AI System DoViews
DoView Outcomes Diagrams can be Used to Spell Out AI Systems Outcomes and the Steps They are Using to Achieve Them
Quick Overview
This page provides some examples of the use of DoView strategy/outcomes diagrams in regard to AI systems and agents. There is more discussion of this topic on the AI Systems Management page. The problem of understanding what AI systems are designed to do and how they will attempt to achieve their objectives is directly analogous to a long-standing problem in human organizational science: organizational interpretability. Similarly, ensuring AI systems pursue intended outcomes rather than drifting toward unintended behavior, the AI alignment problem, is functionally equivalent to the organizational alignment problem. Both require making implicit behavioral logic explicit, ensuring actions connect to stated outcomes, and maintaining accountability as systems scale and operate in varied contexts.
DoView strategy/outcomes diagrams were developed to solve these exact problems in human organizations. The methodology provides a standardized visual format for making "This-Then" behavioral logic transparent: what outcomes are being sought, what steps will be taken to achieve them, and how these connect causally. When AI systems' behavioral logic, whether embedded in training documents, reward functions, or constitutional principles, is represented as DoView outcomes diagrams following standard drawing rules, it becomes rapidly interpretable by humans, systematically comparable across systems, and verifiable for alignment with stated intentions.
The DoViews below provide some examples of how DoViews can be drawn for AI systems. In addition, there is an example of how a DoView strategy diagram can be used to plan the development of an AI system.
Only illustrative
The DoViews below are only illustrative use responsibly and at your own risk regarding IP issues and accuracy. Anyone can use the free AI DoView Drawing Prompt to create an AI-related DoView for any purpose. If you want help doing this, get in touch.
The AI Paper Clip Problem and How DoView Outcomes Diagrams Can be Used in AI Alignment
A classic thought experiment illustrates a problem in AI alignment known as the Paperclip Problem. The commonly understood version of this problem is that an AI system is given the job of making paperclips without any other specific guidelines for its outcomes. It then does everything possible to produce as many paperclips as it can, inadvertently using up all the world’s resources and eliminating humans in the process. This is why it is important to have ways of clearly specifying AI systems’ outcomes. The outcomes of any AI systems can be represented using a DoView outcomes diagram. To illustrate this, below are the DoView diagrams for a Bad Paperclip AI Agent and a Good Paperclip AI Agent. This issue is discussed in more detail in an Outcomes Theory Substack article.
A Real World Example - Claude’s Soul Document [/a050]
A document which outlined the outcomes being sought by Anthropic’s Claude was leaked. It set out the outcomes that Claude should seek and how it should go about seeking these. The Claude’s Soul Document DoView Outcomes Diagram below was created from the text version of the Soul Document. The DoView version is a lot easier to overview and has the potential of being able to be compared to the DoViews of other AI systems to assess the alignment and safety of such systems. This issue is discussed in more detail in an Outcomes Theory Substack article.
Using a DoView Outcomes Diagram for Planning AI Agent Development [/a031]
In the same way that DoView strategy/outcomes diagrams are used to plan anything else, they can be used to plan the development of AI agents. The example below is of using a DoView Strategy Diagram to plan the development of an AI agent swarm.
