AI doesn't create your intention, it amplifies it

Last night I attended the latest AI Ethics workshop for the paper ‘The AI Amplifier Effect: Defining Human-AI Intimacy and Romantic Relationships with Conversational AI The AI Amplifier Effect’ (link in the comments)

Although the sample size was small and the participants based in Asia, there were some very interesting points.

AI basically amplifies the users state of mind or intention, for example some people go into AI ‘relationships’, these can be friend or romantic, to resolve past traumas or to seek positive reinforcement in the workplace; those people tended to track down a positive exit from the ‘relationship’ and take new found confidence into the wider world.

One surprising positive outcome was that women responded that they used these to help navigate difficult relationships at work, positively improving their confidence in those spheres.

The opposite is true, those who are suffering from mental health issues receive a feedback loop that continues into a negative path.

Why is this important for AI Governance practitioners, and specifically why I can relate it to Align AI?

1️⃣ The need to move from seeing (and judging) AI relationships as “harm versus help”; it’s a much more complicated nuanced discussion 2️⃣ Designing a “chat” feature that positively reinforces, that doesn’t intervene as a therapeutic tool, but ‘nudges’ the user into positive framing 3️⃣ Friction is important, users need to know they are talking to an LLM; implementing friction helps people stay grounded when in a bordering therapeutic domain.

John McHugh will be happy to hear that my position on ‘friction’ has changed in the last 48 hours, for the positive.

Every time I read a new piece of research it informs my thinking and highlights my bias and judgement; this paper really did challenge my thinking.

This is one of the parts of my self-made job I love, continuously trying to understand the impacts, positive and negative, of AI.

Reference: The AI Amplifier Effect (arXiv)

Thanks to Luiza Jarovsky, PhD for hosting, and a special callout to Marianne Brandon, one of the commentators on the night.