Arjun Moorthy - Dec 2, 2019
A solution for algorithmic transparency, as outlined to the US Senate
In June 2019 Stephen Wolfram, a pioneer in computer science research, was asked to testify to the US Senate Commerce Committee about one of the more pressing problems of our time: how internet platforms persuade us to spend too much time on them. His testimony was measured and thoughtful, as befits the youngest MacArthur grant recipient ever. Wolfram’s solution to the problem of algorithmic transparency turned out to be similar to what The Factual has built and what many of our daily newsletter readers already benefit from every day. Could The Factual be an answer to this critical issue facing society?
Tristan Harris, Maggie Stanphill, Stephen Wolfram, and Rashida Richardson at the US Senate - June 25, 2019
Wolfram started his testimony saying that Artificial Intelligence (AI) engines do indeed “run the world” as billions of people are being fed content selected by AI. The problem is every AI has biases and because of the black-box nature of these algorithms, it’s hard to say what the AI is optimizing for.
Because AIs are too complicated to “open up” Wolfram suggested that third parties software providers can take “pre-digested feature vectors from the underlying content platform, then use these to do the final ranking of items in whatever way they want.” In essence, a layer of software is added atop your favorite news sources to select content along dimensions that you, as a reader, value. Wolfram calls these “final ranking providers” and, on the surface, it sounds similar to what The Factual is.
Wolfram's diagram of where a Final Ranking provider fits into the content flow
Readers of The Factual’s daily newsletter know that stories are automatically selected based on two dimensions:
The most widely reported news topics.
The most credible stories on each topic.
Selecting the most widely reported topics is easy to understand.
To select the most credible stories on each topic The Factual automatically calculates a Credibility Grade for each article. This grade is based on attributes that people already associate with credible news.
The Factual evaluates each article for the author's historical topical expertise, the extent of sources in the story, how objective the tone of writing is, and the site's historical reputation. Every news article on our site transparently displays these four feature vectors so that readers can easily understand how and why an article was selected.
The Factual Credibility Grade's four feature vectors
In addition, users can toggle filters for political leaning preference, or story coverage, to adjust their feed as suits them.
The Factual's filter controls
The result is a manageable, transparent, and useful news digest instead of the seemingly random and endless assortment of news in people’s feeds.
Wolfram envisions multiple final ranking providers so that users have a choice of how they want their content selected. We hope to offer a compelling option to users as the volume of content in our feeds increases.
If you have feedback on ideas to improve our model please leave a comment below. Thank you.
Postscript: On a personal note, as a computer engineering student in the ‘90s I had heard of Stephen Wolfram’s work with Mathematica, though at the University of Waterloo we used Maple and Matlab - two competing software packages. Wolfram occupies such hallowed halls of computer science that even now I am hesitant to think anything we've built might be relevant to such a giant in our industry. If we have indeed built what he envisioned I concede it is coincidental and I hope that it measures up to the scrutiny he would subject us to.