Picture Credit: Shane Rounce, Unsplash.
Google Deepmind Challenge, March 2016. Tension is building as the undefeated “Go” world champion, Lee Sedol, appears to be on the verge of losing against its opponent, Alpha Go. There are no emotions on the winner’s face: Alpha Go is just a machine. But quietly spectating from a distance, computer scientists from the Google Deepmind team are bursting with excitement. Considered as the pinnacle of board games for its inherent complexity and intuitive nature, Go has long been out of reach for modern AI. Today, however, Alpha Go is making history with a handful of unusual, “non-human” moves; moves that appear to be seemingly bad decision-making or outright mistakes. Fast forward a few years after the triumph of the machine and Lee Sedol announces his retirement, saying “AI cannot be defeated.”
We may have mixed feelings about this story, especially since it happened ten years earlier than expected. It sheds light on the rapid growth of AI, but also on the promises of deep learning technology. Beyond the theories of “mankind’s obsolescence” as well as the ethical concerns, this raises a number of questions as to how AI can be used for creative purposes.
Is AI creating?
To better understand the role and potential expansion of AI in the creative realm, AI specialist Arthur Miller suggests that we distinguish AI in two groups: symbolic machines and artificial neural networks.
The first one, evocatively labelled “Good Old-Fashion AI” (GOFAI), runs on encoded rules and “works through a series of logical steps to solve specific problems.” When given the task to produce art, for example, symbolic machines create neat and nice pieces without ever overstepping the boundaries of the code. The famous “Lovelace test” would consider humanlike intelligence (creativity serving as a proxy) the capacity to produce something “new, surprising, and valuable” the machine was not designed to do. As Miller puts it, what we see here is “pre-programmed art”.
The “Lovelace test” would consider humanlike intelligence the capacity to produce something “new, surprising, and valuable” the machine was not designed to do.
Neural networks, on the other hand, are more “experimental and unpredictable”. They “are loosely inspired by the wiring of neurons in the human brain and need far less input.” Deep learning is precisely the type of technology that powered Alpha Go. After being fed basic knowledge, Alpha Go repeatedly played against different versions of itself, learning from its own mistakes and uncovering various tricks and strategies. When putting this technology to the test for art-related purposes—screenwriting or poetry in this case—experts point out that the results often sound nonsensical. But if AI-generated art appears to be eerie, opaque, disjointed, or just “not quite there,” there is a consensus around the fact that it might be a powerful tool to stimulate human’s imagination and free up time for ideation.
…or assisting creativity?
AI has already stepped into creative territory, not as a truly creative entity per se, but because of its supporting role in the industry. In a 2018 paper titled “Creative Disruption,” McKinsey and the World Economic Forum outlined the roles of AI in domains such as content distribution, marketing, journalism, music, and filmmaking. Real-life examples include:
- matching content on Netflix to the user profile
- speedwriting political and sports news (i.e. Heliograf, Cyborg)
- composing and adapting (mostly background) music (i.e. Amper, AIVA, Magenta, Amadeus Code)
- processing sound to achieve studio quality standards (i.e. Landr).
In the fashion industry, AI solutions (i.e. Autogenerate.ai) aim at helping designers keep up with people’s changing habits and a fast-paced marketplace. Further experiments have been conducted in the field of painting to promote artistic crossovers by extending a specific master style (ex. impressionism) to photos and videos “through the style transfer technique,” and the list goes on. But what about the film industry?
AI for filmmaking
From the first AI-created trailer for the horror movie Morgan to the first AI-scripted film Sunspring, the audio-visual industry has sought to leverage the power of machines to refine their content. Aside from its experimental role in editing and screenwriting, AI is more generally used for production and marketing purposes. Here are some of the most prominent applications of AI in the filmmaking industry:
About four years ago, IBM used machine learning (the Watson platform) to screen images, sounds, and composition of a large collection of horror movies. Watson’s subsequent mission was to extract appropriate scenes of the full movie Morgan so that editors could quickly piece them together in a dynamic trailer, speeding up an otherwise lengthy process.
In the case of the short film Sunspring (2016), an AI system digested the scripts of a large number of science fiction movies before it was tasked with creating a screenplay, complete with dialogue and stage direction. Taken together, AI-generated dialogues did not quite create a meaningful unit, but the result was considered strangely entertaining.
Production and Marketing
Artificial intelligence has certainly been more successful when trying its hand at analyzing, assessing, and selecting potentially profitable screenplays. The Belgian company Scriptbook claims to “assist film studios, producers, sales agents, distributors and financiers with their greenlighting decisions.” After gobbling up enough data, the machine starts to detect patterns, then predicts success rates: storyline potential, character likeability, target demographics, and expected satisfaction. Branching out into AI-generated storytelling, Scriptbook has also developed a generative AI hoping that it can become “a worthy co-creator.” With such a tool, we circle back to the above-mentioned application.
If AI is data-driven, it’s only logical that its applications would include box office forecasts and consumer-taste predictions in order to drive content creation and facilitate audience targeting (consider the cases of Vault and Qloo). Predictive analytics can also erupt into a crucial phase of the creative process: talent selection (casting). Case in point: Cynelytic Inc. has partnered up with Warner Bros. to “use [a] AI-driven content and talent valuation system.” Again, it all comes down to optimizing the greenlighting process and guiding upstream creative decisions, something that didn’t fail to raise concerns among creators of all stripes.
Perhaps, the most pressing question is not so much whether AI can be creative, but how it contributes to the creative process and what impact its applications might have for the future of the industry.
An ancient Chinese board game and the longest continually played game in history. Before taking up the Google Deepmind Challenge, Lee Sedol won the world title 18 times in a row.
See Alpha Go , A Documentary [available on Netflix].
A former world champion of the game Go says he’s retiring because AI is so strong: “Even if I become the number 1, there is an entity that cannot be defeated,” Business Insider, 17/11/2019.
Miller, A. , Creativity and AI: the Next Step, Scientific American, 1/10/2019.
Du Sautoy, M. , The Creativity Code. Boston: Harvard University Press.
Miller, A. , Creativity and AI: the Next Step, Scientific American, 1/10/2019.
See Deepmind website and Alpha Go dedicated blog. Online [https://deepmind.com/research/case-studies/alphago-the-story-so-far]
Also see Miller, A. , The Artist in the Machine: the World of AI-Powered Creativity. Boston: MIT Press.
Mc Kinsey and the World Economic Forum , Creative Disruption: the impact of technologies on the creative economies, White Paper.
Introducing the next generation of music makers, The Guardian, 29/08/2017.
IBM, What’s next for AI. Online [https://www.ibm.com/watson/advantage-reports/future-of-artificial-intelligence/ai-creativity.html]
IBM, The quest for AI creativity. Online [https://www.ibm.com/watson/advantage-reports/future-of-artificial-intelligence/ai-creativity.html]
This is what happens when an AI-written screenplay is made into a film, The Guardian, 10/0/2016.
Data driver Cynelytic engages Warner Bros Pictures International to Utilize their Revolutionary AI-driven content and talent valuation system, Business Wire, 08/2020.
Hollywood is quietly using AI to help decide which movies to make, The Verge 28/05/2019. Also see Can AI Predict a Movie’s Success? Algorithmic Screenplay Service ‘Scriptbook’ Causes Major Backlash, No Film School 21/04/2017.