The most important aspect of today's media are the recommendation systems that guide our attention and ultimately our thoughts, opinions & culture at scale. Most algorithms we use are black boxes - we don't know what they are nor how the ML models are parameterized & trained and if there's any (intentional) bias. And even if that was public information - there would be no way to check because we don't have access to the data either. Furthermore, nobody can build new competing indexes, models & recommendation engines and we're left with the lowest common denominator that optimizes mostly for engagement & time spent in the attention economy.
Is it a coincidence that one of the most important papers in AI/ML that introduced transformers was called
"Attention is all you need"?
- Verifiable algorithms, indexes & models
- Feeds & home pages
- Signaling preferences
- Aggregate sentiment - the big picture
Within Headjack, all public data is freely accessible and the data network effect is shared by every actor in the ecosystem - anyone can train new ML models, compute views and analyze data in new ways. There will be demand for all kinds of indexes & models but because entry is permissionless a market will appear and the ones that process data the most efficiently will be rewarded. The results of such batch processing jobs could be public or private (available only to those who pay for access) and the product would be verifiable - either optimistically (trust by default but with the option to recreate them and check for equivalence at some cost) or through verifiable computation.
"Rather, many, many different individuals and organizations would be able to tweak the system to their own levels of comfort and share them with others—and allow the competition to happen at the implementation layer, rather than at the underlying social network level." - Protocols, Not Platforms: A Technological Approach to Free Speech
We have no control over what social media shows us - we're presented with their best attempt at engaging us and nothing more. Feeds are primitive/limited and have their own agenda. Once access to data has been democratized we could finally have a choice as new entrants will differentiate themselves in ways that weren't possible before. What if:
- you want the first thing you see to be a dashboard with graphs & charts about the things you care about instead of an endless feed to scroll?
- instead of a feed you were shown a map with events grouped by category/people/activity so you could choose selectively which zones to "zoom" into based on your interest?
- you're interested in all events from the past week - not just those from the last 24 hours?
- you want things sorted differently, or to tune out certain kinds of content?
- you want your video feed to show content that people you follow have engaged with?
Events coming from an application are an advertisement for it because the URIs point to it as the source - users are effectively signaling what UI, filtration & content moderation they prefer.
But this can be pushed a lot further - using public message types more specific preferences within one application (and compatible with others) could be signaled. The choice of recommendation algorithms, filtration criteria & moderation levels could be displayed in account profiles throughout all applications either as badges or in some other way. An application could accommodate a wide range of preferences by using different indexes & models which would be handled by infrastructure layers beneath them - reusable by all other applications that need them too.
This way of signaling fine-grained preferences could allow us to collectively migrate to better algorithms - by showing which interests & perspectives we've adopted ourselves. One value to signal would be to use algorithms & models that are transparent and not some black box - we could study how virality happens. What if some algorithm promotes antisocial ultra viral content close to the borderline of acceptability just for the sake of engagement? We could boycott it. And why not choose indexes & algorithms that up & down regulate specific accounts? Moderation could be done through a set of filters to which users opt in/out by toggling & layering - choosing what to amplify and what to tone down. Such preferences can also be private.
Currently social media skews our perception of the world because the vocal become viral - most people are quiet and rarely (if ever) post anything, but they do consume. We could let them signal their preferences and better gauge our values. Preferences are a form of expression.
TODO: https://twitter.com/balajis/status/1581635513886253057 https://pbs.twimg.com/media/FfMYkpLVEAAOo_d?format=jpg&name=large
With open data, systems could show the overlap of communities and focus on what's shared and unites them - bridging the gap, making “the others” less foreign, and reducing polarization. We should be taking notes from Taiwan's civic hacktivism and their Computational Democracy Project & G0v movement - building better tools for social consensus & relaxing the culture war. Everything could be visualized through dashboards & graphs - including global heat maps of interests & activity based on topic/region/etc. Learning from history & making analogies could be much more quantifiable and precise if everything was in the ledger of record. We could leverage AI to surface human collective intelligence at scale in a transparent way.
"The general trajectory of institutionalization associated with steadily increasing specialization, urbanization, and bureaucracy may mean that mass media will continue to rise in importance, playing the role of the juicy gossiper in our increasingly separated existence from one another." - The Importance of Gossip Across Societies