Anthropic's AI fluency chief says the best AI users know when to do the work themselves

Anthropic's Kristen Swanson, the company's AI fluency expert, argues that the most proficient users of AI understand when to employ the technology and when to rely on their own abilities. Swanson contends that using AI for every task is not the optimal approach, as excessive reliance can yield diminished results. She explains that AI fluency entails discerning the appropriate tasks for which AI is best suited and when human intervention is necessary.

The discernment tax, she refers to the process of evaluating AI's output, which can sometimes be more labor-intensive than completing the task independently. Swanson notes that while AI can be advantageous for tedious data analysis, utilizing it to generate content in areas where the user possesses expertise may result in unnecessary work.

Companies are striving to establish clear guidelines for when employees should leverage AI, while also recognizing that the effectiveness of the technology may evolve as it advances. Swanson emphasizes that AI users must periodically reassess the suitability of tasks for AI, as a task that is deemed appropriate for human completion today may be better suited for AI in the future.

This approach is crucial in preventing capability overhang, a phenomenon wherein users become overly reliant on AI based on earlier experiences with the technology.