If you torture your data too much it will confess to anything, and most often, your algorithms will make an undesired decision.
Living in a world of Big Data, it may take you several days to realise that your Artificial Intelligence (AI) code erred on a matter of public interest.
That just happened to Facebook when its algorithms flagged down President Uhuru Kenyatta’s congratulatory message on Uganda’s president Yoweri Museveni last Sunday.
The social media behemoth was forced to apologise to State House on Thursday, regretting that its algorithms failed it in an electioneering period where social media platforms are expected to protect citizens against misinformation.
“In this case, the post shared by State House Kenya featured a photo of Yoweri Museveni that had been marked as false by one of our independent fact-checking partners in December 2019. This caused our automated systems to mark the new post as fake news,” Facebook said in a statement sent to the Nation.
Matching error fixed
The company then automatically applied a label stating that “the same information was checked in another post by independent fact checkers.”
“The issue has been resolved, with the matching error fixed, we are in communication with the State House Kenya. Our goal is to protect more people from harmful content across our platforms. However, we know that our systems are far from perfect,” Janet Kemboi, Facebook Spokesperson in Kenya said when she spoke exclusively to the Nation.
Facebook has been using AI to kick off similarity detection processes that identify duplicates of debunked stories and reduce their distribution.
It feeds these ratings as a signal back into its machine learning model, which helps in detect potentially false items in the future and do so faster.
However, even as research on AI intensifies across the globe, achieving human capabilities is still miles away, especially in the emotional and empathetic aspect of machine learning.