Stakeholder Participation is Key to Successfully Leveraging Artificial Intelligence (AI) and Machine Learning (ML)

According to Daniel Sanchez-Reina, VP Analyst at Gartner, during his presentation at the IT Symposium/Xpo 2022, October 17-20 in Orlando, "The pressure on CIOs to deliver digital dividends is higher than ever. CEOs and boards anticipated that investments in digital assets, channels and digital business capabilities would accelerate growth beyond what was previously possible. Now, business leadership expects to see these digital-driven improvements reflected in enterprise financials."

Respondents to the the 2023 Gartner CIO and Technology Executive Survey Daniel was quoting ranked their executives’ objectives for digital technology investment over the last two years. The top two objectives were to improve operational excellence (53%) and improve customer or citizen experience (45%). In comparison, only 27% cited growing revenue as a primary objective and 22% cited improving cost efficiency.

“CIOs must prioritize digital initiatives with market-facing, growth impact,” said?Janelle Hill, Distinguished VP Analyst, Gartner. “For some CIOs, this means stepping out of their comfort zone of internal back-office automation to instead focus on customer or constituent-facing initiatives.”

The survey further revealed that CIOs’ future technology plans remain focused on optimization rather than growth. CIOs’ top areas of increased investment for 2023 include cyber and information security (66%), business intelligence/data analytics (55%), cloud platforms (50%) and (32%) are increasing investment in artificial intelligence (AI) and machine (ML).

One of the key components of success that CIOs, who are under pressure to deliver on the promise of Digital Transformation, may want to consider to reduce risk and increase time to value, is the "human factor". Introducing transformation technology such as AI and ML into an organization without addressing the misunderstandings, fears and continued support of stakeholders is a formula for failure.

According to a Research Briefing published on the MIT Center for Information Systems Research website on July 21, 2022 by Ida A. Someh, Barbara H. Wixom, and Cynthia M. Beath (Dr. Boo), entitled Building AI Explanation Capability for the AI-Powered Organization, "Organizations must strengthen and draw on five advanced data monetization capabilities—data science, data management, data platform, customer understanding, and acceptable data use—to deploy AI?solutions that are operationally, economically, and ethically sound.?Yet these five capabilities alone are insufficient for building stakeholders’ confidence in an AI solution. AI teams must also regularly engage with a variety of stakeholder groups to explain how the organization is managing characteristics of AI that impede trust in AI solutions.

According the MIT Research Briefing, the four characteristics of AI make it difficult for AI project teams to build stakeholder trust in AI solutions:

• Unproven value:?Because the market still lacks a broad selection of proven AI use cases, leaders are uncertain if and how their company could create meaningful returns from AI project investments.
• Model opacity:?AI draws upon computational mathematics and statistics that, due to their sophistication, make it very hard for even some data scientists (let alone laypeople) to readily understand if a model is producing accurate results in ways that are compliant and ethical.
• Model drift:?An AI model will produce results that are biased in the same way as the data used to train it. Over time, as the real-world changes, or if error is introduced into data sources, the accuracy of the predictions an AI model produces from new data will deviate from the model’s performance during the training period—a phenomenon known as “drift.”
• Mindless application:?AI models evaluate a new case and draw conclusions about it with some degree (or lack) of confidence. Because there is some level of error in transferring learning from historic case evaluations used for model training to new cases, AI model results are not definitive; applying the results as if they were definitive to new cases or in new contexts can be risky.

This group found that organizations that adopt practices that resolve problems arising from these challenging AI characteristics successfully build stakeholder confidence.

• Value formulation practices?tackle the problem of unproven value by helping AI teams substantiate, in terms that appeal to both direct and indirect stakeholders, exactly how AI can be an attractive investment.
• Decision tracing practices?contend with model opacity by helping AI teams unravel model computations and mathematics, to convey the how behind model results to those who consume and make use of the output.
• Bias remediation practices?enable AI teams to redirect model drift by exposing model decisions, helping the teams to reduce bias in a model’s training activities throughout its life.
• Boundary setting practices?address mindless application by guiding AI teams in applying AI model output mindfully to avoid unexpected outcomes and unintended consequences.

Collectively, these four sets of practices build an organization’s AI Explanation (AIX) capability, which you can define as?the ability to manage AI initiatives in ways that ensure models are value-generating, compliant, representative, and reliable.The AIX capability is multidimensional. It includes the ability to articulate a model’s value for different stakeholders, to make the inner workings of models understandable, to create explanatory interfaces to expose and enable rectification of biases in AI model outputs, and to set boundaries for the safe application of AI models.

Unlike the five data monetization capabilities, which are well understood and formalized, practices that build AIX capabilities are still emerging. In fact, most AI project teams are creating their playbook as they go. To establish AIX as an enterprise capability, it's crucial that an organization surface and share good practices, which takes effort and leadership. New, effective AIX practices should be seen as building blocks for an enterprise AIX capability. With this capability in hand, projects can stop crafting practices from scratch.

Therefore, organizational leaders must seek to proactively develop an AIX capability. A good starting point is to identify units and other organizations that are already driving effective explanations. Next, identify practices that the organization’s own AI project teams have invented or adopted to create AI models that are value-generating, compliant, representative, and reliable, and enable sharing and reuse of these practices across the company. Finally, continue testing the most promising practices, and institutionalize the best ones. These actions will go a long way toward making AI trustworthy and consumable in an AI-powered organization.

Understanding and addressing the Organizational Change Management (OCM) involved in leveraging AI/ML within the enterprise is paramount to success. However, OCM is not just required for complex and mysterious technologies like AI/ML. According to an article published by Eric Kimberling on October 8, 2021, entitled, Top 5 Human Factors in Digital Transformation, "Organizational Change Management and the human side of change is by far the most significant piece of any digital transformation puzzle. If you address the human factor well, you're likely to succeed. If you don't address it well, you're highly likely to fail."