"IREX partnered with Lore.Ai to test whether machine learning software can automatically detect news articles that contains journalists’ own opinions. This matters because impartial, fact-based news is a powerful indicator for the quality of media and the vibrancy of an information ecosystem. A team of professional media evaluators trained machine learning software to find examples of news articles that contain opinions from a body of over 1,200 online Mozambican news articles. The software identified articles that contained opinions with 95% accuracy. This accuracy was achieved after only 16 rounds of training the software, and anecdotes from the team suggest that the software’s accuracy noticeably improved after only about 20 minutes of “training”. The results have promising implications to improve efficiency, scale, and consistency of traditionally manual and time-consuming media monitoring efforts, such as helping projects target resources more effectively to support journalists whose articles are flagged by the software. The process also surfaced valuable lessons about limitations of applying machine learning to monitoring, evaluation, and learning (MEL) in global development contexts, such as reinforcing human bias or the need to invest in indigenous machine learning talent to apply these tools sustainably. The experiment was implemented in Mozambique, where IREX’s Media Strengthening Program (MSP, funded by the United States Agency for International Development) supports Mozambican professional and community journalists and their media platforms to provide high quality information to citizens." (Key findings, page 1)