This work is relevant for Innovation because bAIgrapher uses Artificial Intelligence to transform therapy for potentially millions of patients with Alzheimer’s disease.

Reminiscence therapy is considered an effective tool for patients with Alzheimer’s disease, but it is costly and time consuming, making it impractical for most people.

bAIgrapher, which has been developed before the emergence of ChatGPT and GPT4, is about to change the therapeutic landscape. Through the sensible and planned use of Artificial Intelligence, it offers a form of interpersonal reminiscence therapy that is accessible, scalable, medically responsible, and scientifically rigorous.

The project relies on several Natural Language Processing (NLP) tasks, like speech recognition, context analysis, and speech synthesis. As a platform capable of creating a book, it is a new application within the field of Generative AI.

One of bAIgrapher’s main components is a Large Language Model (LLM) that was fine-tuned from a GPT-3 backbone, but it’s much more than that: it’s an amalgam of systems working in sequence and in tandem. When put in motion, data collection doesn’t scrape the internet; it collects a finite, curated, and highly personalized set of data from willing sources. This circumvents a major limitation from popular LLMs available today: “hallucinating” (generating text that isn’t true and confuses historical facts).

bAIgrapher’s fine-tuning process is trained on over 400 existing biographies to improve its writing skills, so it is not only a good read, but also makes sense for patients.

BAIgrapher is the first platform to use generative artificial intelligence to make reminiscence therapy accessible and practical for people with Alzheimer's disease. First, bAIgrapher collects stories from patients and their loved ones. Then, it creates personalized biographies that help patients reconnect with their past, a technique that can help strengthen memory and maintain cognitive functioning.

Developed months before the emergence of ChatGPT and GPT4, the AI behind bAIgrapher is different: trained in over 400 biographies, it’s uniquely equipped to write a compelling life story. Unlike other large language models, which rely on publicly available information, bAIgrapher generates content based on specific information provided by the user.

The resulting biography is delivered as print or audiobook, allowing patients to relive their memories in a way that is accessible and engaging. The platform provides a powerful tool for improving the quality of life and emotional well-being of people with Alzheimer’s disease.

February-June 2022 - Conception of the specific architecture, including AI algorithms, models, and literary research.

June 2022 - Technical and medical discussions about the platform’s design and architecture to ensure it was ethically and medically responsible. These conversations included, for example, decisions about the inputs, the people who would be allowed to participate, the form of the participation, how each input would be used, and the overall focus of the project around the patient’s health and life.

June-August 2022 - Architecture development of platform's 1st version.

August 2022 - Platform up and running.

November 2022 - Finding the first patient. It required around two weeks to contact medical doctors and family members, obtain consent, and confirm availability.

December 2022 - Submission of inputs. This includes time for gathering a robust number of inputs from patient, patient’s family, and friends.

December 2022 - First bAIgraphy production: this includes generation of textual content, audiobook generation, photo and image placement, production and design, physical book printing, and full team review.

January 2023 - Patient gets her own bAIgraphy in print and audiobook form and begins her reminiscence therapy. Family and friends also receive copies.

Data quality is critical in any well-designed AI-based project.

bAIgrapher is centered around the patient’s personal story. Their recalling of previous experiences, especially those of pleasant nature, is essential, particularly during early stages of the disease.

bAIgrapher’s platform sequentially guides this process, asking specific questions, and suggesting diverse topics. Close family and friends are a fundamental component by contributing with testimonials. Inputs can be text, audio, or images. Depending on audio quality/quantity, the patient’s voice can be synthesized to produce an audiobook.

This guided process is essential because it associates inputs with specific dates, places, events, and other contextual information. The resulting input data is organized on a clear timeline and well-labeled, which enhances bAigrapher’s performance, generating high-quality and consistent outputs that preserve original context and information. Images are also automatically added to the book when pertinent.

Since the AI mustn’t create new false information or distort the context of the inputs, the amount of content in the book is, ultimately, proportional to the amount of input data. The more data imputed, the longer the book. In case of excess input data, bAIgrapher can write a summarized book, but we don’t recommend writing a longer book from little input data.

UFCSPA, the top medical university in Brazil, sees great potential in bAIgrapher as a tool that can help thousands, if not millions, of patients with Alzheimer’s disease, and they intend to make it part of their treatment protocol. UFCSPA will also extend use of the tool to affiliated medical colleges, including Ivy League universities like Harvard and Brown.

The platform can revolutionize reminiscence therapy, making it truly scalable and accessible, which is not possible in its current format.

bAIgrapher’s technology is highly scalable while maintaining medicine’s four “Ps”: It is predictive, preventive, participatory, and personalized to different cultural and family contexts. bAIgrapher challenges what’s possible and accessible in terms of medical care and quality of life for patients and families.

bAIgrapher pushes the boundaries of AI, offering a more humane and responsible approach. Contrary to the perceived AI threat of replacing people, bAIgrapher highlights how irreplaceable we all are.

Alzheimer’s disease strikes families, not only individuals, and addressing the needs of the patient’s family is essential to optimize the quality of life of the affected person and their caregivers.

Reminiscence therapy uses written or oral life histories to improve psychological well-being. In its interpersonal variation, it is a form of group therapy that can be very effective but also costly and impractical.

bAIgrapher is the first scalable interpersonal reminiscence therapy for patients with Alzheimer’s disease, but it is also a powerful healing tool for their families and close friends. Because of its practical nature and seamless interface, bAIgrapher allows for an entire community to contribute their stories about the patient, which in turn allows them to become a fundamental part of that person’s fight against the disease.

To enhance care for someone by effortlessly telling that person’s stories can be highly cathartic for everyone involved.