Skip to content
Daniel Sosebee

Human Programming Writings and Definitions

Methodable, the human-programming environment

Writings on human programming

⭐ The ‘Meet Methodable’ program offers a good intro to human programming, as do the following posts from my human programming blog:

Also see The Origins of LIFE for discussion of a particular human program.

Definitions

Human-programming system

A human-programming system is a system that aids in the development and execution of abstract human workflows.

Human-programming language

A human-programming language is a formal language for specifying human programs. Human-programming languages are a superset of computer programming languages, because human-programming languages can define computational processes as well as human tasks and interactions.

Human-programming development environment

A human-programming development environment is a set of tools that aid in the creation and modification of human programs.

Human-program software runtime

A human-program software runtime is the software component of the human-program runtime (the full runtime includes the human user). The software runtime is responsible for navigating the program’s control flow and interfacing with the human user in order to deliver and monitor tasks.

Human programming for AI agents

'Ganch' art by Daniel Sosebee

I believe human-programming systems could be very effective for use by AI systems. We can think of an LLM as an AI task-completion engine, devoid of intrinsic executive-functioning ability. A well-constructed human-programming system could provide the executive functioning and reasoning capabilities necessary to allow an AI system to better accomplish difficult objectives.

I’ve started a research project amps.run to explore this intersection.

Ethics of advancing AI agent capabilities

Creation of superintelligent AI agents is an existential risk to humanity. I expect these kinds of agents to be built regardless. Therefore developing technology to help align AI systems with human values is critical.

I believe that AI-agent systems that rely on program execution for their executive-functioning tasks will be easier to align than systems which rely on the inscrutable matrices of deep learning models for their reward systems, because language-based systems will be easier to interpret and to govern. With that said I have every intention to proceed with care and to occasionally step back and evaluate the moral consequences of my work.

Assorted notes

Data collection

2024-02-22

I would like to give myself goals and then attempt to document the entirety of my thought and executive-functioning processes as I attempt to compete the goals. In doing so I will learn more about how doing things actually works. I can use these learnings to later program AI agents that do good executive functioning.