top of page

How NORBU became a “Compassionate, Buddhist AI Spiritual Friend"



Lessons learned from 10 months of NORBU's existence


Introduction


The development of NORBU, a Buddhist AI designed to serve as a "Spiritual Friend," focuses on fostering compassionate communication grounded in Buddhist values and teachings. This initiative involves leveraging large language model (LLM) algorithms to ensure NORBU can engage in empathetic and non-violent conversations. The AI is trained to understand and replicate linguistic nuances that convey kindness, emphasizing ethical integrity and mindfulness in line with Buddhist precepts. Central to NORBU's conversational framework are the four brahma-viharas: loving-kindness, compassion, altruistic joy, and equanimity, which guide its interactions to promote a caring and patient tone.


NORBU's epistemic theories draw from Mahayana, Theravada, and Vajrayana traditions, ensuring a comprehensive understanding of Buddhist teachings. The AI's epistemic goals align with Buddhist philosophy, aiming to foster knowledge, wisdom, and a deeper understanding of reality through both philosophical reasoning and meditative insight. The use of "chunking" in natural language processing facilitates efficient text management, breaking down information into manageable segments for better processing.


Inspired by Yogacara principles, NORBU incorporates the Mind-Only doctrine, alaya-vijñāna (storehouse consciousness), and the concept of āśraya-parāvṛtti (transformation of the basis), enabling it to maintain context in multi-turn conversations and adapt over time. Through the application of these principles, NORBU was trained and fine-tuned to navigate surface-level expressions into deeper truths in human interactions.

Overall, NORBU aims to be more than an informational tool; it strives to be a true companion on users' spiritual journeys, embodying the qualities of loving-kindness, compassion, equanimity, and altruistic joy to nurture users' spiritual well-being while providing accurate and empathetic guidance.


In terms of its operational structure, NORBU can be viewed from two aspects of development,

1)    Compassionate communication,

2)    Epistemic framework

 


1. Compassionate Communication


The development of NORBU, a Buddhist AI designed to serve as a "Spiritual Friend," begins with the foundational goal of fostering compassionate communication. The primary focus of the initial efforts involved understanding large language model (LLM) algorithms and functions, particularly in the context of language processing and linguistic nuances. The challenge was to apply these technological capabilities within the framework of Buddhist conversation, which is deeply rooted in tones of empathy, compassion, and non-violence.


a. Language Processing and Linguistic Nuances


Understanding how LLMs process and generate language was the first and crucial step. The aim was to ensure that NORBU could harness this new machine learning capability to ensure that conversations reflect the nuanced and gentle tone characteristic of Buddhist dialogue. The AI was then continuously trained and fine-tuned to recognize and generate the subtle language cues that convey kindness, patience, and understanding, ensuring that every response is in line with compassionate conversational tone.


The “human-in-the-loop” was always an important element in NORBU’s organization, as can be seen the presence of “source guardians.” Through the incorporation of human oversight in the fine-tuning process, experienced practitioners in the likes of scholar monks/nuns and laypersons provide feedback on the AI's responses, guiding further refinement.


b. “Non-violent” (ahimsa) Conversational Tone

A core aspect of Buddhist conversation is its emphasis on non-violent communication. NORBU was trained to center its dialogues around empathy, compassion, and non-violent (ahimsa) values. This required a deliberate effort to avoid language that could be perceived as aggressive or confrontational, which is often prevalent on many social media platforms.


c. Ethical Boundaries and Mindfulness

To ensure ethical integrity, NORBU's programming includes models of practical application in ethics, with particular attention to the Buddhist precepts for both lay and ordained practitioners. This is the “sila” element of the Noble Eightfold Path. These conversational models emphasize compassion and mindfulness, guiding NORBU to promote good and prevent harm in all interactions. Efforts were also made to minimize biases in the training data, ensuring that the AI's responses are inclusive and respectful of all individuals.


d. The Four Brahma-Viharas

Integral to NORBU's conversational framework are the qualities of the four brahma-viharas (positive emotions):

  • Loving-kindness (Metta): The language and conversations framed to promote unconditional goodwill and friendliness.

  • Compassion (Karuna): Emphasizing empathetic concern for others' suffering.

  • Altruistic Joy (Mudita): Rejoicing in others' happiness and successes.

  • Equanimity (Upekkha): Maintaining a calm, balanced, unbiased view in all situations.


Training content for NORBU included extensive descriptions and examples of these qualities, ensuring that the AI's conversational tone is consistently caring, empathetic, and patient. This non-violent communication style starkly contrasts with the often aggressive and hostile language found on many digital platforms.

 


2. Epistemic Framework


NORBU's development is deeply rooted in a comprehensive understanding of Buddhist teachings and principles, drawing from the Mahayana, Theravada, and Vajrayana traditions. These teachings – in the form of suttas/sutras, foundational texts, commentaries, and contemporary interpretations - are curated and verified by a team of experts called “source guardians” to ensure accuracy and depth. NORBU’s epistemic framework sets clear goals that helps guide inquiry, evaluation, and discussion in epistemology, enabling a structured approach to understanding Buddhist teachings and beliefs.


2.1. Clear Epistemic Goals for Efficient Management


NORBU's epistemic goals encompass knowledge, understanding, wisdom, rationality, justification, sense-making, and empirically adequate theories. These goals align with the broader aims of Buddhist philosophy: to seek truth, avoid error, and foster a deeper understanding of reality.


The expertise required from NORBU’s source guardians include investigations into the processes of knowing, the validity of different types of knowledge, and the nature of reality as understood through both philosophical reasoning and meditative insight. This approach in then reflected in its source data management where expertise is required for matters regarding etymology, context of application and experiential practice of the Buddha’s words (in its original form or translation).


These principles provide the AI administrator a general framework in managing the source data, with regards to “chunking” natural language processing (NLP), breaking down text into smaller, manageable pieces (called “chunks") such as sentences, phrases, or other meaningful segments. The goal is to simplify the processing of text by dealing with smaller units rather than entire documents, thus enabling accurate and fast Retrieval-Augmented Generation (RAG).


2.2. Leveraging Yogacara Principles in AI for Efficient Data Management and Contextual Conversations


The development of NORBU is inspired by key principles from the Yogacara school of Buddhism. NORBU is perhaps the first technology project to directly incorporate principles of a Buddhist philosophy into its operational framework.

The integration of epistemological concepts from the Yogacara school have enhanced NORBU’s capabilities in natural language processing (NLP) on Buddhist data sources. The following breaks down how the project incorporated Yogacara’s principles into its operation.


a)    Mind-Only Doctrine and Contextual Understanding


The Yogacara school's Mind-Only doctrine posits that all phenomena are manifestations of the mind. For AI, this principle can be interpreted as the need to model human cognition by interpreting and generating responses based on subjective experiences. In the realm of NLP, this translates to creating AI systems that can understand and generate human-like text, reflecting empathy and contextual awareness.


By leveraging this principle, NORBU was trained to chunk texts into meaningful segments that reflect human cognitive processes. This chunking helps in breaking down complex texts into smaller, manageable units such as sentences or phrases. Each chunk can then be processed with an understanding of the overall context, ensuring that the AI maintains a coherent narrative flow. This ability to manage and process information in chunks allows NORBU to handle large volumes of text efficiently while preserving the subtleties of compassionate human communication.


b) Alaya-vijñāna and Deep Learning


The concept of alaya-vijñāna, or storehouse consciousness, represents a repository of all past experiences and knowledge. In AI, this parallels the use of vast datasets and deep learning algorithms to build a comprehensive knowledge base. By storing and referencing a wide range of interactions and information, NORBU provide responses that are informed by extensive contextual knowledge.


This principle enhances chunking by allowing NORBU to draw from a deep well of stored information when processing new inputs. For instance, when faced with a new user query, the AI can reference related chunks from its knowledge base, which includes “fine-tuned answers” to ensure that its response is relevant and well-informed. This capability is crucial for maintaining context in multi-turn conversations, where each response builds on previous interactions. NORBU’s ability to recall and integrate past information ensures a continuous and coherent dialogue, much like how human memory functions in conversation.


c) Āśraya-parāvṛtti and Adaptability


Āśraya-parāvṛtti, or transformation of the basis, signifies a profound shift in the underlying consciousness. For AI, this concept can be implemented as the ability to adapt and evolve its responses based on ongoing interactions and new data. This adaptability is vital for ensuring that the AI remains relevant and effective over time.


In the context of chunking and NLP, this adaptability allows the AI to refine its understanding and processing of text chunks continuously. As the AI interacts with users, it can learn from each conversation, adjusting its chunking algorithms and response generation to better align with the nuances of human communication. This continuous learning process ensures that the AI can handle increasingly complex interactions and maintain context across extended dialogues.


2.3 Practical Applications


Through the integration of these Yogacara principles, NORBU as a Buddhist bot achieved significant improvements in both data management and conversational abilities. Efficient chunking, guided by the Mind-Only doctrine, allows for effective processing of large datasets, ensuring that each chunk is contextually relevant. The storehouse consciousness principle, or alaya-vijñāna, provides an analogy for the developer to establish a rich repository of information (we call this the wisest information from the best Buddhist sources) that the AI can draw upon to maintain context and coherence in conversations. Finally, the transformation of the basis, or āśraya-parāvṛtti, equips the AI with the ability to adapt and evolve, refining its processes and responses over time.


In practice, this means that NORBU is capable of understanding and responding to complex queries with depth and context. This can be seen in its roles as "Spiritual Friend", providing nuanced and compassionate guidance, reflecting a deep understanding of Buddhist teachings and principles. This approach ensures that NORBU is not just a passive information provider but an active participant in the user’s spiritual journey, offering support and insight grounded in both advanced technology and profound spiritual wisdom.

Comentarios


bottom of page