Information Wants to Be Found.
Algorithms to make Elon Musk's Twitter acquisition more than a vanity project
Tying strategic business decisions to grand ideals is risky in the best of times. “Free speech” is as nebulous as ideologies get. Twitter has at times resembled a gigantic echo chamber of hate; allowing QAnon vitriol to propagate before the 2020 election, then abruptly banning more than 150,000 accounts after the January 6 insurrection. The service lost money in 2020 and purging its far-right membership didn’t exactly solve the problem.
Elon Musk has been pretty clear that he is making this $44B purchase with motives other than pure profit, but surely there must be some way to turn Twitter around. A few hours after the deal was announced on Monday, an empty public repo folder mysteriously labeled “algorithms” appeared on the Twitter Github account. There were no no way to contribute. The folder vanished a few hours later and was apparently a joke, but assuming the new regime makes good on its promise to open-source its algorithms, here are three high-level suggestions:
Go for the Venn. Existing social sharing algorithms tend to polarize online communities and privilege the loudest, shrillest voices. The reason can be demonstrated mathematically; this model (to which I was a contributor) suggests that when nodes passing on information and making decisions represent people, a totally distributed social organization will tend to favor more exclusionary, polarized viewpoints over time. If Twitter wants to differentiate its service, it should find ways to promote “Venn Diagram” content that appeals to the middle, and not the edges.
Long-tail marketing has had its moment in the sun. As a society and as a culture, we really need to find ways to rediscover our common ground. Think Prince and Outkast; there is nothing uncool about “crossover appeal.”
Don’t sweat the bots. In an age when AI services exist to automatically write blog posts and other online content, when social media managers may manage thousands of accounts for clients, there is nothing magical about human-created authorship. One of my favorite photo accounts, @archillect on Twitter, is entirely AI-curated. Creating new levels of CAPTCHAs, making it difficult for users to maintain multiple accounts, and introducing other time-consuming and burdensome identity tests obscures any privacy advantages Twitter may once have boasted and will only drive users away.
Click fraud remains a real problem for any business model based in online advertising, but it is basically background noise. Fake bot traffic may result in some slight distortions of CPM for impression-based advertising campaigns but large advertisers certainly aren’t going to abandon Twitter in droves; this sort of phenomenon is everywhere on the open web and has very little bearing on user experience. A better way to establish Twitter as a premier online destination is to promote content which will appeal to users and keep them engaged and interacting. This brings us back to the algorithm conversation.Speech is not a commodity. Free speech is meaningless when it is not seen or heard. When the supply of information is free and almost infinite, our current social sorting algorithms fail miserably at separating high-value content from clickbait and spam. News reports that frame the value of Twitter’s service in terms of its 217 million daily users miss the point; what differentiates Twitter is content strikingly different from the pablum and family photo albums found on larger rivals Facebook and Instagram. Relevant, topical many-to-many conversations spanning politics, business, and technology are what have kept Twitter alive over the years.
Now all the company needs is a way to amplify signal vs. noise—in short, an algorithm for merit. The concept of “merit” is threatening to some. But without some means to evaluate and gauge the relative importance of information, all content (on social media and elsewhere) looks the same to machines. New and fresh viewpoints have no way to find an audience. How to solve for merit? It’s actually not that difficult. Simply create a model that predicts bias among users; then weight results to correct for that bias. Remember that bias means more than racism, sexism, and ableism. Machine learning has been used to detect political bias in news media going back as far as 2018.
Rather than suppress the most extreme offenders, do free speech advocates a favor and turn up the volume for those currently getting shouted down. Good ideas will outcompete bad ones, but only if they are heard in the first place.
Further background, source materials, and preliminary work for creating algorithms to counter the problem of network bias may be found at networkbias.com.
Lotus Rose is not a registered investment, legal or tax advisor or a broker/dealer. All investment and financial opinions expressed on Lotus Trader are from the personal research and experience of the owner of this newsletter and intended for educational and informational purposes only. Although every effort is made to ensure that all information is accurate and up-to-date, occasional unintentional errors may occur.