Life Tips Miscellaneous Travel & History Zen and Life Tips Read Classics Philosophy Navigation of this blog
The world is made up of relationships.
Carlo Rovelli’s*Helgoland* presents a ‘relational interpretation’ of quantum mechanics and extends them to the origins of our world.
The following section describes its contents.
Rovelli’s relational interpretation of quantum mechanics explains that matter does not exist as an ‘independent entity’, but only makes sense in ‘relations’. In other words, objects and particles do not exist on their own, but their properties are determined only by their interactions (relations) with other objects. For example, from a quantum perspective, the position and momentum of an electron are determined by its relationship with the object it observes, and it can be considered that different ‘realities’ exist for different observers. In this way, the basis of reality is the relationship between objects rather than the objects themselves. This view also leads to the idea of possible worlds, which is discussed in ‘Possible worlds, logic, probability and artificial intelligence’.
Based on this relational view, our intuitive understanding that there is an entity and it can be observed is overturned; rather, the essence is how observers and objects relate to each other. In this relationship, the ‘reality’ we perceive is not a fixed entity, but is composed of a series of interactions. In other words, Rovelli suggests that there is no independent ‘absolute reality’, but that matter and existence have a ‘relative reality’ based on their respective relationships.
The title ‘Helgoland’ comes from Werner Heisenberg’s 1925 idea of matrix mechanics, the basis of quantum mechanics, on the German island of Helgoland. Rovelli reflects on Heisenberg’s revolutionary discovery and reveals the enigmatic nature of the quantum world.
Rovelli states that this relational interpretation of quantum mechanics also has implications for philosophical questions. In questions such as human consciousness, what the self is and what it means to exist, Rovelli offers suggestions from physics and sees the existence of the self as also being constituted in a relational way.
Thinking about consciousness and the existence of the self from this relational perspective, it is possible to consider that the self does not exist independently, but emerges from relationships with others and the environment, and that consciousness is not one isolated ‘thing’, but a phenomenon that emerges from the complex interactions with the surrounding environment and other people.
Heavy Imperial Net and ‘relational interpretation’
This relationship-based worldview is also shared by Buddhist worldviews such as ‘The Internet and Vairochana Buddha – The Kegon Sutra and Esoteric Buddhism’. The Kegon Sutra presents a view of the world in which the various worlds are interconnected, with ‘one being many and many being one’, for example, in the Chong Tei Net, the world is described as a vast net of countless jewels, each jewel reflecting all the others, each jewel existing in relation to the others, and all beings depending on each other. Every being is said to be interdependent and connected to everything else. ‘Reality’ here is not one independent thing, but is constituted by an infinite number of relationships with others and the environment.
This perspective is similar to Carlo Rovelli’s ‘relational interpretation’ of the world, in which physical existence also has no meaning in itself, and existence is only defined by relations with other objects and observers, and shares the view that reality is not an isolated entity but exists through a network of relations The two share the view that reality is not an isolated entity, but exists through a network of relations.
Rovelli argues that reality is only defined by interaction, which is also expressed in Tei-net’s notion of ‘one-is-all and all-is-one’ (one contains the whole and the whole is contained in the one). Just as in Tei-net each being mirrors all others, so in Rovelli’s quantum mechanics each observation or interaction is said to affect all other elements, all being inseparably bound together.
These worldviews are not fixed, but interpret the world as a dynamic field of interaction.
Extension to AI technology
The following are possible applications of this perspective, which interprets the world as a dynamic field of interaction, to AI technology.
1. relational-based AI architecture: in Rovelli’s ‘relational quantum mechanics’, the existence of objects and information is defined in terms of relationships rather than as independent entities. Similarly, incorporating a ‘relational-based’ architecture in the design of AI systems enables AI to generate meaning through constant ‘relationships’ with humans and other systems. For example, AI systems could be developed that take into account the context of user inputs and actions and respond flexibly accordingly.
In this approach, it is possible to realise systems where AI does not process pieces of data as independent entities, but where each piece of information is defined by its relationship to other information, which goes beyond the linear flow of conventional AI of ‘input → processing → output’ and allows the entire network to dynamically share information and 1. functions as a structure in which the entire network dynamically shares information with each other.
2. knowledge graphs as networks of relationships: the idea of the ‘Tei-net’ is the concept that all entities influence each other and exist as an overall network structure. One way of incorporating this concept into AI technology is through the use of knowledge graphs, which are network structures that describe data not as a single piece of information, but including relationships with other data.
Such knowledge graphs can be extended to reproduce an ‘Indra’s web’ structure, where information is interrelated with all other information. In this approach, each node (information) has diverse relationships with other nodes, and new insights and patterns may be discovered based on the relationships. For example, by finding new relationships between individual data points, a knowledge inference engine can be built that provides deeper insights.
3. interdependent AI models: traditional AI models are dominated by approaches based on explicit causal relationships and independent tasks, but by incorporating Rovelli’s relational approach and Tei-net’s network perspective, it is possible to build AI models that are interdependent. In this model, the aim is not for each subsystem to operate independently, but for the system to constantly adapt while being influenced by each other.
For example, a ‘multi-agent system’ could be considered, in which AI agents exchange information with each other and learn and make decisions based on their understanding of each other’s situation and state. In this system, individual agents learn about their relationships with other agents, enabling them to be adaptive and flexible even when the environment changes dynamically.
4. personalised AI based on relationships with users: based on Rovelli’s theory, it is also possible to personalise AI methods based on ‘relationships’ with users, for example, a system where the AI remembers past interactions and actions with users as relationships and adjusts its responses and suggestions accordingly. This helps to build a ‘relational AI’ that makes decisions based on the entire relationship with the user, rather than just analysing data.
By changing the AI’s responses according to the depth and type of relationship, it is possible to achieve more natural and user-friendly interactions and provide a user experience where the AI feels as if it is a partner with whom the user has had a long-standing relationship.
5. indra’s web data correlation and pattern discovery: based on the holistic networked thinking of ‘indra’s web’, AI techniques can also be devised to discover correlations and patterns across data sets. Such AI would not merely detect salient relationships within a data set, but would also be able to detect potential interdependencies and non-linear relationships between data.
For example, in market research, disease diagnosis or climate change analysis, it can serve as a tool for discovering patterns where multiple factors are intricately intertwined, as AI learns not only independent patterns but also a ‘web’ perspective, where every data point interacts with all others 5. can provide correlations and new insights not found by traditional methods.
6. relational-based ethical AI design: based on Rovelli’s relational perspective and Indra’s web philosophy, a new direction for AI ethics can also be seen: AI can be designed with an awareness of interdependence, considering the impact of AI decisions on individual users and society as a whole, so that AI decisions and learning processes are not only relevant to specific users but also to It is possible to deepen ethical considerations by taking into account how AI’s decisions and learning processes relate not only to specific users, but also to other systems and society as a whole.
For example, a ‘relational-oriented ethical AI’ can be proposed, which considers how, if an AI has some impact on a user, that impact will also spill over to other users and society as a whole. Such an AI could improve ethics and transparency, as the system would be less prone to the spread of prejudice and erroneous decisions, as it would focus on the effects of relationships.
By extending Rovelli’s relational quantum mechanics and Indra’s web philosophy to AI technology, the potential for AI to evolve from a traditional ‘data processing machine’ to a ‘relational, interactive entity’ becomes apparent. This will enable AI to assume a new role as a ‘partner’ that grows in symbiosis with humans and society, influencing each other, rather than simply being a tool.
implementation example
This section describes basic implementation examples of relational AI design. The following implementation examples aim at systems in which AI learns and dynamically changes its interrelationships with users and the environment.
1. personalising users using relationships
A system that tracks the ‘relationships’ of each user and learns and personalises specific behaviour patterns and reactions. Based on the user’s past choices, questions and dialogue content, the AI forms a relationship with each user and provides more appropriate responses.
Example implementation: relationship-keeping AI assistant
# Importing the required libraries.
from collections import defaultdict
import random
# Dictionaries holding user relationship data.
user_relationships = defaultdict(lambda: {'preferences': {}, 'history': []})
# Functions to generate a relationship-based response.
def generate_response(user_id, message):
# Get user relationship data.
relationship = user_relationships[user_id]
# Inference of user interest from past messages.
if "recommendation" in relationship['preferences']:
response = f"Based on your interest in {relationship['preferences']['recommendation']}, I recommend..."
else:
response = f"What would you like to know more about, {user_id}?"
# Save message in history
relationship['history'].append(message)
# Updated guess data.
if "book" in message:
relationship['preferences']['recommendation'] = "books"
return response
# examples showing the use (of a word)
user_id = "user123"
message = "Can you recommend me a book?"
print(generate_response(user_id, message))In this script, when a user sends a message asking for a ‘book’ recommendation, the AI learns the relationship and uses it in its next response. By continuously updating the relationship, personalised responses can be generated over time.
2. knowledge graph systems that mimic Indra’s web
Based on the idea of an ‘imperial web’, a knowledge graph is constructed in which all information is networked together, allowing the AI to relate information and draw new knowledge. For example, if user input is associated with a particular node, the system will also refer to the surrounding nodes to generate a response.
Example implementation: knowledge graph construction and relationship retrieval
import networkx as nx
# Creating graphs
knowledge_graph = nx.Graph()
# Definition of nodes and relationships
knowledge_graph.add_edges_from([
("AI", "Machine Learning"),
("Machine Learning", "Deep Learning"),
("Deep Learning", "Neural Networks"),
("AI", "Ethics"),
("Ethics", "Transparency"),
("Transparency", "User Trust"),
])
# Function to generate a relationship-aware response.
def generate_graph_response(topic):
if topic in knowledge_graph:
# Examining relationships from connected nodes.
related_topics = list(knowledge_graph.neighbors(topic))
response = f"The topic '{topic}' is related to: {', '.join(related_topics)}"
else:
response = "I don't have information on that topic."
return response
# examples showing the use (of a word)
topic = "Machine Learning"
print(generate_graph_response(topic))In this example, a user asks a question about ‘Machine Learning’ and a response is returned that considers the relevance of ‘Deep Learning’ and ‘AI’. This approach allows the AI to flexibly draw knowledge from the network structure of information and include it in the response.
3. interrelational learning in multi-agent systems
Making use of Rovelli’s ‘relational’ perspective, a multi-agent system is constructed in which agents learn by interacting with each other. Each agent learns the relationships with other agents and cooperates with them to carry out tasks.
Example implementation: cooperative multi-agent system
import random
# Initial state of the agent.
agents = {
"AgentA": {"knowledge": set(), "history": []},
"AgentB": {"knowledge": set(), "history": []}
}
# Agents learn by exchanging knowledge with each other.
def agent_interaction(agent1, agent2):
# Knowledge sharing.
shared_knowledge = agent1["knowledge"] | agent2["knowledge"]
# Add new knowledge based on relationships
if "Task1" in shared_knowledge:
agent1["knowledge"].add("Skill1")
agent2["knowledge"].add("Skill1")
# Update exchange history
agent1["history"].append(f"Interacted with {agent2}")
agent2["history"].append(f"Interacted with {agent1}")
# Running agent-to-agent exchanges.
agents["AgentA"]["knowledge"].add("Task1")
agent_interaction(agents["AgentA"], agents["AgentB"])
# Display of results
print("Agent A Knowledge:", agents["AgentA"]["knowledge"])
print("Agent B Knowledge:", agents["AgentB"]["knowledge"])In this code, if AgentA knows ‘Task1’, then by interacting with AgentB, new knowledge called ‘Skill1’ is added to both agents. In this way, a system can be built in which agents mutually learn about each other’s relationships and grow together.
4. relationship-based decision-making AI
This is an example where the AI makes decisions not only on individual data, but also on relationships with surrounding data. For example, more flexible and adaptive AI systems can be achieved by using interdependent algorithms to make decisions in the context of multiple factors.
Example implementation: decision-making algorithm considering relationships.
# Decision-making functions depending on the situation.
def make_decision(context):
decision = ""
if "UserMood" in context and context["UserMood"] == "happy":
decision = "Suggest new challenge"
elif "UserMood" in context and context["UserMood"] == "stressed":
decision = "Offer relaxation advice"
# Consideration of relationships with other factors.
if "TimeOfDay" in context and context["TimeOfDay"] == "evening":
decision += " and share evening tips"
return decision
# examples showing the use (of a word)
context = {"UserMood": "stressed", "TimeOfDay": "evening"}
print(make_decision(context))The code makes flexible decisions based on the user’s situation (‘UserMood’ and ‘TimeOfDay’). By taking into account the user’s emotions and the relationship with the time of day, more appropriate actions can be presented.
References
References are discussed below.
1. quantum theory and relations of Carlo Rovelli
– Rovelli, Carlo. *Reality is Not What it Seems: The Journey to Quantum Gravity*. Penguin Books, 2016.
-Rovelli’s introduction to relational quantum mechanics and understanding reality. It details the theory of quantum gravity and the concept of relationality.
– Rovelli, Carlo. *The Order of Time*. Riverhead Books, 2018.
– This work describes an approach to interpreting the flow of time and reality from a relational perspective, which is also useful for understanding dynamic temporal change in AI and systems.
2. relationships and AI technology
– Russell, Stuart J., and Peter Norvig. *Artificial Intelligence: A Modern Approach*. Pearson, 2016.
– A comprehensive textbook dealing extensively with the construction of relational-based AI architectures and multi-agent systems.
– Wooldridge, Michael. *An Introduction to MultiAgent Systems*. Wiley, 2009.
– Suitable for learning about the design of multi-agent systems and how interactions between agents affect the overall system.
3. network theory and knowledge graphs
– Newman, M. E. J. *Networks: An Introduction*. Oxford University Press, 2010.
– It deals with the basic concepts of networks and relationships and helps to understand network structures such as knowledge graphs and Indra’s webs.
– Barabási, Albert-László. *Network Science*. Cambridge University Press, 2016.
– Exploring relationships and link structures from a network science perspective enables the generation of knowledge graphs similar to Indra’s web.
4. literature on Eastern thought
– Watson, Burton, trans. *The Vimalakirti Sutra*. Columbia University Press, 1997.
– Fundamental resource for a deeper understanding of the concept of the imperial network and its interdependence through the Buddhist scripture, the Vimala Sutra described in “Vima Sutra and Musho-Ho-Nin, the teaching of the Undivided“.
– Cook, Francis H. *Hua-Yen Buddhism: The Jewel Net of Indra*. Pennsylvania State University Press, 1977.
– There is an explanation of the concept of the imperial network of the Kegon sect and the concept of interrelationships, which provides inspiration for designing systems that incorporate a relational perspective.
5. decision-making and optimisation in AI systems
– Sutton, Richard S., and Andrew G. Barto. *Reinforcement Learning: An Introduction*. MIT Press, 2018.
– Learn about reinforcement learning algorithms that dynamically learn optimal behaviour based on relationships.
– Silver, David, et al. *”Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm.”* *Science*, vol. 362, no. 6419, 2018, pp. 1140–1144.
– Research on AI systems that make optimal decisions through self-interaction. It is also informative from the perspective of relational interaction.
コメント