Artificial Intelligence Gains More Trust in Confidential Environments

"When people can't see how artificial intelligence works, they may trust it more," wrote an article in the latest issue of Harvard Business Review. The article demonstrates how understanding the workings of an unknown model can help people trust the process more.

A similar pattern can be observed among leaders in the tech industry. As one of the many image-conscious companies in Silicon Valley, Apple also ensures secrecy around their artificial intelligence and machine learning practices. OpenAI follows a similar approach, striving to hide its technology while actively attracting enterprise clients.

During this year's WWDC, CEO Tim Cook noticeably avoided using the term "artificial intelligence" and instead opted for the milder "machine learning." Apple's aversion to labeling its technology as "artificial intelligence" is not new, as the company has long been cautious about hyping its magical capabilities. Instead, Apple focuses on the practical functionalities of machine learning, emphasizing the tangible benefits for its user-centric audience.

The company's executives stated in an interview with "Good Morning America," "We do integrate it into our products, but people may not necessarily perceive it as artificial intelligence. This gives Apple an advantage over competitors like Microsoft and Google, who are currently boasting about their artificial intelligence products but struggle to adapt the entire company."

OpenAI follows the same secrecy strategy as Apple. The company's 98-page technical document lacks basic details about artificial intelligence model data or architecture. Despite receiving harsh criticism for its shallow content, this secretive approach seems to work in favor of the company.

Trust in the Process

A group of researchers from Georgetown University, Harvard University, and MIT analyzed inventory decisions for 425 products across 186 stores of luxury retailers in the United States. Half of the employees received recommendations from an easily understandable algorithm, while the other half received recommendations from an opaque algorithm.

Comparative analysis of the decisions showed that employees were more likely to align with recommendations from the opaque artificial intelligence. The conclusion drawn from the results is that people exhibit higher confidence in artificial intelligence systems when they do not fully understand how they work.

Professor Timothy DeStefabo emphasized a recognized phenomenon that decision-makers are generally reluctant, consciously or unconsciously, to accept guidance generated by artificial intelligence and choose to override it. This is not the first time. Historically, resistance to new technology has been the norm.

DeStefabo and his team collaborated with Tapestry, a $6.7 billion company and the parent company of Coach, Kate Spade, and Stuart Weitzman, to study the roots of this reluctance and find strategies to mitigate it.

The company had long relied on rule-based algorithms to assist allocators in estimating demand. The model was based on users' daily experiences and visible inputs. The company developed a more complex predictive model, a black box to users, to achieve better accuracy. Surprisingly, shipments aligned with recommendations from the latter model by approximately 50%, indicating greater trust from users in the black box model.

Prior to this, the company had long relied on rule-based algorithms to assist allocators in estimating demand. These algorithms were understandable to users, based on their daily experiences and inputs. However, the company developed a more complex "black box" predictive model for improved accuracy. The surprising result was that shipments aligned with recommendations from the latter model by approximately 50%, indicating greater trust from users in the black box model. This result suggests that users have more trust in black box models.

One reason allocators reject less complex systems is due to "overconfidence in troubleshooting" - users believe they have a better understanding of the model than they actually do. DeStefabo wrote that although employees do not know how the model works, as it was developed and tested with input from some colleagues, it gives them confidence in the model.

In conclusion, tech companies need to focus on meeting customer needs rather than selling technological tricks to customers.