Ensuring AI Safety with Ontological Anti-Crisis

Ensuring AI Safety with Ontological Anti-Crisis

Table of Contents

1. Introduction
2. Ontological Crisis
   1. Definition
   2. Examples in Machines and Humans
   3. Comparison
3. The Ontological Anti-Crisis
   1. Assumptions about Ontological Development
   2. Ontological Hierarchy
   3. Completeness and Incompleteness
4. The Center for Safe AGI
   1. Introduction to the Organization
   2. Interdisciplinary Team
   3. Connections with Universities, Governments, and Companies
5. Air Safety Research and the Donation of Massive Compute
6. Using Ontological Crisis to Understand Machines
   1. Tools to Design and Develop Ontological Structures
   2. Using OAC for Detecting Incompleteness
7. Ensuring Safety in AGI Development
   1. Gradual Change and Robustness
   2. Efficient Detection of Incompleteness
   3. Making AGI Understandable, Traceable, and Beneficial
8. Comparative Perspectives
   1. Ontological Security in Politics and War
   2. Lessons from Legal Interpretations and Contracts
   3. Coordination and Prediction Markets
9. Challenges in Aligning Systems Smarter than Humans
   1. Addressing General Difficulties
   2. Distinction between AGI and Advanced AI
   3. Creating a Hierarchical Structure of Safety

Ontological Crisis: Shaping AGI Development for a Safer Future

In the ever-evolving landscape of artificial general intelligence (AGI) development, the concept of an ontological crisis plays a crucial role. This crisis refers to the challenge an agent, be it human or machine, faces when its model of reality undergoes a significant change. Peter de Blanc first brought Attention to this problem, highlighting its universal nature. When an agent upgrades or replaces its ontology, it can lead to a crisis where its original goals are ill-defined with respect to the new ontology.

For instance, self-driving cars experience an ontological crisis when their world models change, making them prone to accidents. Humans also encounter ontological crises, such as a loss of faith in God or a shift from fundamental physics to quantum physics, leading to changes in actions or decision-making processes. Anders Sandberg argues that the worry of radical changes affecting our values and identity signifies a form of ontological crisis in humans.

While machines can detect and address ontological crises relatively easily, resolving the crisis in humans is more complex due to biological constraints and other restrictions. In contrast, fixing ontological crises in machines may be more straightforward. This brings the focus to the concept of the ontological anti-crisis (OAC) and its potential role in shaping the development of AGI.

The OAC approach acknowledges the fundamental urge for ontological development and the inner process of individuals' ontological development, which may not be observable. By using a Simplified ontological hierarchy for machines and humans, it becomes possible to examine the structure and dynamics of ontological development. The two key principles of completeness and incompleteness Shape the ontological structure. AGI ontologies should aim for completeness, mirroring humans' ontology to ensure survival, while also embracing incompleteness to facilitate exploration.

To develop AGI with enhanced safety, the Center for Safe AGI (CSA) has emerged as a non-profit organization in China. CSA boasts an interdisciplinary team with strong connections to universities, governments, and local companies. Recently, they received a substantial computer donation, which they plan to utilize for air safety research. The organization views OAC as a tool to make AGI understandable, traceable, and beneficial. By using finite factor sets and robust frameworks like Scott's from MIRI, CSA designs open ontological structures that can detect incompleteness, handle phase transitions, and safely interpolate models.

By leveraging OAC, the CSA aims to overcome challenges related to aligning systems smarter than humans with human interests on a global Scale. Eliezer Yudkowsky's discourse networks highlight the difficulties of aligning systems with different purposes. However, the distinction between AGI and advanced AI allows for different standards of safety and hierarchical structures to address these challenges. By actively tracking and discussing problems related to safety and clarifying them, progress can be made in ensuring the safe development of AGI.

In conclusion, the concept of ontological crises exposes the need for a proactive approach to AGI development. The OAC framework, supported by the CSA, embraces the complexity of ontological crises and proposes strategies to detect, address, and overcome them. By understanding the nuances of ontological crises, drawing insights from various disciplines, and promoting safety standards, AGI can be developed in a manner that aligns with human values and secures a safer future.