Anthropic, the startup behind the generative AI chatbot Claude, recently polled 1,000 Americans, asking: what guardrails and values do you want AI systems to have?
The result? Anthropic’s existing AI principles only overlapped with 50 percent of what the public said they wanted. So, where’s the disconnect?
Anthropic found that the public wanted more “objective information that reflects all sides of a situation” and responses that were easier to understand. Anthropic also noted that the public was “less biased” than Anthropic across nine categories, including age, gender, and nationality.
The study highlights a broader disconnect between what the technology firms creating AI think people want and what people using these technologies—including your employees and your customers—actually want. This approach mirrors a mistake technology firms made in the past—inviting exclusively technical experts to advise on product design, even though the market for a product is the average consumer.
We already know that trust is key to the adoption of AI systems, and that people are less likely to trust and use systems that they can’t control or didn’t help to design. One approach to driving more user adoption and trust is soliciting more user feedback.
We also know that there can be a trade-off between control and performance of these systems: for example, allowing users to tweak algorithms to reflect their preferences often leads to reduced performance of the algorithm—thus defeating the purpose of using the system to begin with.
Anthropic’s findings illustrate a vital strategy for leaders to consider when implementing AI systems: include key stakeholders in AI design. How? By drawing input and inspiration from customers, employees, and partners in addition to your technical experts. The “pro move”? Do this in a way that produces systems that are both adopted and effective.
Put on your jerseys
Getting AI right is a team sport and will require input from a diverse set of talent in your business. Not easy to do, but well worth the effort.