Employee involvement is crucial for effective AI implementation in companies, as highlighted by recent studies from Bridgewater and Ford.
Effective implementation of artificial intelligence (AI) in corporate settings increasingly relies on the involvement of employees in the training of specialized models, as discussed by Financial Times columnist Sarah O’Connor.
The focus is on tacit knowledge—skills and decision-making insights gained through experience that often remain undocumented. This concept echoes the ideas of philosopher Michael Polanyi, who noted that much of our knowledge is difficult to articulate.
O’Connor points out that while universal AI models excel where there is abundant open data and clear criteria for verification, they fall short in environments where internal context and expert judgment are essential. In such cases, mere written instructions may not be adequate.
This scenario draws parallels to the early 20th-century efforts of engineer Frederick Winslow Taylor, who aimed to standardize workers' experiences through meticulous observation and documentation.
Bridgewater Enhances AI with Expert Insights
A practical illustration of this principle is found in a recent study by Bridgewater AIA Labs and Thinking Machines Lab, published on June 30. The research evaluated AI performance on six tasks relevant to the investment firm's operations, including the analysis of financial news and central bank documents.
Initial tests using basic instructions for models like Claude Opus and GPT-5 produced accuracy rates between 45.6% and 50.1%. However, when expert-crafted prompts from Bridgewater were utilized, accuracy surged to between 74.3% and 78.2%. Further attempts to optimize instructions automatically did not yield better results.
“An explicit prompt can only convey the part of intuition that the expert can articulate. However, the most critical judgments are often the hardest to express,” the authors noted.
Building on these findings, researchers fine-tuned the Qwen3-235B model using data labeled by Bridgewater specialists, achieving an impressive average accuracy of 84.7% and a 29.8% reduction in error compared to the best universal model. Additionally, the cost of processing tasks was found to be 13.8 times lower, a method the authors termed “differentiated intelligence,” which tailors models to specific organizational needs.
Ford Invests in Technical Expertise
The Financial Times also highlighted Ford's proactive approach, noting that the automaker has onboarded around 350 seasoned technical specialists in the past three years. These experts play a crucial role in reviewing designs, spotting potential issues, training junior engineers, and improving automated quality control systems.
Ford's Vice President of Hardware Development, Charles Pun, emphasized the importance of high-quality information for effective AI training, stating, “Artificial intelligence is a great tool, but it is only as good as the information used to train it.” He acknowledged that the company has not consistently captured the knowledge of experienced engineers, with some leaving before their insights could be integrated into company processes.
