Businesses need confidence that the mission-critical decisions and outputs generated by their AI solutions are trustworthy and reliable. They must consider if the AI is securely working only off their proprietary data, address the related cybersecurity concerns, and account for the environmental impact of training large language models (LLMs). They need to understand how their system reached a decision, addressed compliance with business and regulatory requirements, and mitigated biased or inflammatory behavior.
The criticality of trust is even clearer today as generative AI is changing the landscape of what's possible with AI. For example, generative AI models can produce highly believable, well-structured responses so it can be hard to identify an incorrect response without the right subject matter expertise. Generative AI does not have test-retest reliability, meaning you can ask the same question many times to an LLM and get different answers each time. It does not understand meaning, which only exacerbates these risks.
Many business leaders are feeling pressure from investors, clients, employees, and others to accelerate their adoption of AI, including generative AI, but they're weighing the urgency to act against the potential risks. In a recent IBM Institute for Business Value report titled "CEO decision-making in the age of AI," three-quarters of surveyed CEOs reported they believe that competitive advantage will depend on who has the most advanced generative AI, but 48% of CEOs surveyed worry about bias or data accuracy.
IBM Consulting's experience helping clients in every industry and geography scale AI has shown us that building trust in AI in the enterprise requires a socio-technical approach that spans people, process, and technology.
Here are three actions leaders should consider to help drive trust as they scale AI:
1. Don't just publish your principles — operationalize them
Fairness, explainability, privacy, robustness, and transparency are IBM's pillars for trustworthy AI. These are powerful and simple words to frame the functional and non-functional requirements for AI models, but complicated to bring to life.
For one example, consider transparency. Users should be able to see how the AI service works, evaluate its functionality, and comprehend its strengths and limitations. This helps a user of the model determine whether it is appropriate for their situation.
Transparency reinforces trust, and the best way to promote transparency is through disclosure. Operationalizing that means ensuring AI systems share information on what data is collected, how it will be used and stored, and who has access to it. The systems should make their purposes clear to users.
2. Establish AI governance to embed the discipline of AI ethics
It's easy to say that ethics matter, but embedding those ethical principles into your operations requires governance. AI governance is how you define policies and establish accountability throughout the AI lifecycle. It's important to establish organizational structure, policies, process, and monitoring. An AI Ethics Board can be a natural catalyst. At IBM, we embed our Principles for Trust and Transparency across our company through an AI Ethics Board which provides centralized governance and holds our company and all IBMers accountable to our values.
3. Focus on culture and cross-functional collaboration
The goal should be to have AI better reflect the communities that it serves. But making sure that systems of inequality aren't perpetuated in AI requires establishing the right organizational culture to curate AI responsibly.
The teams building AI tools and processes should represent multiple disciplines, skillsets, and lived experiences. Consider bringing in not only engineers and data scientists but designers, legal professionals, and the anticipated users themselves. Collaborative methods, like the IBM Garage, are designed to help organizations with the framework to fast track innovation and drive meaningful, lasting transformation that puts the user at the center and helps mitigate bias. Establishing systems that solicit continuous feedback on AI-driven tools and processes can also help drive better outcomes. All of these elements can help produce outcomes that garner trust from users.
AI can be a powerful tool to fuel business transformation, but successful implementation can depend on AI being used in a trustworthy and ethical manner.
IBM Consulting is uniquely positioned to help businesses navigate this new era of AI, applying AI at enterprise scale while following a human-centered, principled approach that builds trust. The combination of our consulting capabilities in trustworthy AI and our powerful software, like the AI and data platform IBM watsonx, can help businesses with auditing and mitigating risk, implementing holistic governance frameworks, operationalizing AI, education and guidance, and organizational change.
This post was created by IBM with Insider Studios.