Despite the real value organizations can achieve through Artificial Intelligence (AI), many still struggle to address the risks associated with it.

In a global survey of risk managers, 58% identify AI as the biggest potential cause of unintended consequences over the next two years. Only 11% say they’re fully capable of assessing risks associated with organization-wide AI adoption.

Bias, discrimination, fairness, and explainability are areas of paramount concern. And while there are some specific definitions for these problem areas, translating them into action involves tough decisions and application-specific constraints.

To answer these questions, we spoke to Responsible AI practitioners (data scientists, AI engineers, lawyers, industrial/organizational psychologists and project managers) from 19 organizations, across four continents. Our analysis indicates that some organizations have struggled to develop a systematic internal approach to convert principles into practice. And our experience shows this is because they underestimate the technical complexity and scale of people and process change required.

Organizations need to tackle a central challenge: translating ethical principles into practical, measurable metrics that work for them. To embed these into everyday processes, they also need the right organizational, technical, operational, and reputational scaffolding. Based on our experience delivering Responsible AI solutions to organizations worldwide, we’ve defined four pillars of successful Responsible AI implementations. For each pillar, we’ve summarized key practitioner pain points, recommended how to address them, and included a case study showing what good looks like.

Democratize new ways of working and facilitate human+machine collaboration.

Practitioner pain points: Practitioner interviewees highlighted the need for appropriate performance metrics and recognition for their work. They indicated that too little value was placed on risk mitigation, including prevention of reputational harm. And time pressures often mean short-term product success was prioritized by leadership over Responsible AI’s long-term benefits.

Set up governance and systems that enable AI to flourish.

Practitioner pain points: Interviewees indicated that companies struggle with stakeholder misalignment, bureaucracy, conflicting agendas, and a lack of clarity on processes/ownership. Individuals often operate ad hoc, following their own values and assessments. Insufficient clarity on governance and accountability, unnecessary conflicts, and competing incentives across groups ultimately led to Responsible AI inertia and a reactive mindset.

Make systems and platforms trustworthy and explainable by design.

Practitioner pain points: One of the biggest barriers that practitioner’s organizations faced was a lack of expertise in defining and measuring ethical use and algorithmic impact of data, models and outcomes. Without established technical methods to monitor and mitigate these risks, organizations can’t be confident that a system is fair. Companies struggle to integrate academic metrics like algorithmic fairness. Responsible AI cannot be measured in revenue generation or click-thru rates, but many businesses still rely on these traditional performance metrics to track practitioners’ success.

Articulate the Responsible AI mission and ensure it’s anchored to company values and ethical guardrails.

Practitioner pain points: Without the right organizational, operational, and technical foundations, interviewees reported that organizations are forced into reactive approaches to Responsible AI, which expose them to significant reputational damage.

Successful organizations take a systematic approach to Responsible AI from the start. They understand the scale and complexity of change required. And they address challenges in parallel.

Like this, they move from principles to practice with confidence, support the professionalization of AI enterprise-wide, and put in place the structures needed to prove the long-term value of Responsible AI.

Our conversations identified a range of organizational, operational, technical and reputational challenges that hold well-intentioned organizations back. While the initial focus is often on ethical and legal requirements, success is also a function of an organization’s ability to modify its traditional ways of working to support Responsible AI—and AI more broadly. In undertaking this process, organizations also establish the structures needed to demonstrate the long-term value of Responsible AI by scaling it across the organization and enable the essential move from “practice to proof.”

We use a set of 25 questions to help our clients to benchmark their motivators and challenges, together with their maturity in terms of people, process and technology against their peers. Where are you on your Responsible AI journey?