OpenAI’s ChatGPT chatbot has had a huge impact since it went public in November and made many realise the potential of AI to change the industry at all levels.
At VO we have, of course, been using AI throughout our product range for several years now. We were one of the pioneers of deploying AI in the video management technology sphere, and it is at the core of much of our business and the way we are developing our solutions. We use it to analyse data to help provide improved recommendations for TV viewers, we use it to segment audiences for targeted advertising, we use it to monitor video piracy activities, and along with others we are working on ways of using it to improve video quality at ever increasing levels of compression.
We talked in detail about all this just before IBC2022 in a piece titled The increasing importance of AI in broadcast. Most of that article still holds true, but matters in AI have moved forward with huge speed since.
Much of that is down to the release of ChatGPT by US research laboratory OpenAI. Just in case you need to be brought up to speed quickly, Open AI had already made serious waves online with the release of DALL-E 2 earlier in the year. DALL-E 2 is an AI-powered image generator that uses natural language prompts to create images; a user simply types in a sentence and an image is generated.
For example, type in “two cats discuss the benefits of AI in a garden full of flowers,” (CatGPT, as it were) and this is what you get.
ChatGPT does the same thing but for text. Here is what results from exactly the same text string input into ChatGPT:
And it is this that has really set the cats amongst the pigeons, as it were. Text underpins the modern internet in a way that images do not, and being able to generate text at scale and at speed fundamentally alters many things from the way search engines, to the ways that SEO and page ranking are calculated. It is no wonder that Microsoft has invested $10 billion in OpenAI to help it overhaul and improve its Bing search engine.
There is a lot to unpack here and it has to be said not all of it is positive, with the technology opening up a whole range of ethical concerns. But we’re going to narrow our field of focus to the video industry. What can ChatGPT and other similar AI chatbots do for you? The answer is quite a lot.
ChatGPT in the content path
Let’s follow the path content takes from creation to distribution to see where AI processes such as ChatGPT may have an impact. AI is already used in quite a few different areas here.
Production and post production
Increasingly sophisticated routines are being used in the editing process for speech to text and metadata generation. This gives a rapidly searchable index of a programme’s rushes, allowing editors to quickly zoom in on different takes via searching for key moments of dialogue or specific actions such as ‘John opens front door’.
The result is a far more efficient editing process. The same technology is also being increasingly used for automatic caption generation, though work is still to be done before it can reach the accuracy of human operators.
AI is also increasingly been used in live work, automatically following tennis players on remote courts during big tennis tournaments, for example, meaning that more of the more expensive human camera operators can be deployed on the bigger matches.
The obvious area where ChatGPT can have an impact is script-writing. No one is expecting it to generate Emmy-award winning dialogue, at least not this generation of machines (see our cat dialogue above). But it can give writers a tool that they can lean into to quickly generate a lot of content.
As well as scriptwriting and generating ideas and writing scripts for TV shows, commercials and so on, at VO we are currently exploring how ChatGPT can assist with various tasks related to TV content creation, such as:
- Character development: creating and fleshing out characters for TV shows and other forms of media
- Storyboarding: Outlining a visual plan for a TV show or video
- Dialogue generation: writing dialogue for TV shows, movies, etc
- Content optimization: suggesting changes to make a TV show or video more engaging and appealing to a target audience
Marketing and distribution
Marketing and distributing film and television takes a lot of content. Descriptions have to be written, synopsis have to be written for the series as a whole and for individual episodes, press releases, TV listings, social media… the list goes on. A lot of the initial work here can be offloaded to AI chatbots, at least for the initial stages. Indeed, it is quite easy to see a multi-tier strategy emerge in the industry here: ChatGPT used for b-list or long-tail content, with human creativity more being brought to bear on the higher profile material.
That implies a value judgement on the quality of the AI-generated content, and with the current generation of tools that is probably fair enough. Large language models such as ChatGPT are trained on content scraped from the internet, and are limited in their creativity. Many times their content can be detected by human readers as AI-generated and is already subject to bans and restrictions in several areas such as education and conferences. It will get better, but it is not there yet.
(And for those worried about the provenance of content, Google says it has a tool that can detect whether it has been generated by AI. Meanwhile, the Chinese government is insisting on the watermarking of all AI-generated content to indicate where it came from.)
In our Anti-Piracy Center, we also investigate how to differentiate between human-generated TV content and machine-generated TV content. There are various methods for doing so, including analyzing the language patterns, tone, and writing style of the content. However, it is becoming increasingly difficult to distinguish between the two as machine-generated content continues to improve and become more sophisticated.
It's important to note that as an AI language model, all the content generated by ChatGPT is produced by a machine and not by a human.
Engagement
This is possibly where ChatGPT and its ilk are likely to be most impactful in the video industry in the immediate future. The use of chatbots in online customer services is growing all the time where they are used to answer low level queries from customers and acting as a triage service, dealing with the mundane questions and flagging things up for human operatives to take over when necessary. ChatGPT increases the sophistication of these models, raising the intervention level considerably.
We are looking at how we can use it to improve recommendations generated by TV analytics. By analyzing a user's viewing history and preferences, ChatGPT can generate personalized recommendations for TV shows and movies that are likely to be of interest to that individual. It can take into consideration content analysis of TV shows and movies and extract information about the content such as genre, themes, and style, which can then be used to improve the accuracy of recommendation systems.
That is just the start too. It can also be integrated into the user experience to provide a more natural language interface for recommendations, especially when linked with Voice. ‘Hey ChatGPT, I want to watch something that will lift my mood?’ could easily elicit a range of responses depending on user-defined settings from a simple content recommendation to solicitous enquiries about the user’s health and well-being, and more. It is not necessarily changing the fundamental way that the recommendation engine works, but the new degree of sophistication it allows brings a perceived increase in capability.
This paves the way for genuine, personalised virtual assistants that are hugely more capable than the current Siris and Alexas. Integrating them into the wider home environment may well be a challenge the broadcast and streaming industry has to face sooner than later.
ChatGPT: The next steps
So, what comes next? ChatGPT is currently moving into a monetisation phase. While it has been free to experiment with so far, it is in rolling out a paid tier for fast lane access and support
It will be interesting to see the integration business model it develops too for other developers to hook into its services via APIs while it scales up to meet demand. And, of course, it won’t have everything its own way either. Google has recognised the threat to its own hegemony and ramped up its own AI development program, promising 20 new services will launch in 2023 alone.
Either way, expect to see AI and chatbots start to make an increasing impact on the industry as 2023 progresses.
Finally, as everyone does with articles like this, we asked ChatGPT what it thought itself about the subject (and yes, we genuinely did this only after we’d written everything else above). Here’s what it said:
“ChatGPT is an AI language model developed by OpenAI, it's [sic] ability to generate human-like text based on input patterns can be utilized in various ways in the video industry such as:
- Content creation: ChatGPT can assist in generating script and captions for video content.
- Interactive video experiences: ChatGPT's conversational AI abilities can be integrated with video platforms to create interactive video experiences, such as conversational video games or virtual assistants.
- Personalization: ChatGPT can be used to create personalized video experiences by recommending videos or generating unique content based on individual user preferences.
“However, its impact would depend on how well and effectively it's integrated and utilized by companies in the video industry.”
Wise words…
