Table of Contents
Knowing that your child will be affected by diabetes for the rest of their life is difficult, and teaching them how to cope with diabetes in a fun and productive way is even more difficult. Generative artificial intelligence (AI) offers an opportunity to enhance educational games to be intuitive, informative, and engaging.
Learn more about a proof-of-concept application that leverages the Amazon Titan Large Language Model (LLM) through Amazon Bedrock to personalize type 1 diabetes education for children in a gamified environment. The goal is to integrate generative AI with serious games (games designed to teach specific topics) to enhance the education available to young children.
Diabetes, which is characterized by hyperglycemia, is Approximately 1 in 9 people in the United States is affected.Includes more than 300,000 children with type 1 diabetes (T1D). Without proper blood sugar control, diabetes can lead to serious health problems over time, including heart disease and kidney disease.
Gamified learning experiences for teaching diabetes management have been attempted in the past, but they have had limitations. These included complex, action-packed gameplay and presented a high barrier to entry, especially for younger children between 4 and 7 years of age, the peak age for type 1 diabetes. Such games also lacked variety, presenting only a set number of pre-programmed scenarios.
T1D Learning Camp games address these issues by reducing gameplay complexity and increasing interactivity. In the game, users play as a dog character who has type 1 diabetes and goes on an overnight camp. At camp, you will learn how to make healthy food choices and the effects of different foods on people with type 1 diabetes. A character named “Dr.” “Robot'' acts as the player's guide throughout the camping experience.
The game is a combination of:
- Text and response based have a conversation section
- mini game playing section
- island explore section
in have a conversation In the section, players participate in lessons learning about concepts such as blood sugar levels and the difference between a high-carb “rocket diet” and a low-carb “training meal.” You will also learn about the symptoms of hyperglycemia (hyperglycemia) and hypoglycemia (hypoglycemia). Players will talk to Dr. Robot about these topics and the food they are eating.
Rather than having to manually plan hundreds of potential conversation paths for possible user inputs, this project leverages generative AI to create customized responses in near real-time (Figure 1) .
Figure 1: Dr. Robot and the player discuss their current diet in the conversation section.
in playing In the section (Figure 2), children play mini-games that include gameplay mechanics such as sorting food into categories of “rockets” and “trains.” Also, blow away the symptoms with the right kind of food (rocket or train). Points are awarded when users successfully sort items. Future iterations of the game are expected to include virtual medals that users can earn when exceeding certain level points. This reinforces the concepts presented during the conversation sections and encourages continued gameplay.
Figure 2: “Symptom Space Destroyer” mini-game in the play section.
in explore section allows players to control their characters around the island and move between different lessons and mini-games (Figure 3). As you explore, you can also find non-player characters to chat with.
Figure 3: Non-player characters on the island who interact with the player in the exploration section.
High-level architecture overview
T1D Learning Camp games are Godot game engine It then calls the Amazon Bedrock API to support conversations between characters. Because Godot uses a custom programming language, gdscriptSDKs such as Boto3 for Python cannot be natively integrated.
To manually initiate a connection, Amazon API Gateway is configured to use Python to invoke an AWS Lambda function, invoke the Amazon Bedrock API, and return the results to Godot. This mechanism using API Gateway allows Godot's HTTP request nodes to make these requests without sending the required signatures (Figure 4).
Figure 4: High-level architecture of the T1D Learning Camp game.
To enable personalized chats with Dr. Robot, Godot's custom “conversations” resource stores a set of pre-built dialogs, placeholders for user responses, and spaces for Amazon Bedrock responses. I am.
Use prompt engineering to help your characters provide context-appropriate responses in simplified language. Added parameters to the Amazon API Gateway request module that allow you to specify LLM settings directly from Godot. The parameters length, temperature, and top-p are used to control the maximum tokens of the response. length, randomness, diversity (Figure 5).
Figure 5: Conversation structure.
Additionally, Amazon Bedrock Guardrails is used to control the responses provided to players to avoid medically inaccurate and unreliable information and medication advice, such as how much insulin to take and when to take it. Amazon Bedrock Guardrails allows you to accept and deliver secure prompts and responses with customizable protections in addition to native LLM protection. T1D Learning Camp's conversation system, powered by Amazon Titan and Guardrails, is an important example of how large-scale language models can be used in games.
Figure 6: Dr. Robot returning a response generated by Amazon Titan. This is the answer to the player's question: “Can I eat watermelon forever?”
Future improvements
In addition to using Amazon Bedrock in the Conversations section, you can also leverage generative AI further to enhance the content, engagement, and accessibility of your T1D Learning Camp, including through the use of the Amazon Bedrock knowledge base. Amazon Bedrock Knowledge Bases is a fully managed feature that helps you implement the entire retrieval extension generation (RAG) workflow, from ingest to retrieval to prompt extension. RAG allows LLMs to create responses that include additional details from the type 1 diabetes resources provided. Applications can use Amazon OpenSearch Service as a vector database and Amazon Simple Storage Service (Amazon S3) as a data source for Amazon Bedrock Knowledge Base.
Future iterations will include the use of Amazon Transcribe, an automatic speech recognition and speech-to-text machine learning service that allows users to converse in applications rather than typing. This feature lowers the barrier to entry for young children who are not proficient in writing or typing, and also improves accessibility for the visually impaired.
Another planned improvement is to leverage Amazon Polly's generative AI text-to-speech capabilities to provide more natural and understandable voices for game characters. Many of the text-to-speech voices currently built into computers sound robotic and can be difficult for children and people with hearing loss to understand.
Finally, improvements could be made to extend the use of the application to non-English speaking children. Amazon Translate can work with the ability to generate speech in different languages through Amazon Polly to translate responses from Amazon Bedrock. Additionally, the Amazon Titan Image Generator foundation model dynamically generates stylized artwork for any food item and helps you personalize the experience based on user input. A child's caregiver can create a list of foods specific to the child's diet or culture and add or replace these foods to the basic set of foods. These changes have the potential to make games more accessible not only to young children but also to children with different languages, cultures, and diets (Figure 7).
Figure 7: High-level architecture with future improvements planned. Amazon S3 is used as the data source for the Amazon Bedrock knowledge base, which uses Amazon OpenSearch Service as the vector database. Amazon Titan Image Generator is the foundational model within Amazon Bedrock.
conclusion
Generative AI has the potential to change the way games interact with players, including serious games for children with various medical conditions. Ultimately, bringing generative AI to projects like T1D Learning Camp can improve patient outcomes and provide a more personalized, engaging, and accessible experience for children with type 1 diabetes. Possibly.
contact address AWS representative To find out how we can help accelerate your business.
For more information about AWS for Healthcare & Life Sciences, a selection of AWS services and AWS Partner Network solutions used by thousands of healthcare and life sciences customers around the world, see AWS for Healthcare & Life Sciences and AWS Visit the Healthcare Solutions webpage. You can also read more blogs about AWS Healthcare stories.
