Sponsored By

Instead of peering too far into the future today, let’s try to understand how AI can help game designers right now.

Mariya Aksyutina, Blogger

August 21, 2023

7 Min Read

2023 has brought with it a ton of new AI-based tools, and game developers around the world are discussing how these new tools will affect the industry in the future: proponents are excited about new opportunities, while others worry that AI may strip games of their originality and creative uniqueness.

But instead of peering too far into the future today, let’s try to understand how AI can help game designers right now.

Hello everyone, my name is Maria Aksyutina, and I'm a game designer at MY.GAMES. In this article, I’ll share the results of my research; I tried to find out how well the sensational ChatGPT copes with game design tasks and whether it can actually be used in our work.

I identified five types of tasks:

  • Game idea generation

  • Describing in-game mechanics

  • Game balance calculation

  • Generating in-game texts

  • Creating technical tasks for specialists

There are currently two versions of ChatGPT available: 3.5 and 4.0. For this experiment, I decided to compare how these two versions deal with the same tasks. I ultimately concluded that both had more or less the same results (although version 4.0 performs better with many other tasks). I spent approximately the same amount of time and effort correcting and improving results with both versions, so I decided to use the free version 3.5.

Disclaimer: this article only contains personal experiences using ChatGPT; your own experience might produce different results. Prior to using ChatGPT for commercial purposes, we advise getting a legal opinion.

Writing a pitch

Let's start with the most interesting and exciting task for a game designer – creating a new game.

We’ll generate a task just like it might be given in a test task or at a game jam:

You’ve been tasked with combining two core mechanics (Idle and Merge) and creating a hybrid casual game set in Victorian England. Draw inspiration from current trends and top-grossing games in the genres of Idle, Arcade, Simulation, and Hyper-casual. Write a brief concept document (1000 characters) where you describe the game idea: core gameplay — elaborate on the game cycle, mechanics, and their connections, the setting and narrative, and the unique selling point (USP) — in a way that your game will be better than what already exists in stores.

Here's the result:

The chat managed this task pretty well; let me highlight some strong points:

  • The chat understood the task correctly, and it knew the genres and basic mechanics quite well

  • All the game elements corresponded to the given setting

  • The pitch was structured and all points were taken into account

  • Yes, it could really be used as a pitch

However, there were also some drawbacks:

  • It offered a very simple solution to the problem, and a bit too on the nose

  • The chat gave shallow descriptions of the game loops

  • The pitch was made up of general phrases; it had little details about the product

  • The USP description was based on marketing parameters, not potential unique benefits

Still, in general, a game designer could use a pitch like this as a basis and then refine it further. It could also serve as a template and a starting point for creating your own game.

Describing mechanics

Once an idea has shaped up, it must be described in the documentation so that all developers on the project understand how the new idea works. The description of the mechanics should be clear and complete, but not over and above what’s needed.

Game designers often consider describing mechanics as their least favorite part of the work. I decided to make ChatGPT’s job a bit more difficult. Instead of feeding it with an exact description of the mechanics I needed it to describe, I created a rather abstract task with restrictions: to create and describe a control system for a mobile FPS.

Here’s the text of the request:

Create and describe the character control mechanics for a mobile first-person shooter. The player should have access to the following actions: movement, shooting, dash, and grenade throw.

And here’s the result:

Safe to say, ChatGPT handled this task well. The points of praise:

  • ChatGPT understood and completed the task. All the requirements and restrictions were taken into account

  • it not only offered control through a button system, but also via screen gestures

  • The description was clear and structured

But, as always, there are some things we still need to work on:

  • The control system was described in general terms

  • ChatGPT didn’t consider some exceptional cases, like when movement wasn’t available for a player

  • It also didn’t sufficiently think through the states for the controls


And, of course, these documents should be accompanied by explanatory schemes and mockups, but this is still beyond the capabilities of ChatGPT.

Game balance calculation

ChatGPT’s developers claim that it can help solve mathematical problems. There are tons of examples online showing ChatGPT solving such problems, but game designers often must work with cases where the conditions of a task can sometimes be clear or incomplete.

Let's ask ChatGPT to do a task that a developer might face:

Present configurations of currency generators in an incremental idle game. Simulate the optimal choice of buying generators, and plot the income per second from generators from playtime in minutes. p.s. The player starts with one generator. The costs and revenues of the generators are described by the following formulas: costnext = costbase *(rate growth)^owned productiontotal = productionbase * owned * multipliers

Here's the solution:

And that’s just great:

  • The chat understood the task correctly, noted that some data was missing and supplemented the solution with its own values

  • In addition to the solution, ChatGPT provided comments on its actions

  • After getting the solution, you can enter your own values, instead of those chosen by the chat, and adjust the calculations to fit your needs

That said, it can be inconvenient to work with mathematical parameters in the chat format with a large amount of data, but it absolutely suits our request.

Generating in-game texts

Let’s move from mathematical problems to a more creative area of development, generating in-game texts..

ChatGPT handles writing marketing and journalistic texts very well, but can it ace writing in-game dialogues?

Here’s our task for ChatGPT:

Write 3 small dialogues for F2P Match-3 game. Concept: cafe renovation. Characters: cafe owner, assistant, disgruntled landlord. Choose the characters who will participate in the dialogues on your own. Audience: women 35+ years old.

Solution:

Advantages of this solution:

  • ChatGPT generated three short dialogues with a small number of remarks and short phrases. This is fully consistent with the formal recommendations for game dialogues.

  • According to the short description, the chat added a simple plot and used all the proposed characters.

Disadvantages:

  • The dialog requires some follow-up work, anyway. Some phrases need to be changed, to make them look more plausible. And we need to take out mentions of Match-3 from the characters’ remarks.

Creating technical tasks for specialists

For ease of evaluation, we asked ChatGPT to describe the mechanics of the popular 2013 game Flappy Bird and decompose it into tasks for specialists. To make sure that the game materials were included in the training dataset of the model, we asked if the chat knew this game.

We then asked the chat to both write a technical description of the mechanics involved with controlling the bird and to make a list of tasks for the programmer, artist and animator:

  • From the first answer, it’s clear that ChatGPT is really familiar with the game and understands its basic elements

  • ChatGPT correctly correlated the roles and tasks of specialists on the project

  • Tasks contain sufficient details and comments

Concluding today’s chat

ChatGPT is a powerful and promising tool that can help solve a wide range of tasks. But, as it has been proved today, it cannot fully replace the work of a game designer. Modern AI algorithms aren’t yet capable of creativity and intuitive decision making. For game designers, the emotional aspect of games plays a large role, and this area remains a unique human skill.

Nevertheless, ChatGPT will be a great help for new game designers. It can serve as a tool for analyzing and searching for information, as well as for creating templates and examples of design documents. Experienced developers can use ChatGPT to help generate ideas and quickly solve well-defined tasks.

Read more about:

Blogs
Daily news, dev blogs, and stories from Game Developer straight to your inbox

You May Also Like