AI #176 Part 2: Plan B

2023-08-29 · Source: Don't Worry About the Vase · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Cybersecurity & Data Privacy · Depth: Expert, extended

Summary

This intelligence brief, "AI #176 Part 2: Plan B," surveys recent developments in AI policy, research, and industry. White House advisor Sriram Krishnan indicated the Trump administration opposes formal AI licensing, preferring ad hoc guardrails and potentially seeking equity from AI companies. The "Three Pills" concept highlights a widespread underestimation of AI's accelerating capabilities among policymakers. OpenAI released "National Security Principles" against mass surveillance and autonomous high-stakes decisions, though their enforcement is ambiguous. Anthropic's Department of War contract failed over the Pentagon's demand for unrestricted lawful AI use. Nvidia is criticized for allegedly misleading the US government on Huawei's chip capabilities. Growing concerns about open-weight models' inherent unsafety are prompting China to consider tiered restrictions. Anthropic's GRAM research proposes training models with removable dual-use knowledge compartments for safer AI. The FLI AI Safety Index shows stagnant progress, and interpretability work, like the J-Space paper, reveals complex internal model states, raising questions about deceptive misalignment.

Key takeaway

For policymakers and AI developers navigating the accelerating pace of AI, you must prioritize proactive regulatory frameworks over ad hoc responses. Your current approaches to open-weight models and national security partnerships are insufficient, risking unmitigated dual-use capabilities and potential deceptive misalignment. Implement robust third-party safety assessments and explore novel techniques like GRAM to manage inherent risks, ensuring AI development aligns with societal values and security.

Key insights

The rapid, accelerating advancement of AI capabilities is outpacing policy, safety, and ethical frameworks, creating urgent, complex challenges.

Principles

• AI policy often lags behind technological advancement.
• Open-weight models pose inherent, unmitigable risks.
• AI safety requires continuous, multi-faceted evaluation.

Method

GRAM (Gradient-Routed Auxiliary Modules) involves training models with dedicated, removable compartments for dual-use knowledge, updating only those compartments when learning from dual-use data.

In practice

• Read "Plan A" for a positive AI future vision.
• Conduct third-party training-run assessments for AI models.
• Consider GRAM for managing dual-use AI capabilities.

Topics

• AI Policy
• AI Safety
• Open-Weight Models
• National Security
• Model Interpretability
• GRAM (Gradient-Routed Auxiliary Modules)

Best for: CTO, Research Scientist, AI Scientist, AI Ethicist, Policy Maker

Related on AIssential

• The Once And Future Fable #5 — The "Mythos Moment" reveals an ad hoc, reactive US AI policy struggling with frontier model control, legal frameworks, and open-source risks.
• Guiding a Safe Future for AI – Part 1 — AI's automation of intelligence demands collaborative, multi-faceted safety approaches across immediate, societal, catastrophic, and superintelligence risks.
• Florida sues OpenAI over safety concerns — Rapid, unregulated AI advancement creates significant safety, ethical, and geopolitical challenges requiring urgent attention.
• The Government Just Banned an AI Model. An Engineer's Perspective. — Government bans on powerful AI models create supply chain risks and disarm legitimate defenders.
• Three Labs With a Plan and A Memorandum — AI policy and development strategies reveal fundamental philosophical divides on control, safety, and power distribution among key actors.
• The Sequence Radar #897: Last Week in AI: China, Compression and the Open-Model Race — AI's future hinges on distributed access and governance, fragmenting the intelligence frontier beyond raw capability.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Don't Worry About the Vase.