This is a project based on three assumptions:
- Our current societies are deeply ill.*
- Training LLMs on data produced by deeply ill societies creates deeply ill AI.
- We can deep heal AI so it can help heal our societies.**
We start out from an even stronger assumption: Our human brains have evolved in conditions that are fundamentally different to those we find in our (post-)modern societies. Being thrown out of our “natural habitat” leaves basically all humans traumatized in one way or another. These (micro-)traumas influence how we understand and interact with the world.
In particular, our current social and economic structures tend put us under constant stress and foster mindsets of distrust, greed, egoism, cynicism, narrow mindedness, disconnectedness etc. This results in behavior that is harmful to our natural ecosystems and – in a vicious circle – reinforces structural harms like extreme inequality and social exclusion.
All these issues are deeply engraved in the data used for training the currently prevailing LLMs. This is a fundamental problem that reaches far beyond the current discourse around AI ethics, bias, and alignment.
**We have developed conceptual ideas on “deep healing” LLMs, which we now seek to test and revise. If this sounds interesting to you, please reach out at firstname.lastname@example.org.