Internet Has Nothing to Bet On (Part 1)

When you open a prediction market like Polymarket or Kalshi today, you’ll see most of the liquidity and volume accumulate around real-life events: game scores, election results, weather reports. These markets clearly have product-market fit — not only do traders love playing them, but markets are doing their job: aggregating staked opinions from a large cluster of anonymous actors, in real-time.

Among these topics, election result is the most interesting, because it is trying to predict a big group of people’s behaviour. Polling and surveys are expensive, time-intensive, and slow to update. Prediction markets offer an alternative data point — more like “what do predictors think voters think,” not “what voters actually think” — but more robust and continuous, and their popularity is very much self-proved.

We don’t need to decide who to vote everyday, but we do make numerous decisions to allocate our time, money and trust, with an increasing amount of them made online. Billions of people decide whether to retweet this post, listen to this new song, or join this fundraising campaign for a new water bottle. These activities produce what I’d call emergent outcomes from anonymous collective behavior — outcomes that only exist because the Internet, as an engagement platform, enables mass anonymous coordination.

Yet if you look at Kalshi, there are very few markets with meaningful liquidity that let users bet on these online collective behaviour, and they are either movie or music related, such as Artist(s) with a #1 Song on Spotify USA: July or Odyssey’s Rotten Tomato score. Mention markets, which let users bet on who says what, might seem like a step in this direction — but what’s being predicted is an individual’s publishing act, not an outcome from collective behaviour. What’s genuinely new — millions of anonymous actors interacting, reacting, amplifying — remains largely unpredictable.

I used to think the core problem was cheating. Clicks, views, likes — the marginal cost of faking any of these are close to zero. After some research, however, I found that this is not the case. Both examples above are strongly tied to one particular platform, meaning these markets are offloading anti-cheating practices to the referred platforms, trusting them not to inflate the numbers. This makes sense because platforms like Spotify and Rotten Tomatoes have subscriptions or verified accounts, which functions as some kind of proof-of-stake design — enough friction to make manipulation costly.¹

Maybe the real challenge brought by frictionless participation, is that emancipatory content creation and community organization do not leave us with a center stage like the physical world. Instead there are millions of niche spaces, divided by algorithms, each with its own micro-audience. A tweet going viral in tech circles is invisible to K-pop fans. A Twitch streamer’s milestone means nothing to the Substack crowd. There just isn’t enough common ground to sustain a liquid market.

On the contrary, events like World Cup or general election takes only once every four years, and is an amalgamation of money, human-hour, venues, transportation, and nationalistic honor. They are bounded events with massive shared attention, and the entire audience agrees on what happened. This boundedness creates natural focal points — events worth betting on because enough people care about the same thing at the same time to create liquidity.

And scarcity is not only about the frequency of an event. The exclusivity of access to the event also plays a part. You need expensive ticket to watch a football game, or an American citizenship to cast your vote. This scarcity give birth to a big group of spectators, the vast majority, in fact, only able to watch but cannot directly participate. They are the ones who broadcast the event into the wider society: coworkers who don’t follow football get pulled in, taxi drivers have opinions.

It’s precisely these spectators — not the core fans — who give prediction markets their thickness. The overflow creates a casual layer — people with low-conviction hunches who are willing to place a small bet. The bet is the only way a spectator can have a stake in the outcome.

Online communities doesn’t run like this. K-pop fandoms rival football audiences in sheer size, but their energy is self-contained: intensely felt within, nearly invisible without. No one outside the fandom is idly wondering whether aespa or IVE will top the chart this week — because if they care, they would’ve clicked and streamed and become fans already. Without that layer of curious bystanders, a market stays thin.

And even if you could attract enough liquidity, you’d face a deeper problem: the gap that prediction markets fill in the physical world — a way for spectators to participate in something they can only watch — simply doesn’t exist here. On the Internet, money buys participation directly. Buy the single, it counts toward sales. Open a premium account, stream on repeat, it counts toward the chart. These are the same proof-of-stake mechanisms that platforms use to prevent manipulation, but they are also the channels through which fans step onto the stage. A football fan can buy a jersey, buy a beer, scream at the TV — none of it gets him any closer to the pitch. A K-pop fan opens Spotify and she’s already playing.

This post is not written to propose new growth source to prediction market companies, but to discuss the nature of our online collective behaviour. At least we are safe to say that it is, and will never be a mirror of the physical world. There’s no such thing as “mainstream” online, as everything is long-tail, including popular ones like K-pop.

AI agents could, in theory, trade thousands of niche internet markets simultaneously — no shared attention needed, making sub-dollar points. But this concession proves the point. If the only path to liquidity requires replacing human participants with bots, that’s an admission that humans don’t share enough common ground to sustain these markets themselves. And in practice, anyone capable of building such agents now would rather deploy them in larger, more liquid pools. The fragmentation problem doesn’t disappear — it just moves up a level.

If there is a way forward, I suspect it lies not in building the online version of Super Bowl, but learning from communities that are so good at mobilizing people and create shared focal points where none exist naturally. Bitcoin has a potential answer for that, and we will discuss later.