Congratulations on your IDG grant! Can you tell us more about this new phenomenon of robo-news?
Historically, journalists have been the ones writing news and generating content. We’re seeing a shift now where a lot of the news is not written by human journalists but actually written by robots. This can be seen in all forms of media; articles, press releases and also on social media. Trusted news sources, such as the Associated Press and the Washington Post, use robots to generate some of their content as well. And it’s sometimes not easy to determine if the information is from a human or a robot. Accountability is therefore a big problem, especially when we think about other forms of media – such as Twitter. In these cases, the onus is now on the consumer of information to verify the reliability of the source of information; it’s an added responsibility for the public. This is very relevant right now with Elon Musk going back and forth about acquiring Twitter before he eventually took the helm. One of his main criticisms is that there is significant robo-generated content on Twitter. He’s looking at this as a negative—and that may certainly be the case—but we don’t know until we research and find out!
What is the goal of your research?
My focus is on the impact of robo-generated news on financial markets. For example, in finance, when there’s news about companies or on the economy there’s typically a market response. Now we want to know what is the market response to robo-generated content vs. the market response to human-generated content. That’s one of the most important research questions—which one really drives financial markets more?
The initial study will take place in the context of social media and will use a dataset of 350+ million financial tweets about thousands of North American firms. First, we need to identify robo-generated content, and secondly, look at their impact on financial markets—and whether they are in sync. They may be better than human-generated content, worse or similar. It may be the case that they may be good at certain times but bad at times of recession or when the economy is moving downwards for example.
How do you identify robo-generated news?
In this case, we’re using artificial intelligence algorithms—specifically unsupervised learning—to identify robo-generated content. Sometimes, people declare that they are robots but oftentimes that’s not the case. We’ll be identifying robo-generated content vs. human-generated content based on patterns and habits. For example, if somebody is generating 500 tweets a day that’s unlikely to be a human. And if somebody follows a million people and is not followed by anyone in return, that’s likely to be a robot. These kinds of features will help identify whether the tweet is from a robot or a human.
Once we identify which content is generated by humans and which content is generated by robots, then we’re going to construct a daily index of robo-content and a daily index of human content. The index includes the amount of content—how much information is generated by each type—and also what’s the sentiment. Is one positive and one negative or are both neutral? And the next part is looking at which one appears to be more correlated with financial markets. For example, if humans are optimistic and robots are pessimistic, and then markets turn out to be negative, that may suggest that robots are a little bit more in sync. But if it’s the other way around, then maybe humans are more in sync than robots. That’s an important part of the study, looking at which one is more in sync and is more likely to predict markets—is it humans or robots?
How will the research be conducted & who will you be working with?
We’ll be utilizing Sprott’s Social Media & Finance Lab. I recently received funding through the Canadian Foundation for Innovation (CFI) to strengthen and expand the capacity of the lab. We need a tremendous amount of computing power for this project as we’re using a very large dataset. The funding from CFI really makes this work possible and will make this lab the first of its kind in Canada. With the help of one of the world’s most advanced AI Supercomputers, this research will become possible.
I’ll also be working with two collaborators from Carleton University — Olga Baysal, Associate Professor in the School of Computer Science and Sandra Robinson, Assistant Professor in the School of Journalism and Communication. Since this project deals with AI and social media, it was important to have research collaborators who are specialists in these fields. The work will benefit incredibly from Olga’s computing expertise and from Sandra’s knowledge of journalism and the media profession.
What do you predict for the future of robo-news?
First and foremost, there’s going to be an impact on the profession of journalism. Can people in that profession still survive? Does the world need journalists anymore or can everything be robo-generated? Once we get some initial insights, we’ll be able to look at different implications for the media profession. From the perspective of investors, they rely on public news sources to make investment decisions. This research is really important for them and for the broader investing public to understand the impact of robo-news so they can better assess their sources of information and make better investment decisions.
In terms of predictions, I don’t know who will be more in sync with financial markets—we’ll have to wait for the results—but I think there will be more robots over time. I’m not sure if I’m ready for robots to play a big role but that might be the case and I’m very curious about the answer myself.
Sprott School of Business