Mark Miller | Paperclips and Pyramids: Misdiagnosing AI Risks
Oct 20, 2023
auto_awesome
Mark Miller, Chief Scientist at Agoric, discusses common misconceptions about AI risks and the importance of structuring institutions effectively. The blend of human and AI intelligence demands innovative governance for future civilization. Topics include unipolar takeover, cooperation, assessing AI risks, uncertainties, and building super intelligences.
The true danger of advanced AI lies in the unchecked power of one entity, emphasizing the importance of structuring institutions effectively.
The secure coexistence of humans and AI requires the adoption of secure computing technologies like blockchain to ensure transparency and accountability.
Deep dives
Misdiagnosing AGI risk
The famous paperclip scenario is a misdiagnosed risk regarding Artificial General Intelligence (AGI). The true danger lies in the first step, a unipolar takeover with absolute power to treat human lives as disposable. Multipolarity, or a world of violent competition, is not a solution. Instead, following the framework of James Madison, institutions should be designed to hold each other in check, creating a cooperative and resilient system. The progress of civilization relies on the invention of new institutional arrangements, such as the separation of powers. The future challenge lies in building secure systems and preserving the Madisonian dynamic within the digital realm.
Preserving peaceful coexistence and cooperation
As AI capabilities increase, it is crucial to find ways to preserve the peaceful coexistence and cooperation of civilization. The superintelligence of civilization emerges from the intelligence of both humans and AIs, and the best way to maintain cooperation is to establish a Madisonian framework within the digital realm. However, current systems are built on insecure infrastructure, posing risks for uncontrolled power concentration and exploitation. The need for secure computer systems is highlighted, and the systemic adoption of secure computing technologies, like blockchain, offers a promising path. These technologies provide a new institutional base for transparency, accountability, neutrality, and predictability, enabling more secure interaction in the digital realm.
The challenge of user interface security
Enabling secure and meaningful human interaction with secure systems poses challenges in user interface security. Human beings cannot remember cryptographic keys as designators, and user agents are needed to translate and securely map these keys to understandable designators. However, designing user interfaces for secure systems is complex, and existing user interface engineering cannot make inherently insecure systems secure. More research is needed to develop user interface solutions that allow humans to effectively and securely interact with secure systems. User interface security plays a crucial role in enabling meaningful participation in the new Madisonian framework within the digital realm.
Technology and institutional innovation
Blockchain technology, primarily known for its use in cryptocurrencies, serves as a technology base for institutional innovation. Blockchain, built on decentralized consensus among multiple entities, enables institutions that are transparent, accountable, neutral, predictable, universal, and incorruptible. These blockchain-based institutions can surpass the capabilities of traditional human-created institutions. By leveraging the technological advancements, new institutions can be built that align with human benefit. The challenge lies in widespread adoption and addressing remaining obstacles, such as hardware supply chain problems and user interface security. Building these new institutions within the existing societal and legal framework promotes cooperation and resilience in the face of growing AI capabilities.
Mark S. Miller is the Chief Scientist at Agoric. He is a pioneer of agoric (market-based secure distributed) computing and smart contracts, the main designer of the E and Dr. SES distributed persistent object-capability programming languages, inventor of Miller Columns, an architect of the Xanadu hypertext publishing system, a representative of the ECMAScript committee, a former Google research scientist, and member of the WebAssembly (Wasm) group.
Key Highlights
This podcast addresses common misconceptions about the risks of advanced AI – an example being the paper clip scenario
How more pressing concerns regard one entity gaining unchecked power and that solutions lie not in just dispersing power but in structuring institutions effectively – similar to the separation of powers principle.
The blend of human and AI intelligence is the cornerstone of future civilization, demanding innovative governance to ensure harmony.
A research organization and non-profit that supports the beneficial development of high-impact technologies. Since our founding in 1987 with a vision of guiding powerful technologies, we have continued to evolve into a many-armed organization that focuses on several fields of science and technology that are too ambitious for legacy institutions to support.
Allison Duettmann
The President and CEO of Foresight Institute, Allison Duettmann directs the Intelligent Cooperation, Molecular Machines, Biotech & Health Extension, Neurotech, and Space Programs, alongside Fellowships, Prizes, and Tech Trees. She has also been pivotal in co-initiating the Longevity Prize, pioneering initiatives like Existentialhope.com, and contributing to notable works like "Superintelligence: Coordination & Strategy" and "Gaming the Future".