The Content Turing Test
Answering the question “Can AI be used to make a real movie?”
Applying a Turing Test to Video Content
The “Content Turing Test” puts someones name on some form of published media content - like a movie or novel - that was thought up and written mostly by AI software. The audience thinks a human made what they're seeing, to see if they enjoy watching and would pay for more. Try it for yourself on friends using some of the AI-written short films below and judge for yourself. Do you conclude that "yes", audiences would buy movie tickets and invest their evenings watching more of these types of short-films? Definitely not.
But could an AI-written screenplay win the Academy Award for Best Picture in the near future? Many think it’s not only possible, but inevitable.
In my last post I made the claim that Artificial Intelligence (AI) has?not?to date successfully created a feature film worthy of theatrical release (read?How AI Can Benefit Hollywood). In this post I go into further detail and explain how I devised an answer to this question, and how I plan to judge future attempts (including my own). But what is the definition of Cinema, and how do we classify videos as Films? One way is to apply a variation of the Turing Test but for media content an AI created and intended to pass for its best work. The term "Turing Test" has been around for decades, and is constantly being used in new forms as AI gains new skills (e.g. AI-generated images as works of art). There have also been similar terms coined for slightly different definitions like "Lovelace Test" [1]. I would posit my definition below is slightly unique and specifically defined for filmmaking.
The definition of an AI entity has been sliding for years when humans judge if something “is really Artificial?Intelligence?” or not. People's conclusions change as AI succeeds at new tasks. This is known as the “AI Effect”, that anything?only?humans can do requires intelligence, until a computer can do it — at which point it’s simply considered software [2]. Experts half a century ago claimed it would take?true?AI to be able to play chess, but when such a machine was invented it was no longer considered AI — simply a “chess-playing machine” merely built on?dumb algorithms?and?brute force?computation. This sliding test continued as software accomplished more tasks considered under the domain of humanity like language translation, efficient travel directions, image or facial recognition, and self-driving cars.
Now we see this sliding test applied to software-created?artwork, from images to songs and yes, even the movies. Debates rage online as to whether machines are able to produce true “art”, and some argue this can only be a human skill. Yet in recent years we’ve seen examples of people using AI to make parts of films like scenes and dialog, with ground-breaking efforts such as “Sunspring” and more [3]. However, a discerning audience (and likely even the film-makers themselves) would admit these can’t be considered a?real?movie?yet. I would wager that if screened to a Film Studies professor as works of their students, these films would receive a failing grade. While every new movie is unique, history shows that epic stories tend to follow recognizable patterns and structure. This format evolved and perfected over centuries and is what we teach aspiring Directors and Screenwriters as they hone their craft.
These examples of AI-written screenplays instead tend to ramble insensibly with no clear direction or lesson about humanity. At the very least, we can assume they would not be competitive with the latest superhero movies at the box office. There is no conclusive?Act and Scene structure. The protagonist follows no discernible?Character Arc, exhibiting no?Want, Need, “Ghost”, or “Lie”. These aren’t true Stories but simply a filmed collection of words read by actors on sets. So when does content cross the line and become a real?film?
The?Turing Test?is a simple way to answer the question: “Is this machine able to think as intelligently as humans?” [4].?Alex Garland’s?2014 blockbuster from?A24 Films?Ex Machina?was based on this concept. A two-sided test hides the fact that the subject may be a machine pretending to be human. The two enter into a conversation. At the end, if the questioner is convinced they were talking to another human, it proves the machine is a true?Artificial Intelligence.
Imagine extending this concept, but instead of testing a?conversation?(proving an ability to?think),?we present?artistic content?and see if it passes for a human work as any other.?Put another way, we show people a movie without telling them an AI made it, and see if it’s a hit. Authors have used pseudonym “pen names” for generations, including Stephen King, Isaac Asimov, J.K. Rowling, Benjamin Franklin, Mark Twain, and Dr. Seuss. So have actors, comedians, and directors. This could be for a variety of reasons, including privacy and the desire to overcome unconscious biases of gender, race, or religion. Because it’s subjective and inconclusive to simply ask an audience “Was that a good movie?” (subject to Agreement Bias [5]), we instead apply the same standards as Hollywood:
If you tell someone “an AI made this film” there will be bias coming from both sides. Some people may be so impressed a machine could accomplish this feat and lower their standards for quality entertainment. Some may even?purchase?AI-created content, but I posit today this is for the gimmick and not solely quality of work, similar to people buying paintings made by animals [6]. Conversely, many people (usually older veterans of the entertainment industry) who are against the very?concept?of Creative AI, may decide consciously or subconsciously that they hate what they are watching, knowing in advance it was not human-made.
Here’s the fun part: there’s a good chance someone has already done this! There could be a show or movie on streaming someone created using AI and just said they made it. For all we know, it’s one of the biggest hits of the year. Do you think you could tell if there was? In the end for me it doesn’t matter?who?(or what) makes a film — as long as it’s great. If I could watch a movie I absolutely love each and every night of the week, I would. There’s no upper limit for new great content each year. For now we’ll just have to accept rare masterpieces, though I can’t help but imagine…
What unprecedented epic stories could tomorrow’s AI-assisted creatives produce for us to enjoy?
References
Cover image from: TIME magazine (time.com) Blog article: https://time.com/3609585/the-true-story-of-the-imitation-game/
[1] Self-described as related to AI and Creativity as new term: "Lovelace 2.0 Test" - published 2014 by Mark O. Riedl see https://arxiv.org/abs/1410.6142
[2] What counts as artificially intelligent? (The Verge - James Vincent)
领英推荐
[3] Short Films made with AI include:
Sunspring?— End Cue Production (Oscar Sharp, Ross Goodwin)
More examples and stories made by AI and published recently:
[4] The term "Turing Test" was used later, as defined originally by Turing in the 50s as "The Imitation Game":
[5] “Agreement” Bias:
[6] Other smart species sell artworks: ElephantArtOnline.com - elephantartonline.com
????
Copyright ? 2022 CYBERFILM.AI CORPORATION
Engagement Lead @ inVerita | Bachelor's in Accounting
2 年Russell, thanks for sharing!