Elas is the Mascotte of ASReview and your Electronic Learning Assistant who will guide you through the interactive process of making decisions using Artificial Intelligence in ASReview. Elas comes with some essential principles:

Humans are the Oracle

The interaction between humans and machines will take us a significant leap forward. We believe a human should decide whether to mark a record as relevant/irrelevant (hence, the human is the Oracle). The AI merely orders the records on the relevance score as predicted by the model in each iteration of the active learning cycle.

Code is Open & Results are Transparent

We are strong proponents of open science, and therefore all the code is available and accessible (and Libre) on Github. We value the user’s privacy and do not get to see any of your data since everything stays on your device (or your docker or server). All the project information, the settings of the model, and each decision made by the human are saved, making the entire pipeline transparent. The project file can be used to retrain the model at each stage of the active learning phase making the whole active learning process reproducible. We encourage you to follow the FAIR data principles and publish your data, results, and project files on a data repository with a permissive license.

Decisions are Unbiased

We present only the essentials needed for unbiased decision-making. When screening, for example, academic papers, we show titles and abstracts only and do not present authors or journal names. This way, you can focus on what is truly important (the content)!

The interface shows AI is used  

Simplicity is the ultimate sophistication (Davinci), and, therefore, we keep the front end as clean as possible. BUT, we do explain which parts are based on the results of the AI and why this is the case. This might be boring, but essential for the ethical use of AI-aided tools so that you always know which part of the process was influenced by an AI.

Users are responsible for high-quality data

ASReview focuses on the machine learning part of the reviewing pipeline. The preprocessing of the data is left to the user. Be aware of the principle GIGO (Garbage in = Garbage out) and check the quality of your data first if the results of the AI seem to be disappointing.