay for production, and those licenses are predicated on a complex variety of ways the MVPD expects to exploit its limited or exclusive access to the content.

The simplest and most obvious example is advertising. If, under the FCC proposal, the MVPD that has licensed programming is forced to deliver that programming free of charge to Google, which may then re-distribute the content however it wants and then advertise against the programming from its own ad services, the MVPD’s ad revenue will go down almost immediately.  So, when a new slate of shows is produced, the MVPD’s incentive to pay current market-value license fees is diminished while Google, which captured part of the ad market, isn’t paying anything at all.  Secondarily, new-market distribution channels like Hulu would see no incentive to license programming under such a regime, which gives lie to the notion that this entire proposal is about competition to benefit consumers rather than what it is, which is a government giveaway to Google.

I never quite understand why it should be hard to recognize that less is less—that if license fees for programs go down or if new channels for licensing are cut off, there can be no result other than less production or production of lesser quality. And the FCC proposal would appear to create exactly these conditions—possibly more quickly than people think.  The producers and MVPDs aren’t blind.  If the FCC proposal were to pass, they’ll revise their business strategies immediately, and that could include producing a lot less work within just a couple of years.  At a time when we’re clearly seeing a Golden Age of the small screen in quality writing and production—and in flexibility of viewing options—it is unfathomable that the FCC would advocate unraveling the licensing regimes that have made all this bounty possible.

What’s in it for Google?

I know I’ve repeated Google in this post despite the fact that there are other manufacturers hoping to sell boxes under the FCC proposal.  But if the value of getting into this line of business is predicated on advertising and data mining—which it has to be—it’s pretty hard to imagine that Google would not very quickly dominate this space and become the only game in town.  I understand Radio Shack, for instance, plans to make boxes, but as that company doesn’t have an online ad business or a data mining business, their boxes would presumably serve Google’s pipelines for a piece of the revenue.  If that’s how this would shake out, the “competition in boxes” story is pure illusion.

If the producers, the MVPDs, and the unions are all correct that this plan can only undermine the means of production and inevitably reduce production quantity, quality, or both, what does even Google—let alone the rest of us—gain in the long-run?  When variety of quality content is reduced, then advertising value is reduced and so is data mining value.  Google has a long track record of earning revenue by exploiting works it hasn’t licensed; but in this case, its parasitical model can actually limit the means by which the company typically generates revenue.  So, what’s the long-term plan here?  It’s hard to say. But it’s not hard to see in the short term how this proposal is bad for creators in the film industry and bad for consumers who want to see great television continue to thrive.