INITIAL RESEARCH AGENDA

Initial Projects

Initial research concepts

DPIM is examined under the hypothesis that it may contribute, in the first instance, to reducing harm caused by wrong judgments, and KAC under the hypothesis that it may contribute, in the first instance, to tracing evidence and execution results and to responsible operation. To say that the two projects also contribute to human capability or to gaps in use, a separate evaluation that includes people and the field is needed.

Two research agendas

What do the two projects ask?

RESEARCH AGENDA I

DPIMDecision Process Intervention Model

Researching where to intervene in the judgment process

A lens examining the point where several transparent, branching paths meet

Observing the stateReading the state of the judgment process
Choosing an interventionIncluding no intervention
Evaluating the effect of the interventionRecording results · time · cost · errors

Meaning

DPIM is short for Decision Process Intervention Model, meaning ‘a model for intervening in the judgment process.’

A research agenda that seeks to read the observable states in the judgment process of an AI team and to learn where, and with what kind of intervention, results improve.

What is looked at separately

Diagnosing a state and choosing an intervention should be kept distinct. Because a diagnosis that finds a problem well does not always lead to a useful intervention, the comparison should include cases with no intervention.

Read more — Designing the initial experiments

The initial experiments of DPIM can be designed so that several options are compared separately from the same state of progress. Required evidence can be requested, the current evidence can be checked again, other alternatives can be reviewed, or the process can be allowed to continue as it is. Along with the final result, the time and cost the intervention used and any new errors are recorded. The effects and conditions of application of DPIM are verified through such comparative experiments.

RESEARCH AGENDA II

KACKnowledge Action Chain

Connecting knowledge, execution and verified results

A thin green band that links three glass panes at different depths and loops back

Knowledge and methodsConnecting the knowledge needed
Execution recordsLeaving outputs
Verifying resultsDistinguishing outputs from confirmed outcomes

Meaning

KAC stands for Knowledge Action Chain, that is, ‘the connection between knowledge and action,’ and is a research concept that deals with how knowledge leads to execution and to verified outcomes.

What is looked at separately

The knowledge, methods and skills needed for goals and tasks are organized and connected with actual execution records, outputs and results, while the fact that an output was produced is kept distinct from the fact that the intended outcome was confirmed. When reviewed experience is reused, the conditions under which it can be applied should be recorded with it.

Read more — Outputs and confirmed outcomes

In the purchase review case, the creation of a report file counts as an output. Whether the report meets the requested specifications and the applicable rules and deals with the necessary exceptions needs separate verification. Improvements obtained through verification can be reflected in knowledge and procedures within the approved scope and tested again. When KAC is studied, it can be examined, in comparison with a system that lacks this connection, what differences arise in tracing evidence, confirming actual completion and reusing experience.

Criteria for choosing projects

In what order are projects chosen?

A meaningful question.
A testable proposition.

First
The difficulty to reduce and the change to check
It should be made clear whose difficulties, and which ones, are to be reduced and how that change is to be checked. A priority goal is chosen from among the four social goals, and the experience and conditions of use of the people affected are checked.
Then
Scholarly grounds and verifiability
Whether a problem is scholarly important and whether it can be verified are reviewed. What failures recur in the field, what explanations the relevant scholarship offers, and whether the materials and environment needed for comparative experiments can be put in place are looked at. Separating problems that are already well solved from problems that still lack explanation helps in deciding where to use research resources.

The other agendas of the Scholars’ Council for AI are chosen by the same criteria, that is, the problem, scholarly grounds and verifiability. Agendas such as learning support in education, users’ requests for review and the conditions of use in small organizations can also be chosen according to the same social goals, and whether to adopt a particular technology is judged by whether it contributes to achieving the goals.

See the change the Scholars’ Council for AI seeks

Comparative research design

When the two projects are dealt with together, what is looked at separately?

Research that deals with both projects together is also possible, and it can be designed so that KAC records the relationship between knowledge and execution and DPIM examines intervention in a particular state of judgment. Whether the two concepts can be combined is reviewed together with independent research on each.

A design that compares four conditions side by side

Comparison condition	Intervention according to state	Connection between knowledge and execution results	Effect to check
Baseline condition	None added	Existing method kept	Current results, failures and costs
DPIM applied	Chosen, including no intervention	Existing method kept	Added effect of choosing an intervention
KAC applied	Existing judgment procedure kept	Source, execution and verification records connected	Differences in tracing, confirming completion and reuse
DPIM and KAC combined	Chosen according to state	Connected through to interventions and execution results	Differences beyond each one’s effect, and added cost

This is a hypothetical comparative design. The comparison is made under the same task, material and model conditions and a stated budget, and added costs are also recorded. It is not assumed in advance that the combined condition is the best.

A research structure through which DPIM and KAC can be connected

DPIM

Observing the state
Choosing an intervention
Evaluating the effect of the intervention

KAC

Knowledge and methods needed
Evidence and conditions applied
Execution records and outputs
Confirmed results and limits

This is the functional relationship in the initial research concepts. DPIM chooses and compares interventions in an observed state of judgment, and KAC connects the knowledge needed, execution and verification results. The effect of each and the added effect of combining them are verified separately.

Explore the research principles, the faculty and joint research too

See the research principles See the faculty

Collaborate