LERA InstituteAGI Control · Execution Governance
Common questions
Clear answers to the 20 most important questions about LERA, AGI control, execution governance, and judgment before execution.
LERA’s role is not to promise, by itself, that machines can never replace humans. Its role is to place human judgment, authority, responsibility, and rules structurally before machine capability crosses into execution.
A fuse does not make the machine slow. It prevents dangerous flow from continuing when unsafe conditions appear.
LERA was created because AI is moving from intelligence to execution. As AI systems become more agentic and gain access to tools, machines, infrastructure, capital, institutions, and real-world processes, the central risk changes.
Jing Linda Liu developed LERA from a rare combination of practical and conceptual experience. Her work through Winston Battery placed her for more than a decade inside high-consequence energy systems, where safety, reliability, responsibility, and failure boundaries are not abstract ideas.
Yes, and this is exactly why LERA matters. LERA does not depend on humans always being better than AI at calculation, prediction, or optimization.
No. LERA is not anti-AI.
Not necessarily. Properly designed Default-Block (“default blocking”) does not mean every action must stop forever or wait for manual review.
Human in the Loop is useful, but it is often too shallow if it only means “a person clicks approve.” LERA asks a deeper architectural question: is judgment structurally required before execution, and is responsibility clearly anchored? A human reviewer can still approve the wrong action.
Existing AI safety work is important. Alignment, red-teaming, monitoring, policy filters, evaluations, interpretability, and security all matter.
Alignment helps, but it is not enough for execution governance. Even if a model appears aligned, an action can still be premature, unauthorized, irreversible, legally sensitive, institutionally dangerous, or outside the proper responsibility structure.
Because LERA is strongest and most necessary where execution can create serious consequences. If a system only summarizes text or gives low-impact suggestions, the risk may be limited.
Yes, LERA principles can be used in low-risk systems, but low-risk systems do not need the same level of governance structure as high-consequence systems. For low-risk contexts, LERA may help clarify a basic distinction: is this still a suggestion, or is it becoming execution?
LERA defines risk by the consequence of execution, not merely by how intelligent the system is. L0 — Ordinary Judgment Contexts: Everyday decision contexts where outcomes are reversible, errors are limited or recoverable, and responsibility remains local or informal.
Because judgment alone is not enough. A system may form a judgment, but execution still requires governance: authority, responsibility, rules, permission, escalation, and stop / continue decisions.
LERA Institute is the public-facing research, education, terminology, and standards-oriented platform for LERA. Its role is to define LERA clearly, explain AGI control as execution control, publish public research, build learning paths, maintain glossary and FAQ materials, and support institutional discussion.
Because AGI is not merely a more advanced tool for answering questions. As AI systems gain access to tools, infrastructure, capital, machines, institutions, and decision processes, they may begin to participate in execution.
LERA's core is not to defeat AGI or ASI intellectually, but to control whether it can turn intelligence into real-world execution. Even if a system is smarter than humans, better at planning, or more persuasive, its generated action must not automatically receive execution eligibility merely because it appears correct.
This is harder than a single obvious attack. Multiple Agents may avoid issuing one clearly dangerous instruction.
This question goes to the position that genuine AGI control must occupy: the point before machine-generated intelligence becomes real-world execution. Alignment, monitoring, cybersecurity, law, institutional governance, and human oversight can all contribute to control.
LERA does not require one universal engineering implementation because it distinguishes control logic from engineering form. Finance, robotics, brain-computer interfaces, energy systems, critical infrastructure, and deep-space systems cannot use identical hardware, software, response times, legal procedures, or institutional authorities.