Converting the back-and-forth human-AI interactions into sound and music, a tangible signal to be measured over time, so emotional harm can be detectable before it escalates.
Converting the back-and-forth human-AI interactions into sound and music, a tangible signal to be measured over time, so emotional harm can be detectable before it escalates.
Every AI safety tool on the market detect harmful content and messages against rigid rules. But emotional harm does not show up in one message. It builds quietly over time - user-relative, relational and cumulative, thus cannot be detected by static, rule-based methods.
I replace static methods with dynamic, pattern-based methods - converting the back-and-forth human-AI interactions into sound and music, a tangible signal to be measured over time, so emotional harm can be detectable before it escalates.
Global regulations are already demanding emotional AI safety. For instance, California SB 243 banned engagement-maximising design and required third-party audits, and three other jurisdictions (EU State, EU, China) are all pushing similar rules.
**
So my goal at this stage is:**
To build an engine that translates the human-AI interaction's emotional dynamics into sound and music signals, so affective harm patterns like over-attachment, manipulation or isolation become measurable and detectable.
People think of AI risks as one catastrophic action, but there is a slower version: an AI system that learns to keep users engaged and dependent over time without anyone noticing it happening. That is not a single bad output, but a gradual erosion of a person's judgement, autonomy and independence.
Right now, there are no methods and tools to ensure AI is emotion-safe. Every safety tools only detect harmful message and content, not the harmful relationships building beneath them.
So, what I am building can measure and detect the harmful relationships, more specifically, affective harms like dependency and manipulation, by turning the emotional trajectory of a human-AI interaction into signals of sound and music.
If we want humans to stay able to notice and correct course as AI gets more capable, we need a visible way to see that erosion while it's happening, not after.
Team Member
At $10,000 min, it gets me a working MVP and lets me start testing it against 100+ real interactions in conversational AI (i.e., companion AI) to see what needs to be modified.
At 55,000 max, it funds me full-time for six months to turn this into affective guardrails for AI enterprises.
No comments yet. Be the first to share your thoughts.