3 ‘states of consistency’ every AI program needs to succeed

Of 52 AI models analyzed by MIT’s Center for Information Systems Research over the past two years, 31 have been deployed. The rest were in pilot phases or under development.

The authors of a research briefing on the analysis found that organizations taking AI projects from raw concept through meaningful deployment achieve three “interdependent states of consistency” while operating in the midst of “dynamic, changing forces.”

The three states:

  • Scientific consistency, which produces an AI model that can generate accurate outcomes;
  • Application consistency, which creates an AI solution that can achieve goals in situ over time; and
  • Stakeholder consistency, which generates AI benefits across a network of people with a keen interest in the system’s success.  

The authors propose that organizations pulling off the triple feat are, knowingly or not, demonstrating an adaptive management approach called AI alignment.

They describe several real-world examples of such success, noting that the rewards of AI alignment are not always measurable in monetary terms.

The report was released in November but refreshed in a new summary from MIT’s Sloan School of Management Jan. 6.

Summary here, research briefing here.