Step into the role of a detective and uncover the thief from a list of 12 potential suspects. Use the documents you unlock each time you eliminate two suspects to aid your investigation. Each game you start will be unique, thanks to Generative AI. Embark on your own adventure, where you are solely responsible for solving the case.
​​
ROLE
Game Designer/ Level Designer
YEARS
2023/2024 (1 year project)
TYPE
Point and click
Deduction game
PLATEFORM
PC
What was my role in this project?
​
In this project, my role was to develop the concept of a game entirely generated by AI and bring that idea to life, while ensuring the game remained engaging and enjoyable for the player.
Team
3
Investigation game (Deduction)
Notebook as the main element for the investigation
What possibilities does Generative AI offer?
​
Currently, AI can generate high-quality text and images, depending on the model chosen. With this capability in mind, we decided to create a visual novel game. This genre offers a wide range of creative possibilities, but one type of game, in particular, caught our attention: the detective/deduction game.
Research and References
​
Once we had defined the game genre, the first step was to find references to begin developing a concept. This phase led us to discover three types of detective games: deduction, investigation, and contradiction-based games. After analyzing each, we quickly realized that to easily implement AI, deduction-based detective games were the perfect model for us. This became the foundation for our initial concept.
Investigation games (Investigation)
Simulation Game
Interesting features for our game
Comedic atmosphere + Notebook containing descriptions of events
Connections made using pins/board
Text-based deduction with fill-in-the-blanks
Investigation games (Contradictions)
First Concept
​
In the first concept, the player participates in a criminal investigation where they must interrogate two suspects to determine the culprit. To gather information, the player has documents on the suspects that need to be filled out by asking the appropriate questions. Once the documents are completed, the player receives reports for each suspect to detect inconsistencies in their statements.
Document Analysis in Gameplay
Number of documents the player will need to analyze : 12 ​​
​For the first concept, which consists of two gameplay phases — document analysis and interrogation — I spent a significant amount of time on the second phase, the interrogation. I developed several concepts that were eventually removed from the final version, but they helped us refocus the project's vision. This also allowed us to better align the game with the possibilities offered by AI and our ability to implement them within just one year.
Iteration Phase: Interrogation
.jpg)
In this concept, the player interrogates a suspect and must fill out the "alibi" sheet on the left side of the screen. The player needs to ask the right questions while monitoring the status bar at the top, which shows the suspect's emotional state. If the player makes too many mistakes, the suspect can call their lawyer (indicated by an icon in the top right of the screen), which immediately ends the interrogation for a certain amount time.
.jpg)
In this version, the status bar is still present, but this time the player must ensure that the black icon, representing the suspect's level of stress, stays within the purple area at the top of the screen. To achieve this, the player has tokens, previously obtained through mini-games, which allow them to either threaten the suspect with a "gun" token or reassure them with a "coffee" token.
In this version the bar at the top of the screen, representing the suspect emotionel state and willingness to speack, moves, and the player must keep it centered to get the suspect to talk, either by threatening him with a gun or reassuring him with a flower. However, the player must be careful not to be detected by the camera, which can turn on without warning.
In this version, the player simply needs to press the right button on the keyboard at the right time to get the suspect to talk, while music is playing in the background.
Problems with the first concept
Suspect personality system is too complicated to implement with ChatGPT.
​
Overly simple question system.
​
Document reading is too monotonous and lengthy.
​
Many features are underdeveloped.
​
Notebook usage is nonexistent.
​
Risk of investigation being too long.
Second Concept
​
In this second concept, after considering the issues with the first, we decided to remove the interrogation phase and keep only the document analysis phase. At this point, a new reference came into play to help us: the game Papers, Please, which is entirely based on document analysis. This time, the gameplay involves giving the player all the documents, and the player must cross-check the investigation information with the suspects' statements, identify inconsistencies and correlate the investigation with the suspects' details.
.jpg)
Inconsistency
​
When the player identifies an inconsistency by connecting two pins between a suspect's statement and a document the player has, a red link is created.
Correlation
​
When the player finds a match by connecting two pins between certain documents, a blue link is created.
Number of documents the player will need to analyze : 12 ​​
Improvement
Single-phase: simplified gameplay more in ligne with the investigative vision we wanted for our game.
​
Notebook usage implemented.
​
Different documents => Easier to convey suspect personality details through the use of ChatGPT.
Problems with the second concept
Investigation taking too long per suspect.
​
Document reading is too monotonous and lengthy.
​
Pin system too complex to implement and integrate with ChatGPT.
​
No increase in difficulty.
​
All documents are displayed on the board which could lead to player being overwhelmed with information.
Pre-final concept:
​
In the pre-final version of the game, the player faces 10 suspects and must identify the culprit using the information available in the player's notebook. Additional information about each suspect is unlocked as the player clears them of suspicion. This version is heavily inspired by the board game "Guess Who?", a reference that likely came to mind.
Number of documents the player will need to analyze : 5 ​​
​In this version, we decided to drastically reduce the number of documents from 12 to 5. The player must analyze these documents to find evidence incriminating certain suspects and clear others, removing them from the investigation to get closer to identifying the real culprit.
I selected four documents that could be interesting to include in the game, allowing for some creative freedom in terms of humor and, most importantly, that don't seem too difficult to implement with AI. Once selected, I had them approved by the team.
​​Police reports
ID
Interrogation transcripts
Medical reports
Photographs and videos
Physical evidence
Phone and electronic records
Internet history
Financial documents
Witness statements
Expert reports
Logs and records
Criminal records
Creation of documents for gameplay
​
The first step in creating the documents was to compile a list of the documents that could help solve a theft investigation and identify the culprit in real life. To do this, I used ChatGPT to generate an initial list, which I then verified by conducting my own research online to confirm its accuracy.
In-depth research
​
To create the documents, I used real life documents to first understand their content and then how the information could be presented.
_edited.jpg)
_edited.jpg)
First prototype​​
Playtest
​
Before developing the second concept of documents, I created playtest documents for each of the four documents to identify areas for improvement (issues).
​After the playtest, I presented a series of questions to several people to gather their feedback and help me improve the documents.
Problems with the first prototype
Too much information to read and to consider for the investigation
(risk of the investigation being too long)
​
Monotonous and unpleasant reading
Improvement
​
For the design phase, I reduced the number of lines in the internet history from 12 to 8. Additionally, certain details in the other documents are no longer considered in the investigation, such as the expenses on the bank statement, the conviction dates on the criminal record, and the gender on the ID card.
_edited.jpg)
Selected Typography
Added transparency to the chosen font to mimic the effect of dried ink.
​To improve the visual readability of the documents, and considering that our game relies heavily on reading, I aimed to make information easier and faster to absorb by focusing on typography. I researched the most suitable fonts for quick and efficient reading. Then, I created a table where each sentence was written in a different font and asked several people to read the table and tell me which sentence was the easiest to read and the most visually appealing.
​Graphic improvement by achieving a more realistic effect.
Final documents
Guilty Or Not
