Dec 28, 2020 09:20:00

NeuroIPS, the top AI society, requires authors of papers to describe 'the impact of theory on society'

NeurIPS, an international conference in the field of artificial intelligence (AI) and machine learning, has added 'a description of the wide range of effects of theory' as a requirement for papers submitted to the conference.

The NeurIPS 2020 broader impacts experiment «Statistical Modeling, Causal Inference, and Social Science
https://statmodeling.stat.columbia.edu/2020/12/21/the-neurips-2020-broader-impacts-experiment/

There is a need for a balanced discussion of the potential impact of what computer scientists have created, whether good or bad. The method of discussing the ethical aspects of theory has been considered for some time, but this is the first time that requirements such as 'a description of the impact of theory on society' have been set. This will allow us to be more active in discussing the impact of theory. NeurIPS's setting of these requirements is part of a reform movement aimed at significantly changing the values of computer science from the traditional view that 'algorithms and mathematical formulas are far from ethics.' There is also a view that it is.

However, the criteria are very vague as to what the 'extensive impact' is and how to describe it to meet the requirements. If you refer to the official

FAQ page , 'A description of the broader impact is mandatory, but if the impact is theoretically predictable and general, or if it is unlikely to have any impact, then You can freely describe that. '

Regarding the review of papers, there is a statement that the reviewers only judge 'whether the impact of the research content is strictly described' and not the 'good or bad of the impact'. The FAQ page states, 'If the research content creates unreasonable prejudice or is intended to cause harm, we may refuse to accept it from an ethical point of view, but the research content will only adversely affect society in the future. Even if it brings about, the receipt will not be refused. ' However, even if the research content is problematic from an ethical point of view, it may be peer-reviewed once and received as a result of rigorous examination.

At the Neural IPS 2020 international conference hosted by Neur IPS from December 6 to 12, 2020, no papers were rejected because they did not meet the requirements for 'widespread influence' and were peer-reviewed ethics. There were 13 problematic treatises, of which only 4 were rejected. In addition, by setting these requirements and activating discussions, it will be a psychological support for researchers working on research that has an adverse effect, and researchers will be able to give ethical meaning to their research content. There is also an opinion that it is a big experiment to know and make it more responsible.