Dreading the task of reading through thousands of documents, or sorting through millions of leaked records? Delegate that to Artificial Intelligence, a term for computer processes that simulate human learning, reasoning and self-correction.
Paul Bradshaw, a journalist with BBC England’s data unit, shared 10 ways you can use Artificial Intelligence in your reporting — and where it might fall short.
“One really important thing to remember about machine learning is that it is always an estimate, it will never be 100 hundred percent correct,” Bradshaw warned.
Establish the Scale
You have a hunch that a problem may be bigger than a few anecdotes you’ve collected, but you don’t know the actual scale. “We need to know how many people are affected, or we need to know how long this has been going on,” said Bradshaw.
An algorithm can be trained to look for specific text patterns and analyze numerous documents to identify what’s relevant. The Atlanta Journal-Constitution fed an algorithm with documents from sexual harassment cases. The algorithm then went on to look for similar documents, giving the team more leads to investigate.
Prove a Problem Exists
Zoom in on a problem. Swiss broadcasting outlet Schweizer Radio und Fernsehen trained an algorithm to detect fake social media accounts based on multiple factors, and was able to show that a third of the followers of a Swiss influencer’s accounts were fake.
Unmask a system
Natural Language Processing can be used to analyze linguistic patterns instead of just identifying words. It can include the context in which the words are used, as well as the speaker’s intent. “That has a lot of opportunities in terms of journalism,” he said.
ProPublica analyzed exactly how campaigns customized emails to voters with slight variations, based on what they knew about them.
But NLP has been criticized because, for example, the algorithms are typically trained in a specific cultural context, and potentially with bias with regard to gender or ethnicity.
AI can help detect linguistic patterns in order to interpret sentiment, and of course detect changes. This is especially helpful when dealing with thousands of pages.
The Washington Post obtained draft versions of 12 audits by the USAID’s Inspector General office, and with the help of AI compared finalized versions. They were able to identify that anything that expressed negative sentiment or could be interpreted as critical was removed from draft versions of documents.
Unmask pseudo-human behavior
Artificial actors and automated content are on the rise. AI can identify bots and content generated by a machine rather than a human, which can be useful information for your investigation.
Find leads from huge data
AI can generate leads from large quantities of data. For example, an analysis of salaries can identify the school with the most overpaid teachers. Or gender pay gaps. “And that’s where you have to do your interviews,” Bradshaw said.
Jargon and obfuscation
Government agencies, companies, NGOs, and even journalists love their jargon. But those terms, acronyms and phrases can be difficult to understand, and even more difficult to analyze. AI can look through patterns of texts and documents and decode profession-specific language. “It helps you ask the right questions,” Bradshaw said.
Associations and links
AI can zero in on relationships and networks, and recognize entities that may have inconsistent identifying information. For example, AI can match property records, tax data and other information, even if identifying entities are not an exact match.
Eye in the sky
AI can analyze drone footage and satellite imagery and let you know, for example, where illegal mining is taking place. This can provide leads and targets for your investigation. It can also run sensor data so you don’t have to pore over a huge amount of data.
Conversion and summaries
OCR technology can read scanned or photographed images of handwritten or printed text and convert them into searchable files. AI can produce the same from audio and video content — and also summarize the content into something coherent.
This technology is not perfect, so prepare yourself for errors – but it can help lift the burden.
Make your own bot
You can use AI for personalization, headlining, translation and chatbots. For example, bots can generate targeted, localized coverage based on a big national story or investigation, bringing more readers to the original content. Bots can also interact with your readers on a story 24/7, keeping them engaged longer.
AI has its drawbacks: it’s not 100 percent accurate, it can feed into cultural biases, and you have to think about how much editorial control you retain when automating stories.
“Fundamentally, the final thing to consider here is that it is a tool. Like any tool, it exists within cultures,” Bradshaw said.
Moreover, in reality, media outlets will likely only invest in AI technology what they can monetize over time.
Leonie Kijewski is a freelance reporter based in Phnom Penh, Cambodia. Her work has appeared in The Guardian, Al Jazeera, Voice of America and various other publications. She speaks German, English, French, and Dutch. She previously worked for the Phnom Penh Post as a sub-editor and reporter. She holds a master’s degree in international law.