Fraud is one thing. Continued experiments to learn are another.. even if applied in different domains. Take applied math. It can be written is a math journal, biology, chemistry, physics etc.. I would be suspect of clinical trials specifically but any preclinical work should be replicated. CS work is easy to replicate. So I think the specific threat of fraud is limited. Me too publications suggest learning but probably lack originality. The challenges are the inductive leaps which are usually counter intuitive and outside of the mainstream. Funding agendcies need to promotes those.