We’re at a critical juncture in artificial intelligence. The rapid emergence of generative AI has sparked a flurry of predictions about AI’s growth and impact on business and society. But there’s one problem that’s holding organizations back from realizing AI’s full potential: trust.
More than half of consumers believe AI poses a serious threat to society, and without trust in AI inside and outside of the enterprise, it will never reach its full potential.
The solution to this challenge is “trustworthy AI” — AI that is designed, developed, deployed, and governed to meet the diverse stakeholder needs for accountability, competence, consistency, reliability, empathy, integrity, and transparency. Adopting trustworthy AI is a clearly defined strategy that enables businesses to deploy AI in a way that minimizes risk and doubt while reaping all of its benefits.
The trust gap
In the context of AI, trust means confidence in a particular outcome. Whether users of AI are software developers building applications or consumers interacting with chatbots, they need to be confident that the output from the AI they are using is accurate, unbiased, and useful. Inaccurate or unexpected results, genAI hallucinations, errors in text-based results, and embedded bias are among the top concerns that lower business executives’ trust in AI.
How big is the trust gap within the enterprise? A recent Forrester survey found that 25% of data and analytics decision makers said a lack of trust in AI systems was a major concern in their use of AI, and 21% cited a lack of transparency into AI/machine learning (ML) systems and models.
But consumers are far more skeptical, saying they want to know where AI is in their path to purchase and want more details about how the organizations they interact with are using it. Only 28% of U.S. online adults say they trust companies that use AI models for their customers, while 46% say they don’t. And more than half (52%) say they feel “AI poses a serious threat to society.”
As more organizations see AI as a key element of growth, the trust gap must be addressed.
Building trustworthy AI
Despite much skepticism and distrust in the market, backing down on AI efforts at this critical juncture could be the biggest mistake any company could make. Given AI’s enormous potential, the solution to these challenges is not to adopt less AI, but to have more trustworthy AI.
By definition, trustworthy AI must:Seven Measures of Trust“:
1. Transparency: For many users, AI is a black box. An explainable AI approach makes models more transparent and interpretable.
2. Capabilities: AI is probabilistic: machines learn from real-world data and reflect the inherent uncertainty of the world. Business leaders who adopt AI must embrace the fact that AI predictions are not deterministic.
3. Consistency: “Model drift” occurs when a model’s performance changes over time due to changes in data or other factors. The best way to ensure AI consistency is to adopt ModelOps, a set of tools, technologies, and practices that help organizations efficiently deploy, monitor, retrain, and manage AI models.
4. Accountability: AI will never be perfect, so if your organization’s AI goes awry — like a chatbot on an eating disorder site encouraging visitors to count calories — take responsibility, explain what went wrong and why, and take clear steps to avoid making the same mistake in the future.
5. Integrity: Appointing a chief ethics or trust officer within your organization can help guide your AI processes and build trust both internally and externally. Even if such a position does not exist, your organization should clearly define a role that will be responsible for the integrity of AI.
6. Reliability: Trusting AI means having confidence in its results, and trust breeds reliability, so the most effective way to strengthen trust in AI is to test it in virtually simulated situations before testing your models in the real world.
7. Empathy: Involving a broad and diverse group of stakeholders in testing models and incorporating feedback can help eliminate bias and build empathy for users and customers into AI models.
Of course, which levers a company focuses on will vary depending on its industry, goals, and other factors. But long-term success with AI won’t be measured by how many tools an organization adopts or how quickly they do so. AI success stories are companies that get real value from AI, and that depends, obviously, on how much customers and employees trust the technology.
Are you looking for the right partner to support you on your AI journey? Get to know Forrester.
