A common challenge in science is the human capability to evaluate the real impact of an observation and a data set. This is a complex task due to having only partial information and/or to the complexity of the problem, requiring different fields to be combined. In order to overcome these important limitations, we need to be able to review all the available data and interpretations. This would allow us to evaluate the global distribution of a specific process or phenomenon of interest. The increasing number of scientific publications prevents scientists from being able to keep up with all the available literature especially when scientific papers cross disciplines. These challenges prevent us from evaluating the global impact of a certain process and are particularly relevant today given the impact of our scientific assessment on one of the most pressing issues of our time, which is climate change and its impact on society. We present here an application of artificial intelligence to geosciences: We conduct a systematic analysis of geoscience literature through a hybrid machine-human approach. Such applications are more common in other fields such as biomedicine and are in their infancy in the geosciences because of various difficulties the machines encounter in parsing geologic literature. We describe here some of these limitations and how we overcame them. We then use the following case study as an example to test our approach: We ask whether climate is influenced by volcanism in the geologic past. Our case study results show, as expected, that most analyzed literature in this experiment conclude that volcanism influences climate change in deep time, but there is no complete consensus on this question. Similarly, any question of potential global significance, such as the impact of human activities on climate change, can be posed as an interrogating technique for our vast and fast-growing literature in the field of geosciences. Such an approach has the potential to be applied to a variety of complex problems, hence addressing some of the major limitations with cross-disciplinary research.
You can access the full article here.