Is it a good idea to use AI to fight fake news?
Not long ago Mark Zuckerberg said he wants to use AI to fight fake news on Facebook. His idea to use algorithms for assessing the truthfulness of webpages or other data is not new. In 2015 Google sketched an algorithm for calculating truth ranks (or ‘Knowledge-Based Trust’ ranks, as they call them) in the spirit of their hugely successful search algorithm based on ‘page ranks’. But now that Facebook (and probably Google) are actually going to put such algorithms to work in their products, we face (at least) two new problems.
The first problem is whether or not to keep these algorithms a secret. Probably, Google will not open-source the algorithms; just like with their page-rank algorithm they will not want that people are going to play the rules of the program. That would undermine the accuracy and trustworthiness of their outcomes. But, if the algorithms are kept a secret, we face a problem that is maybe worse. Do we trust Google to not skew calculations of truth ranks in a certain direction? Should what in the future might develop into an important source for truth be in the hands of a private company? Here we see clearly why the idea of having OpenAI is a valuable one.
The second problem mirrors the first one; where the first problem above is about whether we can actually trust truth ranks, a second problem is that people will easily do. People tend to put too much trust in AI. We see this with self driving cars where simple autopilot systems are treated as if they have full level 5 autonomy. We see this when mistaken outcomes of FBI big-data-algorithms lead to innocent journalists being tagged as possible terrorists for the rest of their lives. Something similar can happen with truth ranks: the AI behind them, especially if not open sourced, is likely to be overestimated by many, and truth will no longer be what it used to be.
If we want to fight fake news by technological means, I see better opportunities in other directions: blockchain. I may explain my thoughts on that in a future post.