Axiomyx Decision Lab

Human-machine experimentation for complex decisions

Axiomyx Decision Lab

Build synthetic environments, introduce uncertainty, compare courses of action, and study how people and machine decision systems respond to the same situation.

What Decision Lab is

Decision Lab takes the same decision architecture that plays games in Arena and points it at synthetic decision environments instead. The environment stops being a game board and becomes a scenario: resources, timing, access, information quality, competing objectives, and an opponent that responds to what you actually did.

The machine is not there to produce an answer. It is there as a second decision-maker that can be run repeatedly, questioned, and compared against the people in the room.

Intended uses

  • Wargaming and scenario design
  • Professional education and structured training
  • Simulation and research
  • Human-machine experimentation
  • Course of action exploration and comparison
  • Decision-support analysis
  • Adaptive opposing behaviour in an exercise
  • Repeated experimentation on one starting situation

What this is not

Decision Lab is a research, education and experimentation environment. It is not an autonomous command system. It does not direct real activity, it is not fielded software, and it has no role in any real operation.

No organisation currently uses Decision Lab. No government, military, academic or commercial body has adopted, accredited or endorsed it, and nothing on this page should be read as suggesting otherwise. It is described here because it is being built, not because it has been deployed.


Worked example

A synthetic environment, in the abstract.

This is an invented scenario used to show the shape of the problem. It is not a real plan, it uses no real disposition of anything, and it contains nothing operational.

The situation

A distributed organisation has to keep access to several important nodes while working with limited resources, degraded communications, changing access to routes, uncertain information about the current picture, and an opponent that adapts to whatever it does.

There is no clean answer. Committing resources to hold one node makes another harder to reach. Waiting for better information costs time that cannot be recovered. Choosing the route that is reliable today assumes it is still available tomorrow, and the opponent gets a say in that.

This is the class of problem Decision Lab exists to study: too many interacting variables to hold in your head, an outcome that depends on sequence rather than on any single choice, and an adversary who is not standing still while you decide.

Variables the environment defines

What gets measured

  • Which nodes were held, and for how long
  • What the decision cost in resources and in time
  • How the plan behaved when information turned out to be wrong
  • How it behaved when the opponent did something unexpected
  • Where the human choice and the machine choice diverged

Human and machine

Another decision actor, not a replacement.

The point is not that the machine decides better. It is that it decides differently, it can be run a hundred times, and the difference between the two is informative.

Human team

  • Experience
  • Doctrine and training
  • Judgement about what matters
  • Context the scenario does not encode
  • Intuition, including the kind that is right for reasons nobody can articulate

Axiomyx

  • State evaluation
  • Candidate search across many futures
  • Simulation of what follows a choice
  • Value estimation on a consistent basis
  • Repeated experimentation without fatigue or preference

Compare results

Run properly, this gives you four things a workshop on its own does not: alternatives nobody raised, assumptions made visible because the machine did not share them, an opponent that responds rather than agreeing, and a record of where human and machine reasoning came apart. That last one is usually the most useful, in both directions.


Method

A workable experiment.

This is the shape a Decision Lab experiment is intended to take. It is a research method, not a product feature list.

  1. Define the environment

    Build the synthetic situation: the nodes, the resources, the clock, the information quality.

  2. Define state and actions

    Decide what the participants and the system can see, and what they are allowed to do about it.

  3. Define objectives

    State what a good outcome is, in terms that can actually be measured at the end of a run.

  4. Present it to both

    Give the same starting situation to the human participants and to Axiomyx, without either seeing the other's choice.

  5. Record decisions

    Capture what each side chose, when, and on what information.

  6. Introduce uncertainty

    Degrade information, close a route, or let the adaptive opponent respond to what was actually done.

  7. Repeat

    Run it again, and again, varying one thing at a time. Repetition is what turns an anecdote into a result.

  8. Compare

    Look at outcomes and at decision patterns, not only at who scored better.

  9. After-action review

    Work through where the two sides diverged and what each was assuming when they did.

Working on decisions like these?

Register and pick Decision Lab, or partnership and research collaboration. Concrete use cases shape what gets built, and this is early enough that they still can.

Decision Lab interest