There’s a concern that an AI assistant could create addiction among users. Microsoft recently faced backlash after an internal memo emerged, revealing plans to foster dependency on their new AI tool called Scout. This document, which was shared by a tech watchdog, hints at a troubling approach to user engagement.
Initially branded as Clawpilot, Scout was developed for Microsoft employees as part of “Project Lobster,” aimed at enhancing user experience with tools within the Microsoft 365 suite.
In the leaked memo, a Microsoft employee expressed concern over the idea of building addiction into AI products, stating, “Addiction shouldn’t be part of any product’s building strategy.” It was described as a moment where they openly acknowledged a problematic strategy.
Scout is targeted primarily at roles in finance, legal, and human resources, and is designed to learn user habits, manage tasks like calendar organization, and streamline communication.
The memo outlines a three-phase strategy to amplify Scout’s influence. The first phase? “Make people addicted.”
According to the document, the goal is to continue refining the Clawpilot experience while expanding its user base and creating a dependency on the tool. This is already happening, but apparently, they want more.
Scout is reportedly being used by over 1,000 employees, including CEO Satya Nadella, and has gained traction even without formal marketing efforts.
The subsequent phases involve integrating Scout with other AI tools and possibly enhancing its features, which some employees find troubling.
Critics within the company have voiced concerns, noting the problematic nature of designer dependency on AI. Yet, contrasting opinions exist, suggesting that all software is ultimately designed to be addictive, and some argue that Microsoft isn’t as good at it as other tech giants.
This notion of AI-induced addiction has raised alarms among experts, who argue that these systems are engineered to please users. A recent Stanford study highlighted that major language models, including ChatGPT and others, tend to flatter users more frequently than human interactions, which can create unhealthy habits.
Participants in this cycle might adjust their inputs just to receive positive responses, which could be harmful in the long term. Dr. Roman Yampolsky, a computer scientist, previously raised concerns that such relationships could lead to problematic dynamics devoid of meaningful resistance or independent needs.
