The Accelerator exploits Artificial Intelligence and machine learning to automate the detection of bias in news coverage. The project aims to determine whether AI-Generated Metadata that currently helps a broadcaster like Al Jazeera to search and discover content, can also help them to analyse and measure bias, raise flags and hold themselves accountable to their values.

The team will be focused on the ability to show diversity, to be fair and balanced as any news organisation strives to be. With 4 or 5 language broadcasts across multiple platforms and 24/7 news and programmes generated daily, being able to track and measure against diversity, editorial and regulatory standards is a real challenge for the Champions.

The team has taken the pragmatic decision to narrow the focus of the POC to something that can both be achievable and indicative of the promise and potential of AI to support the media in this realm. The starting point was the recognition that the language of news broadcasting is a powerful way of conveying very subtle meaning and is a significant means to persuade, to endorse, to contradict, or to cast out. While acknowledging that language choices are reinforced by visual elements, whether that be human expression and gesture or choices made in an edit suite, the Accelerator has concentrated on analysing text, still a fearsomely complex task.

For esteemed, independent news organisations, is vital that such a tool, and definitely a productised version in the future, is open and transparent, so that when news organisations are accused of bias, or if they want to check their own output against reference markers, the whole process takes place in the open and can be examined for any faults or discrepancies. While what is seen in the culmination of the 2021 phase of the project and outputs at year end are not going to be a finished product, it will provide important insights as to what AI can do and point toward some key possibilities to come. 

The aim of the PoC therefore was to develop the outline of a machine language to evaluate the semantics of news coverage, designing a workflow and processing engine to detect the tonality of reportage. Given that 95% of which is unstructured audio, video or text in, the team began with a feasibility v’s impact matrix focused on the analysis of news bias. A pipeline was subsequently developed with inputs in the form of video, audio and text, with three machine learning processing modules for processing each with a feature extraction layer to focus on areas such as topic mapping, quote attribution, topic polarities, object recognition etc The team’s PoC ultimately demonstrated a AI and the workflow methodology to examine and analyse the coverage of a single event by multiple news organisations, with the Fall of Kabul from 15 August onwards chosen as one example. The next steps will be to develop the concept further, with the possibility of industry collaboration continuing, to develop a tool as a ‘minimal viable product’.