The Algorithm Behind ‘Steal a Brainrot’

Why Roblox Has Become a Casino for Kids

There was a surreal moment on Hard Fork recently when Roblox’s CEO David Baszucki, in the middle of a conversation about child safety, floated the idea of adding prediction markets to the platform. The comment was overshadowed by his high-fives, but imagining elementary-school kids speculating in miniature “markets” is admittedly kind of hilarious and unlikely to work in practice. Yet, it also highlights something real: the digital environments we build are increasingly shaped by the convergence of algorithms and market mechanisms that sometimes bleed into gambling.

If we want to understand the future of liberty in an algorithmic world, we have to understand how children, our most-online citizens of the future, encounter these systems today. Roblox is one of the first “algorithmic nations” they inhabit (an astonishing 50 million daily active Roblox users are under 13 years old.) Its recommendation feeds determine where they go; its mechanics and rules determine how rewards arrive; its social architecture determines when peer pressure kicks in and what habits get reinforced.

What does this look like, and how will it play out? To get a sense, I spent 100s of hours playing Roblox and similar games with my kids. On the professional side, I also spent four years as an advisor to the President of Global Affairs at Meta thinking about some of the most difficult governance problems related to algorithms and society. Here’s a bit of what I’ve learned from this “field work” and how I think games like Roblox can chart a path forward amidst these concerns.

Grow a Garden, Steal a Brainrot, and online gambling for kids

Grow a Garden is a runaway hit on Roblox that has quietly become one of the most played experiences on Earth (over 33 billion plays and counting, according to the game’s homepage). It looks like a cozy farming game, but if you look more closely, its underlying loops are designed to optimize revenue for developers within the rules Roblox has set regarding how developers are compensated.

Indeed, to understand Grow a Garden is to understand Roblox’s economic engine, and to understand that engine is to see the larger challenge of governing algorithmic systems in a world where the rules that matter most aren’t written in law, but embedded in the platforms that our children use every day.

In Grow a Garden, your plants grow when you’re offline, encouraging you to come back and harvest them regularly

The basics of the game are straightforward: you buy seeds, plant them in a small plot, wait for them to grow, then harvest and reinvest your earnings to unlock bigger plots, rarer seeds, pets, and decorations. Fun and potentially educational! As I quickly learned as my son taught me how to play, though, beneath that surface, the game relies on mechanics that feel familiar from both gambling and modern engagement algorithms.

• Randomized rewards. Many rewards are randomized, such as seed boxes or pet eggs with uncertain outcomes and low-probability “rare” items. The game uses intermittent reinforcement by mixing predictable progress with occasional surprise bonuses that encourage frequent check-ins.

• Time-gating and pay to skip. To make progress players have to wait for their plants to grow, but you can use Robux (the in-game currency of Roblox, which users purchase with real money) to increase the rate of growth in a variety of ways.

• FOMO. Players can get free rare prizes if they log onto live, weekend events hosted by the game’s creators (and once, Travis Kelce). If you miss the event, you don’t get the reward.

• Social pressure. Players can see each other’s gardens, and much of the competition is about having the coolest garden with the rarest items.

• Stealing. Players can steal plants from one another.

It’s obvious where this goes. Elementary school children are not well conditioned to see how the game is manipulating them, or to handle the emotional ups and downs of gambling.

As evidence of this claim, I submit this viral YouTube video with over a million views that chronicles the many freakouts and tantrums young kids have had after losing cherished items in Steal a Brainrot, a similar Roblox game that is nearly as popular as Grow a Garden now.

The economic structure of Roblox

When you see how Roblox’s economic engine works, you immediately understand why games like Grow a Garden look the way they do. Roblox creates incentives for developers, and developers build games that respond to these incentives.

First, developers need visibility. Roblox is essentially an algorithmic marketplace: many millions of games compete for placement on a handful of high-traffic surfaces such as the home page, “Popular,” and “Top Earning.” Roblox reports that more than 90% of traffic is on the Home page, as of 2024. This is where players find their games.

As an example, here is what my homepage on Roblox looks like today.

Getting into those slots depends heavily on engagement signals. Digging into Roblox’s documentation details shows that it is not only how much time per user is spent in a given game, but how many unique days per week a user plays, how much Robux they spend, and how frequently they play the game with friends.

This all means that developers want to create games that players want to play, which makes sense—but also want to play on a daily basis. All else equal, developers would rather each player play the game 10 minutes a day, 7 days a week, than 70 minutes once a week.

This helps to explain one of the most annoying features of many popular Roblox games: how they encourage users to log in each day to see how much their holdings have appreciated since the day before and to keep them growing (games with this structure are often called “idle games” because they keep changing while you’re not playing).

Some of the key design elements in Grow a Garden meant to create this need to log in every day include: plants grow in real time and have to be harvested and replanted to maintain progress; the seed shop rotates items and rarer sees only appear for short windows; and pets gain value over time, progressing faster if you log in regularly.

Any parent who tries to limit screentime for kids—for example, to weekends only—knows that these daily rewards can be incredibly frustrating. Children who play these games hate to miss out on their daily progress.

Second, developers make money when players spend Robux inside their game. Roblox takes a platform cut, and developers cash out what remains. So the core business model is simple: get kids into the game, and then convert some of them into paying users. Often this means selling time savers, speed-ups, randomized item boxes, and other mechanics that sit right on the edge of gambling.

If you want to get a feel for what I mean, here are screenshots of literally the very first thing I saw after joining the Steal a Brainrot game a couple days ago. I’m greeted by a “radioactive spin” which will give me a randomized reward if I spin it—oh but wait, I only get a free spin 1 hour and 27 minutes from now. If I want to spin it now, I must first go and buy a Noobini Pizzanini! And that’s not my only opportunity to spend: as you can see in the second picture, I can also spend Robux to increase my luck with the spinner.

Kids don’t necessarily understand probabilities or sunk costs. But they do understand the idea that they can spend Robux to get a better chance to win the made-up prizes they want, and to do it faster. Developers understand this too, and they tune their random number generators accordingly. Like a casino, you can be sure that the house always wins in this setup.

The result? A lot of kids are constantly begging their parents to buy them more Robux. IYKYK.

Third, developers must optimize for Roblox’s Creator Rewards. Separate from the in-game transactions that developers can monetize, they can also receive rewards from Roblox for certain behaviors. Roblox recently shifted from a pure “time spent” model to a more complex system that rewards two types of behavior: when users log into a game early in their Roblox session, and when users do something to share the game with others. This means developers are pushed to engineer for quick launches, daily streaks, viral sharing, and constant novelty.

This doesn’t imply that switching the algorithm back to rewarding total time spent rather than daily usage would improve things. Roblox says that they switched to this new algorithm because rewarding time spent too aggressively led to “grindy” games that encouraged players to spend lots of time working to earn points and prizes, and in particular created incentives for “AFK” play in which players leave their computers on and trick Roblox into thinking they’re playing more than they are. Clearly, the ecosystems evolves rapidly to respond to the structure of the algorithm, and a lot of care is required to balance things right.

But when you put this all together, you can see why the CEO of Roblox might think of prediction markets and Roblox as a natural pairing, and why it might not be the best structure for children.

A blueprint for shared governance of the Roblox algorithm

Roblox is an extremely rich ecosystem that leverages creativity at scale to create fun experiences for kids. The question is how to bring the right stakeholders together—the company, the users, the developers, parents, and the government—to design a system that works better for kids. Here are some ideas on how to do that.

What should be parents’ responsibility?

Parents are responsible for setting boundaries for their kids. Game companies are not a replacement for parenting, and if your instinct is “this is too much for my kid right now,” you’re allowed to act on that without waiting for Congress or Roblox Trust & Safety to bless the decision.

At the simple end of the spectrum, that can mean: no Roblox at all for a while; or Roblox only with friends and in certain games; or “no games with loot boxes, spins, random eggs, or stealing.” It can also mean a hard rule of “no Robux spending” until a certain age. If the economic engine is what makes these games so intense, cutting off the fuel is a perfectly reasonable and widely recommended response.

The other tool is explanation. Kids are smarter than we give them credit for. You can sit with them and say, in plain language: “Look, this is designed to make you want to come back every day. These spins are random. The people who made this game make more money when you feel like you almost won and want to try again.” Helping them see behind the curtain doesn’t magically solve everything, but it turns the game from an invisible environment into something they can be at least a little bit critical about.

What should Roblox do?

That said, it’s not enough to offload everything onto families and say “buyer beware.” Roblox is designing an entire economic and social system for children; that comes with real responsibilities. Here are some ideas of what Roblox could do directly.

Re-align the algorithm for kids. Roblox’s discovery and payout systems currently reward short, frequent, high-spend sessions. When you optimize for that pattern, it’s no surprise that your top charts fill with games built around time-gating, paid randomness, and FOMO events.

Roblox could explicitly tilt both its recommendation engine and its Creator Rewards system—at least for under-13 accounts—toward longer, more stable sessions and away from gambling-like mechanics. Think less “idle game that drips dopamine every thirty seconds,” and more “Minecraft-style experience a kid sinks into for an hour on a Saturday.”

This doesn’t mean curating Roblox into a museum of approved experiences. It means being transparent about the objective function. YouTube has moved partway toward a “quality watch time” metric, rather than pure clicks, in part because its revenue doesn’t depend on extracting micro-spends from children inside a game loop. Roblox could make a similar shift: elevate experiences that encourage creative, sustained play and de-emphasize those that function as elaborate slot machines.

Roblox has already acknowledged that its old “time spent” rewards fueled too many grinding and AFK experiences; the broader industry has developed better ways to incentivize deeper engagement that the company could draw on instead.

Experiment with different monetization models for minors. Minecraft makes substantial money up front: you buy the game, and any extras are optional. It’s a richer experience precisely because it doesn’t rely on constant in-game manipulation. Roblox is a more complex ecosystem, but it could create a “kids lane” of experiences clearly labeled as low-monetization, subscription-funded, or Robux-free. Many parents might gladly pay for games that are fun and social without constantly shaking their child’s pocket.

Increase transparency around how kids spend and what they encounter. Right now, parents and outside researchers have almost no visibility into basic questions: How much does an average under-13 spend in a month through different types of in-game purchases? What fraction of that goes into random-reward mechanics? How often do kids buy spins or eggs and immediately regret it? Roblox could publish anonymized data, run randomized experiments with independent experts, and then share the results publicly: “Here is what happens to kids’ behavior when we tone down these mechanics. Here is what happens when we don’t.”

Give parents simpler, stronger control tools. Families need tools that map to the reality of how kids use Roblox. Instead of making parents inspect every individual game, Roblox could offer system-level controls that automatically block games containing gambling-like mechanics—loot boxes, spins, randomized boosts, stealing, and so on. Anyone with a child knows that kids cycle through many games in a single session; parental controls should anticipate that, not force parents into doing their own detective work.

All of this would look less like “Roblox as moral referee” and more like “Roblox as steward of a system it knows is shaping children’s habits.”

What should policymakers do?

Policymakers are already circling this space: there have been several bills across the world aimed at kids’ online safety, rules against dark patterns, and debates about whether loot boxes amount to gambling. The temptation is always to reach for blunt instruments—ban this genre, outlaw that mechanic—and call it a day.

I think the better approach is to focus on structure rather than individual games. We don’t want to kill creative, messy, user‑generated game marketplaces for kids. Nevertheless, we do want those marketplaces not to be junk food by default.

That could mean rules that:

• treat paid, randomized rewards for minors the way we treat gambling, or at least require clear odds and spending limits;

• require large platforms to show that their recommendation systems for kids are not tuned for addiction metrics;

• continue cracking down on dark patterns that lead kids to spend money they don’t understand they’re spending.

We also need better research into the effects of these policies. For example, in countries that have banned loot boxes in kids’ games, have we seen observable shifts in how gambling-like the resulting experience is? Do kids find these games more or less enjoyable?

The point is not that regulators should design games. It’s that they should set the outer boundaries for what kinds of economic tricks are acceptable in products aimed at children, the same way we do in advertising and food.

And what about the kids?

There’s also a more hopeful angle here. Kids are not just passive subjects of algorithms; they’re future citizens of an algorithmic world. It would be a mistake to treat them purely as victims.

One interesting direction would be to bring children explicitly into the governance conversation. Roblox already has a “teen council” that ostensibly weighs in on issues of civility in the game (though the Steal a Brainrot video I shared above certainly suggests this council may not be super effective.)

Roblox could go beyond this and ask children to deliberate on questions like: “Should this game allow stealing?” “Is this spin wheel fair?” “What would make this fun without making people feel bad?” It sounds utopian, but it’s also a very concrete way to teach civic habits inside the systems where they already spend their time.

There is precedent for something like this. In the UK, the Children’s Commissioner has built a youth advisory board for online safety issues. The Australian government has built something similar. UNICEF has experimented with involving children in decisions that affect them in a variety of ways. Obviously, the risk here is that these devolve into pointless exercises in governance theater, and as old research on this topic explains, the key is to make their involvement meaningful and not window-dressing.

In the last few years, Meta, Anthropic, and OpenAI have all experimented with recruiting deliberative bodies of users to discuss important policy issues that affect them. I helped to develop the first ones at Meta—and I’ll be discussing the methods behind these experiments and what we’ve learned from them in a future post—but it is well worth considering how they could be extended to include children in Roblox. Outside of tech, experiments like this are already going on. In the Netherlands, for example, a pilot project uses sortition to recruit students to advise school principals.

The prospects for something like this for Roblox are good, because there is already a vigorous community of young Roblox commentators on YouTube who often discuss the games, their rules, the algorithms, and how it all affects them. As part of my research I’ve watched 100s of hours of these videos with my son, and I can tell you, these YouTube influencers understand way more about how these games work and how they’re affecting them than we do.

As an example, here’s a video with more than 500,000 views that explains quite lucidly why this player thinks that Grow a Garden is so addictive:

At a minimum, we should be teaching all kids basic digital literacy around these mechanics: here’s how random rewards work; here’s what the company is optimizing for; here’s what sunk costs are; here’s why it feels hard to stop. Those concepts are not harder than fractions or cursive. We just haven’t decided they belong in the curriculum yet.

Clear in‑game labels could help too. If a game includes paid random rewards, say that in plain language at the front door. If the expected value of a spin is less than what you pay, don’t hide that fact. Let kids see, explicitly, that they are entering a space where they will probably lose money over time.

What Grow a Garden teaches us about our algorithmic future

We are moving into a world where more and more of life is mediated by systems that look a lot like Roblox: algorithmic marketplaces, user‑generated everything, economic incentives wired directly into the product. The question is not whether those systems are good or bad in the abstract. It’s whether we design and inhabit them in ways that expand people’s freedom or quietly erode it.

At the same time that games like Roblox are becoming more marketized, our markets are getting more gamified. Although the Roblox CEO’s appearance on Hard Fork was ridiculed—primarily for his “high five” remark and his apparent surprise at having to field basic questions about child safety—we should be taking the proposal for prediction markets in Roblox seriously, because it reflects a real trend in society.

As I’ve argued before, prediction markets offer us an incredible informational tool that can give us real-time probability feeds about politics and the world. But embedding betting into every surface of our information environment will have its downsides, too. Exactly what those will be remains to be seen, and I’ll be doing more work on that in the coming months, but one clear place to start is with our kids.

Grow a Garden shows how quickly an innocuous idea can turn into a global, highly optimized behavioral machine once you plug it into the right economic engine. If we care about liberty in an algorithmic age, we can’t just argue about speech or content. We have to pay attention to the engines themselves—the incentives, the ranking systems, the reward structures—and decide, together, what we are willing to normalize for the children who are growing up inside them.