The Center for Humane Technology is working to popularize the term “AI drift,” describing a model’s tendency to degrade in performance over time. I’m not crazy about the term because I don’t think it clearly communicates the dangers that AI presents to people as they go about their daily lives.
In my workplace I’m seeing more and more cases where people are turning to AI to deal with tough problems they don’t want to think about. Problems such as formulating goals, writing memos, drafting proposals.
The danger here is that they are not writing these things themselves, meaning they are not truly thinking them through. Writing is thinking.
If I have an LLM write the goals for my team, I’ll get a list of objectives that have a reasonable probability of being a good match, but they won’t be tailored for the individuals.
I think that using AI to do your problem-solving carries an even greater risk because it can bypass your own sense of knowing, at the center of your being, what is true.
One of the fundamental methods for designing structures and processes that “have good fit” and “feel alive” is to use the Mirror of the Self Test. This is a learned skill that takes practice but produces excellent results by tailoring design solutions to perfectly match the problem space.
I consistently find that AI results often sound good, even insightful. They might trigger a positive gut reaction, such that I feel that they are true. But when I genuinely study the problem space I see that the LLM is lacking context and the solution really doesn’t fit. The only way I can reach that conclusion is to map out the space using what I know in my Self, or my whole experience of life.
And AI generated answers interfere with that Self-evaluation process, making it less human and less reliable..
SEE ALSO
Niklas Luhmann – Impossible to think without writing
