A new preprint on arXiv, Formal Raising: Embedding Verifiable Ethical Properties into the Development of AI, takes up Alan Turing's suggestion that building an intelligent machine might be less like programming an adult and more like educating a child. Turing called this hypothetical system a 'child machine' and argued that it could be shaped through learning rather than direct coding. The paper connects this idea to recent 'model raising' work, which revives the developmental view for large language models.
The title signals the paper's central contribution: a formal approach to raising AI that makes ethical properties verifiable. Rather than treating ethics as an add-on after training, the proposed framework appears to embed ethical requirements into the development process itself. The abstract, however, is truncated in the announcement, so the specific mechanisms and formal methods are not described in the available text.
What is clear is that the paper positions ethical training as a developmental process, echoing Turing's original vision. The emphasis on verifiability suggests a move toward more rigorous, checkable guarantees in AI alignment, though the details remain to be seen in the full preprint.