Game development is undergoing a period of rapid change, and one of the driving forces behind this change is artificial intelligence (AI). AI technologies have already found their way into many areas of game development, from enemy AI in action games to AI companions. But now a new revolution is on the horizon: Generative AI (GenAI). At the devcom developer conference, there were many talks about AI. I have summarized the insights from several of these talks for you here.
What is devcom?
#DDC2024. The Devcom Developer Conference 2024 (devcom) is the annual conference for the game developer community that takes place in Cologne. In addition to networking and exchange, there are specialist presentations on various stages.
- One of the speakers is Jeff SkeltonHead of Technology Partnerships at Electronic Arts (EA), who shared his experience of integrating AI into the development process of a large software company.
- Vitalii VashchukHead of Gaming at EPAM Systems, contributed his expertise in the implementation of AI technologies for more efficient game development.
- Kent KeirseyCEO of Invoke AI, presented how tailor-made AI solutions can revolutionize the creative work of artists.
- Judy Ehrentraut, Creative Content Strategist at Red Meat Games, rounded off the topic by talking about the ethical implications of AI in the creative industry.
What is Generative AI and why is it important for the games industry?
Generative Artificial Intelligence (GenAI) has revolutionized the way content is created in recent years. At its core, GenAI is based on machine learning (ML), a technology that enables computers to learn from data and recognize patterns without being explicitly programmed to recognize these patterns.
Old hat? Machine learning since the 1950s
Machine learning itself is not new; its origins date back to the 1950s. However, it is only in the past decade that we have seen major breakthroughs made possible by the availability of big data and the exponential increase in computing power. These advances have enabled us to train increasingly complex models that can deliver astonishingly precise results.
Advertising
Recognize and generate patterns
Generative AI works by creating content based on previously learned patterns and data. A GenAI model is trained with a variety of data, such as texts, images or pieces of music. In this way, the model learns how such content is typically structured and can then generate new content that follows these patterns. This means that GenAI does not simply copy existing data, but actually creates new content based on the principles it has learned.
Acceleration and optimization
The advantages and potential of GenAI in game development are enormous. This technology makes it possible to create content such as characters, landscapes or even dialog in a fraction of the time that traditional methods would require. AI can also automate everyday tasks that previously required manual intervention.
Hi Jeff! Automation in everyday life.
Imagine you want to hold a 30-minute meeting with Jeff tomorrow. You simply give the command by voice command: "Organize a meeting with Jeff for tomorrow. 30 minutes." And the AI translates the speech into text. Based on this, the AI automatically creates a Zoom meeting that is coordinated with both calendars. It has recognized in the room booking tool that you and Jeff will not be in the same place tomorrow.
Everyday assistants. This ability to independently make helpful and sensible decisions and take on routine tasks is just a small part of the potential that AI holds for all office professions - including game development.
Advertising
Automation and increased efficiency through GenAI
Automation. GenAI has the potential to significantly accelerate the creative process in game development by automating repetitive and time-consuming tasks.
In traditional game development, each phase - from ideation to prototyping to final implementation - requires an immense amount of time and manual labor. GenAI can simplify and accelerate many of these steps by providing automated solutions that increase both creativity and efficiency.
How GenAI accelerates the creative process
One of the biggest advantages of GenAI is its ability to complete complex manual tasks in the creative process in the shortest possible time. Where developers would previously have needed weeks or even months to create prototypes and test ideas, GenAI can complete these processes in days or hours.
Goal: creative flow. This allows you to spend more time in "creative flow" - the state in which you are most productive and can develop your best ideas. "The more administrative things we can take away, the more time creative minds spend in flow and doing what they excel at," explained Jeff Skelton during his presentation.
Advertising
Create game world. One example of this acceleration is the automation of iterative design processes. If you want to create a new game landscape, GenAI can generate a hundred variations based on your specifications: 4×4 km in size, two mountains, a lake.
Fast iterations. You can then select the best variant - version 36 - and refine it further: the eastern mountain slightly flatter and the lake significantly larger and with more bays, with an inflow from the northeast and an outflow to the southeast. The AI again generates prototypes that match these specifications. This approach not only saves time, but also opens up new creative possibilities by generating ideas that you might not have considered yourself. Step by step, you can work your way towards a prototype, which you can then continue to fine-tune manually.
The role of coding tools and test automation
In addition to generating content, GenAI also plays a role in the automation of coding and testing processes. Coding tools based on GenAI can help to program faster and more efficiently by suggesting lines of code, identifying errors and even generating entire sections of code automatically. This not only reduces the error rate, but also speeds up the entire development process.
Automate tests. Another important area in which AI helps to increase efficiency is test automation. In game development, testing is an essential but often tedious process. AI can help automate testing by identifying potential sources of error, suggesting solutions and automatically implementing and testing them.
Advertising
Analysis of test videos. For example, an example was presented at Devcom in which human test players formed an "army of testers" together with AI. The video footage of test games generated in this way can then be analyzed with the help of precisely trained AI: Which parts look like mistakes that a human should take a closer look at?
The combination of automated content creation, coding tools and test automation is intended to speed up the entire development process without compromising on quality.
Independence. Where an automated system used to detect why the program was blocked, she wrote an email or ticket to the person responsible for the program module: "There was an error in module XY. Please correct it and restart the program."
Today, thanks to AI, such messages can already look like this:
- "We had a crash at 9:14:33.
- I have identified module XY as the cause.
- I have found the following errors in lines 2352-2366 [error description]
- and corrected as follows [correction description].
- I have tested the corrections in the test environment
- and restarted the program at 9:15:21.
- Since then it has been running flawlessly.
- Please check the changes you have made and release them for live operation."
Focus on creativity. GenAI should make it possible to concentrate on the creative aspects of the work, while the AI takes over most of the technical and repetitive tasks. This leaves more time for what really matters: creating innovative and exciting game worlds.
Tailor-made creativity: AI as a tool for artists
Generative AI has the potential to fundamentally change the way artists work by providing specialized tools that support and enhance creative work.
Away from the mainstream. A central aspect of this is the development and use of specialized AI models that are tailored to the individual needs and unique style of an artist. These customized models make it possible to make creative processes more efficient and precise without losing artistic control.
The importance of specialized AI models
While generic AI models can already deliver impressive results in many mainstream areas, they often reach their limits when it comes to accurately reproducing or further developing an artist's specific style.
Advertising
One AI per artist. Specialized AI models are being developed to overcome precisely this challenge. They are trained based on an artist's individual style and aesthetic preferences so that the content generated by the AI reflects exactly what the artist intended.
Kent Keirsey emphasized in his presentation at devcom how important it is for artists to retain control over their creative processes: "We train the models in the artist's style. The basic model is open source, but the training results are based on the artist's specific style or characters. They then overwrite the open source model. This allows the customer to use this exact style for new creations."
AI as a digital paintbrush. This approach ensures that the AI functions not just as a tool, but as an extension of the artist's creative expression. This creates a tool for the games industry that is tailored to the specific game for which content is to be generated.
Scenario at Red Meat Games: customization of characters and poses
The importance of specialized AI models is also clear at Red Meat Games. Judy Ehrentraut explained how the studio uses the AI engine Scenario to customize characters and poses precisely and efficiently.
Advertising
"We don't use images from Google, but work exclusively with our own images as the basis for the AI. This allows the developers to quickly bring a figure from their own artwork into the desired pose," explained Ehrentraut. This approach makes it possible to achieve consistent, high-quality results that preserve the studio's original style.
Ethical challenges and the role of the human being
With the rise of generative AI in game development, ethical questions are increasingly coming into focus. One of the central debates revolves around the question: who creates a work of art - the AI or the human? This discussion is by no means trivial, as it touches on fundamental aspects of artistic authorship and creative control.
Who is the true creator? AI vs. humans
Stupid AI? When an AI produces a work of art based on a human-trained model, the question arises as to who is responsible for the creative achievement. Judy Ehrentraut made it clear that current AI has no independent creativity, but merely implements what it has been taught by humans: "AI cannot learn on its own, but is only ever trained with what humans train AI with. Therefore, AI is not really 'intelligent'."
The actual intelligence and creativity still lies with the humans who train and control the AI. Nevertheless, the question remains as to how much the AI contributes to the final creation and whether it could possibly be considered an "artist" in its own right.
Advertising
Ethical considerations in the use of AI-generated art
Ethically clean training data. The use of AI-generated art raises a number of ethical questions. Is it morally acceptable to have an AI create artworks based on the styles and techniques of real artists without them being directly involved in the creation? And what about the integrity of a work that has been partially or completely created by such a trained machine? These questions are particularly relevant at a time when AI is increasingly being integrated into creative processes.
Blurred boundaries. Ehrentraut warned in her presentation that the uncritical use of AI could lead to a blurring of the boundaries between human creativity and machine production.
Responsibility. Reliability. Trust. She emphasized that it is important not to overlook the human factor in the creative process: The responsibility and real intelligence remains with the people using the tool, she said Ehrentraut. Otherwise, the boundaries of reliability, responsibility and trust would become blurred. This statement underlines how important it is to understand AI as a supporting tool that can enhance human creativity without replacing it.
Copyright challenges and the need for ethical data sources
Legal uncertainty. Another key issue is the protection of intellectual property. Are copyrighted works being misused when they are used as the basis for training AI models? The answer is not always clear, and current legislation is often insufficient to grasp the complexity of this new reality.
Advertising
Working ethically and safely. Ehrentraut argued that companies that use AI in art and game development must strictly ensure that their models are trained with ethically unobjectionable data. This is the only way to ensure that the results are both of high quality and ethically acceptable.
Responsibility for people. The challenges of using AI-generated art are manifold, and they require a careful balance between the technological possibilities and the ethical implications. It is the responsibility of developers and artists to find this balance and ensure that AI is used as a tool that supports human creativity without compromising its integrity.
AI overkill: the danger of quality decline
With the proliferation of generative AI in the art and gaming world, we are facing a new challenge: the threat of AI overkill. This refers to the phenomenon of the internet being increasingly flooded with mediocre AI-generated content, which can severely affect the quality and originality of online content. Judy Ehrentraut warns urgently of the risks that uncontrolled use of AI entails.
Risks of flooding the internet with mediocre AI content
Mediocrity. Imagine searching for unique, creative content online, but most of the results are AI-generated and mediocre at best. The content created by humans, which is often more in-depth and original, is becoming increasingly invisible in this flood.
Ehrentraut puts it in a nutshell: "The internet is full of AI. That makes it almost unusable." This statement illustrates the dilemma we find ourselves in. The more AI-generated content appears on the web, the more difficult it becomes to find the really high-quality works created by humans.
The problem of declining quality in AI-generated content
Another problem with AI generation is that the quality of the content decreases if it is repeatedly revised by AI systems. An illustrative example is text generation: if an AI creates a text and this text is then summarized or processed again by an AI, the quality decreases slightly each time. In the end, what remains is a text that is barely legible and only reproduces the original content in a distorted way.
Conclusion: The need for human control
Quality drops. The danger of AI overkill clearly shows that AI generation without careful human control and review can quickly lead to a downward spiral in quality.
Diligence during training. To avoid this development, it is essential that AI models are trained with high-quality, human-curated data and that the results are critically evaluated and corrected if necessary. This is the only way to ensure that the content generated by AI remains not only efficient but also of high quality and that the creative potential of the technology is fully exploited.
