The use of Artificial Intelligence
It's hard to ignore the prevalence of Artificial Intelligence in recent years, which has caused a stir in many industries, including the audiovisual and game sectors. The Flemish Audiovisual Fund (VAF) has been collecting data since 1 October 2023 by including questionnaires about the use of AI in all requests for support of the creation of audiovisual projects. In July 2024 we started analysing the results and data provided.
Methods
This study was carried out in cooperation with Howest DAE Research, in addition to consulting legal experts on intellectual property. Together we were able to define 10 comprehensible categories of AI applications and develop a survey to be completed in 3 different stages.
- Stage 1: Initial support request.
- Stage 2: Request for first payment.
- Stage 3: Request for last payment, in other words, completion of the request.
The questionnaire consisted of three major categories:
- The use of AI: what, why, and when?
- Legal aspects: about legislation and intellectual property
- Opinion: questions we would like your personal opinion on
In total, we received 540 stage 1 questionnaires, 226 stage 2 questionnaires, and 221 stage 3 questionnaires. The survey was completed 987 times.
Most recent update: 07/01/2024
10 AI-categories
- 'Pictures and images: Generative Art'
e.g. generating images through the use of AI models or tools like Dall-E, Shutterstock AI, Canva AI, Midjourney en Stable Diffusion, etc., or editing of images with AI, e.g. with Photoshop. - 'Video / VFX'
e.g. generating videos, editing and splicing of video, converting text to video, (de-)aging and VFX pipelines. - '3D'
e.g. generating models, 3D images, replicating real-life objects with NeRF/Photogrammetry. - 'Virtual Humans'
e.g. generating avatars (facial and/or full-body) - 'Animations'
e.g. motion capture and motion synthesis (articulating 3D models)
- 'Text'
e.g. generating (ChatGPT,...) and translating text (Deepl,...), automating subtitles, interpreting and summarising information, writing assistance, or learning different writing styles - 'Audio'
e.g. generating speech, dubbing, converting text to audio, speech recognition, adjustments and modifications, and generating music. - 'Testing'
e.g. automating simulations, machine learning bots and AI for in-game opponents. - 'Physics / Engine'
e.g. neural physics and engine optimisations, for example, game development or simulation of film sets and camera positioning. - 'Other'
Please select this option if your use of AI doesn't fit in any of the previous categories.
Additional Remarks
- The questionnaire needed to be completed for every type of support for the creation of audiovisual projects. If 1 project requested multiple types of support (e.g. development and production) the questionnaire was completed multiple times for 1 project. This means that there's an overlap of 6.5% of the questionnaires, which were nonetheless included in our dataset because opinions and approaches could change over time, even during a project.
- The questionnaire for the first stage was deliberately shortened to not impact the barrier to entry.
- This survey allowed us to study general trends, however, it's not always possible to zoom in on specific subsectors. For example, we received 10 stage 3 questionnaires for animation support, however only 1 of those was for a feature-length film.
- This report is a snapshot. Stage 3 questionnaires represent completed requests from the past 5 years, while questionnaires from Stage 1 represent future projects and are an indication for the next 5 years. This survey will ideally continue for 5 years to provide a comprehensive view.
The use of artificial intelligence
The reported use of artificial intelligence until July 2024 is pretty limited. 22.2% of respondents in stage 3 indicated that AI was used in the project, meanwhile, 28% of respondents in stage 1 questionnaires indicated that they were planning on using AI. The largest category of AI tools used, by a considerable margin across all stages, was text generation, such as ChatGPT. For both stages, this was around 16-17% of respondents. Tools for Pictures & Images: Generative Art was the second most used in stages 1 & 2. This differed in stage 3 where tools for audio were the second most used. However, this data may be skewed because audio projects, mainly podcasts, have a shorter lead time and are therefore more prevalent in stage 3 questionnaires.
Important: These results are for projects across all categories (fiction, animation, documentary, FilmLab, games, and podcasts).
Observations in Stage 1
While analysing questionnaires from stage 1 (request submissions) we noticed that applicants for series and podcasts are more likely to use AI than applicants for films. There's no clear explanation for this. Studies from France indicate similar results. On average series tend to be more commercial as opposed to films which are more artistic in focus, this may be the reason why innovative tools are more prominent in series.
We can also see that applicants for screenplay support frequently think about using tools for VFX, video, and art. These cases often relate to film d'auteur, where the writer and director are the same person. Sometimes a producer is involved from the start as well.
Respondents who indicated that they won't use AI, find the tools to be less innovative and predominantly serve to complete work quickly and at a cheaper cost, which is opposed to the respondents who indicated they would use AI. The answers from these respondents were more nuanced. They argue that because the use of AI requires extra investment in both time and cost, it does not necessarily lead to reduced costs or work completed quickly.
Observations in Stage 3
107 questionnaires of the nearly 1000 received were stage 3 questionnaires for production support for films and series. In other words: recently completed films and series (projects submitted between 2018 and 2022). Of these 107 questionnaires 22 (20.6%) indicated to have used AI in their project. These questionnaires are more detailed than those in stage 1, which allows us to learn more about the different applications.
The majority, 17 out of 22 (15.9% of respondents or 77.2% of all AI users), used text tools like ChatGPT. Art and Video were each used 4 times.
Of the 22 respondents, 11 used AI to save time, 10 indicated that AI improved the quality of their project and 8 respondents used AI to cut costs. Finally, 5 respondents indicated that AI was used to achieve something that would have been impossible without AI, either not physically possible or due to the time it would take to achieve the same result without AI.
The majority, 15 out of 22, used AI tools primarily to write (accompanying) texts. 9 out of 22 also used AI in the production or post-production stage of a project. A minority, 2 out of 22, used AI during the concept stage, it should be noted that the concept stages of these projects are from before the rise in popularity of AI. We expect that this number will increase in the following years.
In response to the question if there is AI-generated content in the final product, none of the respondents answered yes.
Which tools were used?
Why did they use AI?
When was AI used?
The grey area
The use of artificial intelligence raises a lot of questions about ethics. Some view it as the Wild West, where a lack of legislation allows for a lot of opportunities to experiment, but it can also be risky and even harmful. We currently find ourselves in a legal gray zone.
Let us take ChatGPT from OpenAI as an example. Their terms of use indicate that you remain the sole proprietor of your input and become the owner of the output that ChatGPT generates. These same terms of use however indicate that OpenAI can use all content (including prompts and datasets that you input) to "provide, maintain, develop, and improve our Services". You can disable this once logged in, however you may lose access to certain features.
Of the 22 respondents who used AI in their project, 18 indicated that they made use of free tools. We hope that these people realise that free tools are never truly free and we advise them to verify the privacy settings and terms of use of these tools. 8 respondents indicated that they used paid tools and 8 indicated that they used outside experts for their AI needs. A tool that requires subscriptions or one-time payment does not mean that your data is managed ethically. We always recommend reading the terms of use and any documentation provided to make sure you understand how companies will use your data.
18 out of 22 respondents who used AI in their project indicated that they made use of free tools. We hope that these people realise that free tools are never truly free and and we advise them to verify the privacy settings and terms of use of these tools.
It's concerning that of the 22 respondents, 18 indicated that they don't know if the AI tool they utilised, made use of a intellectual property right. The remaining 4 respondents indicated that no protected works were used.
A deep dive
Considering the limited amount of surveys, it wasn't always easy to find trends in certain categories. Nonetheless, we attempted to provide a few insights at the request of the industry but would like to note that these results are not representative and merely illustrate certain trends based on the received data.
Animation
Of the 107 surveys in stage 3, we received 10 questionnaires for animation projects (short films, feature films, and series). 3 of which used AI, specifically for text. Other categories did not come up. AI was used to save time and money, mainly in the concept stage and while writing documents. All three used free tools. Two of which also used paid tools and one used an external party. The applications of these projects date back to 2022 or earlier when AI was not yet commonly used.
If we take a look at future projects, we received 31 questionnaires in stage 1, of which 6 used AI. The majority (5) will use it for text purposes and 3 will use it for artwork. The categories 3D and animation both appeared once. Reasons why they plan on using AI were not asked during this stage.
This indicates that there is a growing interest in using AI for animation and 3D. However, there's no alarming trend of AI taking over the industry, considering the percentage of AI-users is lower than in stage 3.
Podcasts
Podcasts have only started to receive grants in 2022. Additionally, podcasts on average have a shorter lead time, which means that every project in stage 3 had the opportunity to use AI in their project. We received 15 completed questionnaires, 3 of which indicated that they used AI for Text and Audio in (post)production.
What stood out is that 1 project contained AI-generated content in the end product, although it's not clear what this exactly means. Everyone who used AI did it to save time, and 2 also indicated they wanted to reduce costs. All 3 AI users opted for free and paid services, and 2 of them also hired an external party.
In stage 1 we received 39 questionnaires of which 14 indicated they were planning on using AI in their project. The most common categories were Text (8) and Audio (6).
Games
Game studios have been using AI for quite some time. Motion capturing for example is currently being trained by AI to ensure characters move naturally and AI is often a part of the game engine, a tool used to implement video games without building everything from the ground up.
This is reflected in the questionnaires. For prototype support, we received 23 questionnaires for stage 3, of which 5 used AI. There are no outliers: the categories Art, Animation, Text, Testing, and Other all had 1 or 2 users. In production-related questionnaires (e.g. Vertical Slice and Production) we received 9 questionnaires, and AI was used in 6 of them, with a similar distribution in categories. Stage 1, so this means the future projects (see the graph below), had similar trends and categories.
Everyone who used AI in stage 3 (11) used AI to save time. Additionally, 5 opted for AI to reduce costs, 4 to improve quality and 1 to realise the impossible.
3 out of 11 respondents used AI during the concept phase, 4 to write documents, 7 while making a prototype, and 2 during production. Almost everyone (10) used free tools while less than half (4) used paid tools and 1 hired an external party.
Other reasons for AI use
Following the clean-up of the data, where certain misinterpretations were verified and recategorised, a limited number of respondents indicated that they had other reasons to use AI, besides saving costs and/or time, improving the quality, or realising the impossible. Their reasons are separated into 2 large categories.
Some had creative reasons for using AI, because it was part of their art process or because they wanted to train their own AI model, for example to add handwriting recognition in their game. These are projects centered around AI.
On the other hand, there were projects where AI was the subject. Fiction projects and documentaries about artificial intelligence, where the creators test tools to understand how they work and use these tools as inspiration.
Both categories are not common but it's striking that people use AI as inspiration.
Unconscious bias and AI
Several universities and organisations have researched to ascertain unconscious bias when using AI. An AI model is trained on a specific dataset and as a result, the AI reflects the average of the data in that particular set. If the dataset is not representative and varied, the results of the AI won't be either.
The most famous AI tools rarely or never reveal what data was used to train their tools. This means that we can't determine if and how the results are biased. We've all heard stories of chatbots sending discriminating messages because the dataset unintentionally contained discriminatory and hateful social media posts or image generators that primarily depict white men in senior positions.
Stereotyping and unconscious bias are challenges that AI providers wrestle with daily. However many respondents in our survey are not fully aware of this. Regarding the statement that the use of AI can lead to stereotyping, clichés, and/or unconscious bias, 33% of respondents gave a neutral to negative response, which is worrying at the very least.
Additionally, we see a clear divide between respondents who used AI and those who didn't. Respondents who use AI tend to agree with the above statement. Only 27% of them gave a neutral to negative response. Respondents who don't use AI are less informed and aware of the downsides. It remains important to invest in training to make everyone aware of the pros and cons of using artificial intelligence.
It remains important to invest in training to make everyone aware of the pros and cons of using artificial intelligence.
AI in the end product
There's a common conception that artificial intelligence will replace creative minds and that people will be ousted from audiovisual media and games in favor of AI. This led us to question how AI is used, and specifically if artificial intelligence was used in the end product.
We received numerous reactions in addition to figures concerning this. There are numerous interpretations of the definition of AI and end product. For example, if AI was used to create visual effects, this could be seen as part of post-production or as an end product, depending on who completed the questionnaire. A writer, completing the questionnaire for their screenplay, could see a translated sentence by AI as part of the end product or as research. This means that the data in this part of the survey is not conclusive, however, it does provide us with some insight.
9% of respondents who answered this question (n=143) indicated that AI was used in the end product. 4 of them were in documentaries, the majority of which used some form of AI to write their screenplay. For games (4 projects) AI was used in prototypes to quickly generate artwork and textures.
It's too early to conclude. While the signs seem relatively positive, some parts of the industry could undergo severe changes with technological advancements. In other words, we need to keep a watchful eye and stay vigilant. Qualitative studies and discussions with the industry are paramount to better understanding these results.
The use of intellectual property
We asked whether employees, freelancers, and interns can make arrangements with their employers about the use of their work in (the training of) artificial intelligence.
A large portion of respondents in organisations where AI is not used indicate that it's possible to make these kinds of arrangements, this is in stark contrast with organisations where AI is used. Important note: the majority of these arrangements were agreements where the rights of the work would be transferred to the organisation and employees had no further claims. This means that employees would lose all their rights, which is a restrictive measurement rather than a supportive one.
Some indicated that it was possible to refuse that the work of employees would be used to train AI through a clause in the contract. It's however unclear if this is discussed in advance and if it is factored into contract negotiations.
However, these results should be taken with a grain of salt. We can see a clear lack of knowledge of the legislation and not every respondent is adequately informed to answer these specific questions. This was also established during early conversations with respondents and made it hard to interpret the data.
Additionally, emotions could play a part. A personal bias against AI could cause an employer to be protective of the use of an employee's intellectual property. This does not mean that a bias in favor of AI would lead an employer to intentionally infringe on an employee's intellectual property, however, this may happen unwittingly. In both cases, a lack of knowledge and unconscious bias (positive and negative) could skew the results.
A qualitative study and discussions with stakeholders should provide more clarity.
When should we allow the use of AI?
We asked respondents which reasons justify the use of artificial intelligence. We encountered a multitude of reasons, which we summarised in 5 broad categories.
Administration & support
Automatisation can drastically improve efficienty.
Prototyping & pitches
As long as you remain at the creative helm.
Translation & Transcriptions
When speed is essential.
Inspiration and Research
Researching the appropriate style and direction.
Never!
15% of respondents didn't see any advantage of using AI for their projects.
When should AI never be used?
Similar to the reasons that justify the use of artificial intelligence, we wanted to know when it should never be used. These were separated into 3 large categories.
Concerns about IP and Privacy
Fear of AI being used to infringe on IP rights, and that personal characteristics like voices and images could be copied.
Replacing people with AI
Fear of AI being used to replace writers, artists and designers.
Ethical reasons
Many see a doomsday scenario where AI is used for unethical purposes, e.g. manipulation, misdirection or malicious intent.
Despite a consensus that artificial intelligence shouldn't be used to replace people in creative professions, there's less consensus about exactly which jobs. The notion that AI shouldn't be used to replace writers, artists, and designers, but could be used for tasks like transcription and translations, primarily where speed is of the essence, could be construed as a "not in my backyard" mentality.
Conclusions
- Artificial intelligence is mainly used to assist with administrative and technical tasks. This primarily involves the use of text tools like ChatGPT.
- Hardly anyone used AI in the end product, but there's no concrete definition of what it means to use AI in the end product. This leaves room for interpretation.
- There's a clear lack of knowledge about artificial intelligence and what it should and shouldn't be used for.
- There's a clear lack of knowledge about legislation of intellectual property, privacy, and copyright.
- These results indicate a clear trend, however, more data is needed to gauge actual use. A qualitative study is essential.
Remarks
VAF would like to express its gratitude to Glenn van Waesberghe (Howest DAE Research) for assisting us during this project.
Research by the CNC in France produces similar results. In 2024 the adoption of AI is not yet substantial but it's important to keep an eye on further evolutions.
The data from our report comes from a mandatory questionnaire about Artificial Intelligence, launched in October 2023 and was part of the deliverables of every grant application in the submission, contract, and closing stage within the domain Creation. We would like to emphasise that these questionnaires had no impact on the approval of support or any further progress of the application. Nor did the answers lead to any kind of judgement by VAF about the use of AI.
While compiling the results, any identifying factors were separated from the relevant data. Where identification could have been possible due to a low response rate, no results were published or communicated.
Individual results (regardless of identifying factors) will never be shared with third parties (including commissions and the Board).
Who can I contact with questions about AI?
For more information about our methodology or general questions about AI, you can contact VAF's Knowledge Center.
