TechInformed
With all the hype these past few months you’d be forgiven for thinking that AI is on the verge of taking over Hollywood, but execs, directors and scriptwriters needn’t worry about their day jobs just yet.
article here
“Don’t drink the Kool-Aid,” says Yves Bergquist, Director of AI & Neuroscience in Media at studio-funded think tank ETC, who has written a paper, “Generative AI in Media: What it Means For The Industry,” on AI’s creative and business implications. “The technology is still experimental, highly dependent on large and clean training data (which is expensive to collect and maintain) and ungodly amounts of computation.”
In particular, generative AI is currently very bad at producing anything like a decent story.
“ChatGPT is fresh and still being iterated, but the capabilities displayed so far aren’t a threat to anybody in the media industry. The AI apps that would create (or even optimize) deep narrative content haven’t been invented yet.”
That said, the tech is progressing very fast, and Bergquist thinks it only a matter of time until applications and use cases emerge that will transform the media industry.
In which case, “media executives need to be educated about what AI is and isn’t. Technical executives need to learn to talk about AI to non-technical executives. New skills are needed, such as managing the data-business interface (and the trust issues around it).”
The document is meant to help media professionals think more clearly about AI and its implications for the industry, especially around what skills will be needed in an AI-augmented future.
Good AI Needs Humans
The paper puts forth that the technique that has unlocked the AI genie in tools like ChatGPT is in fact a clever fusion with human moderators. OpenAI, as Bergquist explains in an article in the ETC etcentric blog, used a method called Reinforcement Learning through Human Feedback.
This is essentially a Neural-Symbolic process of continual tweaks and fine-tuning by human teams to train and re-train the AI model.
“When it’s good (often) the machine-output text is indistinguishable from human output, which is probably why the education industry is (justifiably) freaking out,” says Bergquist. “When it’s bad (also often), it betrays inescapable yet fundamental flaws in OpenAI’s methodology: it is hyperscale parroting, not intelligence per se. But it works, which is why it’s garnering so much attention.”
What AI is Good For
How does this translate into practical opportunities or threats to the media industry?
Because ChatGPT is very skilled at outputting basic, low-narrative content, such as an email, website copy, or a press release, ETC thinks the main opportunity for it right now is SEO, “where writing a lot of copy (blogs, websites, etc.) is key to create a lot of links and jack up search engine rankings. Another use is creating bot content on social media on a massive scale. In the media industry, ChatGPT can be used in combination with other generative AI systems to more quickly create previs assets like storyboards and schematics.”
Creative iteration could get a lot easier and faster. And that could be a new way to work. Generative AI models will be powerful (but still pretty dumb) assistants for creatives. Putting a look book together will become a lot easier and faster. Some parts of writing, such as log lines, will become semi-automated.
Marketing functions pivoting on search engine optimization or social media campaigns will need to adjust how they measure success. ETC says they will need to beef up their ability to detect bots and fake accounts.
A little further ahead, opportunities abound, from automating micro-workflows within editing, post-production and CGI to letting machine learning models loose on creating entire scenes from text or image prompts, or automating large chunks of the CGI workflow.
The technology also opens up new creative avenues: voice synthesis can replicate any voice, facial replacement technology can make face blurring in documentaries a thing of the past. Music synthesis can create a score in mere hours. De-aging opens up new narrative possibilities.
What AI Can’t Do
However, ETC also provides a reality check. For while all generative AI tools can automate some of the grunt work process out of the system, its creative capabilities leave a lot to be desired.
“ChatGPT doesn’t comprehend the concepts or symbols present in its output. It can’t abstract or reason in very basic ways. It doesn’t understand anything about the world it’s in, especially causality,” says Bergquist.
“ChatGPT doesn’t have any ability to even understand the basic tenets of a story, let alone build one. Its output is too basic to be even interesting,” he continues.
“ChatGPT cannot perform highly in domains that it has not explicitly been trained and tuned on. It cannot write a novel, or a script, or any kind of text that requires an understanding of the world and its nuances. It’s a complex and highly tuned statistical engine that doesn’t actually ‘know’ or ‘understand’ anything. It can’t write engaging or even slightly narrative text beyond a few paragraphs. It can only have ‘good ideas’ by accident.”
While synthetic image generation is “staggeringly good, and can produce images that could be true works of art,” Bergquist thinks audiences will find its output meaningless as entertainment.
“The point of art is equally the art itself and the artist’s persona and brand. What makes art valuable is the rebellion and talent of a human reaching into deep and previously unknown areas of the human condition.”
In sum, a handful of creators will leverage generative AI models to take their workflows to the next level. As ETC puts it, those creators who can leverage ChatGPT into their workflow to laser-focus their time on the core craft of exploring higher-level human narratives will win the era of augmented content creation.
Studios Need AI Talent
Other winners will be studios, which own their distribution platforms and detailed audience data, and which can use machine learning to understand which content attributes resonate with various audience segments.
“Hollywood will continue to rule,” Bergquist insists. “In an ocean of content, the value lies with curation and personalization. Time being the world’s most precious commodity, exhausted digital denizens will pay a high premium for a service that can deliver them the exact content that they need or that inspires them. And nobody is better at sorting this signal from the noise than the giant talent-filtering algorithm called Hollywood, which also knows something Silicon Valley keeps ignoring: people actually hate technology.”
To this end, studios and content producers are advised to partner with startups to leverage their own data to increase their competitive advantage in the most valuable area of the generative AI.
ETC also suggests that studios should be hiring talent with AI smarts. These include AI engineers with math and computer science skills who can build products, as well as AI engineers with a business (or product development) background who are the interface between the data science function and the C-Suite.
“They can effectively break down business problems in terms of data (for the data science team), as well as communicate to business stakeholders about AI (for the C Suite). We call them ‘hackers’ because they’re scrappy, computational thinkers obsessed with solving business problems with data and computation.”
Senior technical executives with a deep understanding of AI are also important to guide on how budget, plan and staff for it, as well as how to talk about it and which ethical considerations need to be present.
In addition, there will also need to be a greater focus on digital watermarking to ensure content authenticity, ETC finds.
No comments:
Post a Comment