Before we get to the trolls, consider the cynics, who wonder if the change, apparently in place since mid-August, is designed to protect the site’s planned billion-dollar investment in its prequel “The Lord rings”.
OK, companies want to make a profit. But I’m skeptical of the claim that the money spent on “The Rings of Power” drove that decision.(1) Amazon has long used machine learning tools to alter average ratings on his site, demoting posts from those believed to be trolls. , and elevating others – especially those who have actually purchased the affected product. Somehow the republic survived. So when Amazon says it’s embarked on a troll hunt to improve the accuracy of show ratings, I’m inclined to take the company at its word.
But is politics really necessary? And will it help? The answers depend in part on a third question: Why do trolls troll?
Let’s start with online reviews. In theory, their function is to resolve an information asymmetry. In general, sellers have more information than buyers, a challenge that makes buyers wary. Common solutions include branding (if you visit McDonald’s, you know what to expect) and warranties (if you’re not satisfied, return the product for your money’s worth). Another is in-store product review, which is impossible online.
Hence the importance of reviews. In online leaderboards, anyone can play. In a perfect world, these star ratings and user reviews would help resolve the asymmetry by giving potential buyers the perspectives and opinions of actual buyers. Alas, the world isn’t perfect, and neither are online reviews.
Challenges are everywhere. The most obvious is bias. Data shows that buyers at the extremes are more likely to post. Users who hate the product will show a low rating. Users who love the high-end product post. High-end users tend to be more numerous, as the group includes those whose attitude towards the product was favorable before purchase. Meanwhile, the moderates don’t bother to play: hardly anyone posts in the middle. The result is the famous J-shaped curve, which renders the mean and median scores meaningless.(2) (Consumers are obviously aware of the bias and try to adapt to it.)
Trolls, whether human or bot, present a different problem. Always at the height of the threat of comments, often angry and dismissive, often writing in highly offensive terms, trolls are familiar to anyone who has ever been on the internet. A traditional view holds that the online troll not only delights in the expression of unreasoning hostility, but also in posting first, so that its ratings are at the top. What matters is to find and outrage an audience. In this sense, a useful analogy with the physical world is vandalism.
Trolls are also said to delight in causing consternation and even pain to other users. Some researchers claim that trolling suggests a particular personality type, that even those who engage in this behavior offline tend to “everyday sadism.” According to this theory, Amazon’s decision to impose a lag period on publications may well help to eliminate the extremes, at least at the bottom of the scale. For the troll, the waiting period could drain the exercise of its pleasure.
But is the research correct? Recent work suggests that the daily variation in the user’s mood, as well as the tone of previous posts on the same site, influences the tone in which the user posts. The more hostile the comments are already, the more likely the user is to join the trolls.
If this view is correct, what we consider trolling might simply be the complex response to a set of emotional stimuli. The catharsis does not gain so much of an audience as the display itself. In this case, being forced to wait will not change much.
We are therefore faced with a difficulty. If the 72-hour detention is meant to deter trolls, so much the better. But if, as the company has implied, the waiting period is to allow it to weed out trolls, then maybe not.
Whether done by humans or a sophisticated AI or (likely) both, the effort will inevitably run into difficulty in distinguishing between the user who really, really hated the program and the user who just tries to cause problems. There is a risk that the process will simply transform the familiar J-shaped curve into a unimodal distribution where surviving reviews are heavily skewed towards the positive end.
I’m not saying that will happen. Amazon is a smart company and, based on past experience, might just strike the right balance. But just in case, let me offer some alternatives.
First, Amazon could try to crush trolls by persuading the silent majority to get involved in rating streaming videos and other products. How could he do it? Studies show that financial incentives increase the number of individual reviewers. This strategy avoids the challenge of choosing what to remove.
Second, Amazon could deny the potential troll the reward that comes with venting. Instead of waiting three days to post user ratings, make users wait three days after watching a show before rating it. Not only might trolls get bored and go haunt another site; the notes could be more accurate, as they are informed by a cooling-off period.
Finally, Amazon might do nothing. No post delays, no troll hunting, no baby bath water issues. I recognize the risks. When a high-profile project like “Rings of Power” garners good reviews from critics and equally low reviews from viewers, there’s reason to be suspicious of who’s posting. But the premiere drew 25 million viewers on its first day. I think it will survive a few trolls.
(1) Full disclosure: I like JRR Tolkien and I like the Amazon version.
(2) The common (if illegal) practice of buying positive reviews is a separate matter.
This column does not necessarily reflect the opinion of the Editorial Board or of Bloomberg LP and its owners.
Stephen L. Carter is a Bloomberg Opinion columnist. A law professor at Yale University, he is the author, most recently, of “Invisible: the story of the black lawyer who shot down America’s most powerful gangster”.
More stories like this are available at bloomberg.com/opinion