SearcharxivSearch

arXiv subjects

Jonathan Herington

Publications and source records attributed to Jonathan Herington.

2 recordsLinked to original sources

The Effect of Value-Focused Discussions on Scientists' Ethical Decision Making

Many scientists view science as value-free, despite the fact that both epistemic and non-epistemic values structure scientific inquiry. Current ethics training usually focuses on transmitting knowledge about high-level ethical concepts or rules and is widely regarded as ineffective. We argue that ethics training will be more effective at improving ethical decision making if it focuses on connecting values to science. We pull from philosophy and psychology to define ethical decision making using the Four Component Model. This model states that in order to make an ethical decision someone must consider four components: moral sensitivity, moral reasoning, moral motivation, and moral implementation. We formed a moderated fellowship of fourteen science faculty from different disciplines who met for ten sessions over the course of a year, where they discussed the values embedded in different scientific norms. We then conducted interviews before and after the year-long fellowship that involved guided reflection of scenarios where there was some kind of ethical misconduct where the scientific practice required value judgements (e.g using unpublished data in their own work). We looked at how the fellowship affected the scientists' ability to recognize ethical dimensions regarding the scenarios. We found that this fellowship improved moral sensitivity, but their moral reasoning does not improve. We outlined our approach on how to look at scientists' ethical decision making and made recommendations on how to improve our approach. This work can inform future ethical training to align better with what scientists value and introduce useful concepts from philosophy and psychology to education research in physics.

physics.ed-ph

Censoring chemical data to mitigate dual use risk

Machine learning models have dual-use potential, potentially serving both beneficial and malicious purposes. The development of open-source models in chemistry has specifically surfaced dual-use concerns around toxicological data and chemical warfare agents. We discuss a chain risk framework identifying three misuse pathways and corresponding mitigation strategies: inference-level, model-level, and data-level. At the data level, we introduce a model-agnostic noising method to increase prediction error in specific desired regions (sensitive regions). Our results show that selective noise induces variance and attenuation bias, whereas simply omitting sensitive data fails to prevent extrapolation. These findings hold for both molecular feature multilayer perceptrons and graph neural networks. Thus, noising molecular structures can enable open sharing of potential dual-use molecular data.

cs.LG