The rapid advancements in AI necessitate the establishment of guardrails and the development of an ethical philosophy regarding its integration into the workplace. Paula Goldman, the chief ethical and humane use officer at Salesforce, emphasized the importance of AI playing the role of a co-pilot alongside humans rather than functioning on autopilot during Fortune's Brainstorm AI conference in London. Goldman expressed the need for next-level controls and the ability for individuals to comprehend the workings of the AI system. She stressed the significance of designing AI products that consider both AI's strengths and weaknesses, as well as the strengths and weaknesses of human decision-making. One of Goldman's concerns relates to the generation of ethical content by AI, ensuring that it is free from biases and excessive user-generated material like deepfakes that may harm individuals. She believes that unethical use of AI could hinder its funding and development, potentially leading to a loss of trust or user-adoption issues causing an "AI winter. " Goldman stated that the future productivity gains of AI in the workplace will heavily rely on training and people's willingness to embrace new technologies. To instill trust in AI products, particularly among employees, Goldman proposes the implementation of "mindful friction. " This concept entails incorporating checks and balances that ensure AI tools in the workplace bring about more benefits than harm. Salesforce has already taken steps to implement this idea of "mindful friction. " They have developed a marketing segmentation product that generates appropriate demographics for email campaigns. While the AI program provides potential demographics, it is the responsibility of the human to select the suitable parameters to avoid exclusion. Salesforce also includes warning prompts in their generative models that incorporate zip codes, as these can be correlated with specific races or socio-economic statuses.

The aim is to detect such potential biases and encourage humans to review the data more closely. Historically, biases and copyright infringements have raised concerns about trust in AI. For example, a study from MIT Media Lab showed that AI software had a 35% error rate in identifying dark-skinned women, including prominent figures like Oprah Winfrey and Michelle Obama. Biases in facial recognition technology compromise critical tasks such as equipping drones or body cameras with facial recognition software for lethal operations. Additionally, algorithmic biases in healthcare databases can lead to inappropriate treatment plans being suggested for certain patients, as discovered by the Yale School of Medicine. Even in industries where lives are not at stake, AI applications have provoked ethical worries. OpenAI, for instance, scraped user-generated YouTube content, potentially infringing on the copyrights of creators without their consent. Moreover, AI's propagation of misinformation and its limitations in completing basic tasks indicate that it still has a long way to go before fully realizing its potential as a helpful tool for humans. Despite these challenges, Goldman finds hope in designing more intelligent AI features and implementing safeguards led by humans to enhance trust in the industry. The main focus lies in developing products that users can trust while also recognizing situations where human judgment should be applied and given a second look.