Andreas Vlachos on his research into Imitation learning and fact checking.
Everyone should have access to fact checking. It educates people about the way facts are usedAndreas Vlachos
When he started doing work on an automated fact checking system Andreas Vlachos did not anticipate the kind of publicity he was going to generate.
But the advent of Brexit and Trump and the focus on fake news has brought him and the start-up he co-founded a lot of media attention, including a recent feature in the New York Times. “It has become very apparent that there is a need for more fact checking and that automation could be the answer,” he says.
He says that, although currently his algorithms are not able to achieve what human journalists can, there is a real need for them because there are not enough humans to check the multiple sources of news on the internet.
“It’s like translation. No-one thinks Google Translate is as good as a human translator, but it is useful,” he says. There is still a very long way to go to scale it up, but he sees it as ultimately an aid to journalists and to the public.
Andreas did his Bachelors degree in Computer Science at the University of Athens. There he started work on a project on web development and interfacing databases, but he didn’t find it very interesting so his supervisor suggested he work on a project which involved building systems using financial sentences and acronyms. It was his first encounter with artificial intelligence, a subject that was not taught in Greece at the time. Using the rules he had created, Andreas started writing software with a friend.
Alongside Computer Science Andreas was also doing a degree in classical guitar and took part in the selection for the Greek entry to the classical Eurovision contest when he was 19.
Biomedical text mining
In 2003 he began his master’s at Edinburgh University with a focus on artificial intelligence and wrote a paper based on his research on active learning, specifically how computers learn from people, which was published in 2008 and remains his most cited paper. He applied to do a PhD at Edinburgh, but couldn’t cover his maintenance costs so he applied for a research assistant position in Cambridge in 2004 and worked on biomedical information extraction, putting data from science papers onto searchable databases.
Andreas continued this biomedical text mining work when he was accepted to do a PhD in machine learning and natural language processing at Cambridge for which he received a Gates Cambridge Scholarship. He began the PhD in 2006.
While he was at Cambridge, where he was based at Peterhouse College, he was involved in a campaign to get the Gates Foundation to divest from fossil fuels which made a big impression on him. His interest in ethics and his desire to make a difference is what has kept him in academia because he feels he can raise wider social issues around, for instance, diversity in machine learning than might be possible if he was working in industry and he can have an impact on students as a teacher. Andreas did his post-doc at the University of Madison-Wisconsin in the Department of Biostatistics and Medical Informatics which enabled him to advance his methods. He began working on a paradigm for Imitation learning which has a grounding in robotics and computer science. After that he returned to Cambridge for two and a half years, working on dialogue systems which are used, for instance, for helping tourists to find their way around via voice-activated software on their phones.
In 2014 Andreas moved to University College London and began working on a fact checking project with the BBC which he devised with his supervisor. “I was interested in politics and thought that the lack of grounding in facts was a major problem in public discourse,” he said.
He feels the biggest danger comes from misinformation emanating from people who are trusted figures.
Andreas recently moved to the University of Sheffield where he is working on imitation learning for structured prediction, natural language generation to automatically generate weather forecasts from data on weather sites and fact checking and where he, Dhruv Ghulati, Robert Stojnic and Sebastian Riedel set up Factmata, a fact-checking start-up company.
It faces several challenges, including the fact that a lot of information on the internet is outdated so claims based on it are inaccurate and that much important data is not easily accessible.
In October, Factmata applied for funding through the Google Digital News Initiative to develop their idea further. “I want to bring my research to the public so that anyone can use it to battle misinformation,” says Andreas, who has taken on an advisory role in the company. The company will not launch until a plug-in has been developed, meaning it can be more widely used, but it has already received a lot of publicity, including the article in the New York Times.
Andreas is also an adviser on a global fake news challenge organised by two US researchers, Dean Pomerleau and Dedip Rao. That has also brought news coverage. The challenge offers competitors a $2,000 prize to use artificial intelligence to combat fake news. Using a database of verified articles and their artificial intelligence expertise, groups of students, independent programmers and tech industry experts have to accurately predict the veracity of certain claims.
Andreas believes that in a world of overinformation fact checking could change the way we read the news. “Everyone should have access to fact checking,” he says. “It educates people about the way facts are used.”
- 2006 PhD Computer Science
I am a lecturer at the University of Sheffield, working on the intersection of Natural Language Processing and Machine Learning. Current projects include natural language generation, automated fact-checking and imitation learning. I have also worked on semantic parsing, language modelling, information extraction, active learning, clustering and biomedical text mining.