where connection went

puzzle: despite billions $ + decades of attempts, no social platform provides sustainably ¹satisfying social rewards. the graveyard is large (rip clubhouse, houseparty, among us, vine, google+) ² plenty create initial satisfaction but they inevitably degrade. why? ³

covid era as a natural experiment:

the pandemic was an interesting test of some of this. looking back there were a lot of things that didn't quite fit the models we had:

• zoom felt worse despite less total interaction. ⁶
• gaming communities reported increased satisfaction.
• video calls are theoretically higher bandwidth but felt worse (anecdotal).

contradiction? or reveals something about what matters. personal anecdote: during isolation, I discovered something unexpected: long phone calls while walking around my neighbourhood became one of my deepest form of connection. not quick catch-ups rushed between meetings or classes, but sometimes 3-5 hour conversations with family and friends (including ones I'd struggled to meaningfully connect with due to geographic distance). what's counterintuitive isn't that these calls worked – they worked better than pre-pandemic interactions. without the pressure of limited time or the distractions of a physical setting, conversations reached surprising depths. key differences:

• removal of visual attention demands
• parallel physical activity
• time zone differences creating unusual interaction windows
• forced removal of usual social alternatives

platforms seem to misunderstand how human connection works. they focus on delivering discrete high-intensity moments for connection—comments, facetimes, livestreams, etc. but if we think to prehistoric humans, connection came from simply being around each other, and as a result, our evolved reward circuitry would expect a continuous low-bandwidth presence. existing in the same space, sharing the quiet moments, feeling plugged into each other's lives. this isn't about dopamine hits vs. "real" interactions. specifically, i mean: social bonds via ambient presence, not scheduled intensity. feeling connected to our people, so maybe connection strength is better modeled as an integral over time rather than a sum of peak moments.

we could think about connection strength as an integral over time? something like:
connection_strength = (∆U/∆t)dt
where ∆U = reduction in uncertainty about other's mental state..

connection then isn't about the intensity of individual interactions, but the continuous reduction of uncertainty about each other over time. it's not how much you learn in any moment, but how that understanding accumulates.

somewhat unrelated, but personally i've felt that a good litmus test for what deep relationships should aspire to isn't shared experiences or time spent, but building an accurate model of someone's inner world. like "perfect information" but for emotional states. my closest relationships are the ones where i can predict reactions, navigate complex emotions safely, and understand unspoken thoughts. this is why close friends can savage each other, but less close friends and acquaintances would never insult each other. this might explain our deep attachments to fictional characters and even favorite youtubers. we literally know their inner worlds better than most people in our actual lives. younger me felt closer to rand al'thor than many irl classmates for exactly this reason.

it's like the difference between processed food and real nutrition. current platforms provide "empty social calories"—immediate reward without satisfaction.

the future

I think bereal was the most interesting social app of the last 5 yrs and I used it almost daily in 2022-23. skeptics correctly predicted it's failure, but they were wrong about it being the newest photos app wrapper. it was the first social product that got tens of millions of people to share normal moements over highlights. the app eventually died, but it proved something important: people are hungry for authentic digital connection. we just haven't built the right thing yet, and i think we're getting closer.

the misalignment in current social platforms isn't technological but neurological—we've built systems optimized for dopaminergic spike-based engagement (maybe read connection?) when our social reward circuitry evolved for tonic, not phasic, activation patterns. Cf. dunbar's work on grooming behavior in primates → sustained low-intensity social bonding creates stronger neural coupling than intermittent high-intensity interactions. bereal's accidental insight: by forcing synchronous, mundane sharing, it approximates the background presence that characterizes natural social groups. but this is just touching the surface of a deeper principle about human-computer interaction... (add more here?)

watching how gen-z uses tech gives us clues: we oft keep facetimes running while studying just to feel present and connected analagous to how primates maintain cohesion through awareness rather than direct interaction. similarly, the rise of virtual coworking points the same way. See focusmate.com, cofocus.one, timeivy.com. our current social platforms are operating in a local maxima that fails to account for deep structure in human sociality, however the uptick of these examples suggests strong selection pressure for tools that better match our evolved social architectures.

questions to revisit:

• what's the minimum information density threshold for satisfaction?
• how to measure/optimize context preservation?
• is "perfect" digital social reward theoretically possible?

also another random thought to come back to/work in: spacing effects seem important but mechanics seem unclear. maybe a connection to memory consolidation? spaced repetition?

feels like we're finally starting to understand core principles. next wave of social tech might actually work with rather than against evolved architecture.

what would "human-aligned" social platform primitives look like? start from first principles rather than engagement metrics.

misc thoughts/notes

we could think about connection strength as an integral over time? something like:
S(t) = ∫ q(t) · v(t) dt
where q(t) is quality of interaction and v(t) is... vulnerability? authenticity? something like that. how "real" the interaction feels.

breaking down q(t) further:
q(t) = b(t) · (1 - l(t)) · m(t)
where:

• b(t): bandwidth (bits/sec)
• l(t): cognitive load (0 to 1)
• m(t): shared context

interesting implication: you can accumulate more connection with low bandwidth + low load over time vs high bandwidth + high load in bursts. matches the walking calls observation.

thinking about bandwidth based on miller's ±7 stuff ⁷.
text: maybe 2-5 bits/sec
voice: ~20-40 bits/sec (speech rate ~150 wpm × ~6 bits per word)
video: theoretically 50-100+ (facial expressions + body language + voice)

but raw bandwidth is reductive. there's probably a utilization function.

u(b) = min(b, k·(1-l(t)))

where k is how much we're actually able to process (~40-60 bits/sec?).

this might explain why voice feels better than video (at least for me) which initially seemed counterintuitive. the raw bandwidth is higher with video (~50-100 bits/sec). probably relates to attention economics: video demands constant visual processing overhead while voice allows you full focus on the conversation's content. mirrors findings from blind musical auditions and voice-only negotiations outperforming face-to-face interaction ^{8 9}. i suspect this effect is amplified for those of us with adhd-type attention allocation - the high task-switching penalties make purely auditory input more efficient, though i acknowledge this is mostly personal observation mapped onto known mechanisms ¹⁰. the rapid growth of podcasts (~135m monthly listeners above 12 yrs in 2024) ¹¹ could suggest that perhaps we're collectively discovering that voice hits a sweet spot in the attention-bandwidth trade-off, for complex information processing. though one wonders if podcast listeners skew toward certain cognitive styles...

the vulnerability/authenticity thing
v(t) = v0 · (1 + ln(1 + T))
where T is cumulative "real" interaction time.
logarithmic because deep connections keep growing but with diminishing returns?

this actually explains a decent amount:

• why bereal felt different (synchronized context, lower cognitive load, higher authenticity)
• why discord's great (sustained low-bandwidth presence, strong context)
• why instagram feels empty (high bandwidth but low authenticity, poor context preservation)