Paper 01 / Public research brief
Principal-Need-Shaped Personal AI
How Persistent AI May Develop Distinct Capabilities Around a Human Life
Methodology Validation
An AI can learn a person’s preferences without changing what it is able to do. Paper 01 asks whether their changing needs could shape lasting changes in its abilities.
If two AI systems start out equivalent and encounter the same tasks, could different long-term human needs lead them to develop different abilities?
Research hypothesis
Need-Trajectory-Induced Capability Individualization
Can changing a person’s long-term needs change the abilities their AI develops, even when its starting point and task history stay the same?
We call the sequence of a person’s changing needs a “need trajectory”. A “capability profile” describes the abilities an AI has developed. The proposed causal relationship between them has not been established.
Research status
What has been done so far
- Theory
- Advanced
- Prior Art
- Advanced
- Exploratory Tests
- Completed
- Methodology
- Under Validation
- Confirmatory Study
- Upcoming · not completed
- Real-human longitudinal validation
- Not yet established
We have run exploratory simulations to check and revise the method. The progress labels are our own assessment, not an independent review of the findings.
NT-CIE / Proposed experiment
Need-Trajectory Causal Individualization Experiment
NT-CIE is the proposed experiment for Paper 01. It holds task history constant while varying long-term human needs. This is intended to help distinguish learning from repeated tasks from development shaped by needs.
- Same starting AI
- Same task history
- Different long-term human needs
Do different abilities develop and persist?
Differences alone would not settle the question. They would need to last and relate to the assigned needs. We are still checking how to test this reliably. The implementation and experimental parameters are not public.
Research Before Claims
What the experiments can tell us
A simulation can produce the expected result for the wrong reason. We need to check what the measures capture, compare against alternative methods and retain results that contradict the hypothesis.
Early experiments produced positive and negative results. After comparisons with stronger baselines and reviews of the method, some mechanisms were narrowed or removed.
A confirmatory experiment has not been completed. We have not established whether the idea holds over time with real people. These simulations do not demonstrate a working Personal AI product.