Key takeaways
- Many games with the shape of an io arena are populated entirely by AI rivals, which is why a run can begin the instant you tap instead of waiting on a lobby to fill.
- Researchers building human-like agents list the giveaways from the other side: bots tend towards repetitive patterns, superhuman precision, an absence of idle non-goal-directed movement, and no frustration after a loss.
- Authored opponents are learnable opponents. The four ghosts in Pac-Man remain the clearest demonstration — each runs a different targeting rule on the same map, and all of them alternate between scatter and chase on a schedule.
- Arena AI that hunts food and other rivals rather than only the player creates a field with its own dynamics, which you can exploit by letting contests resolve without you.
- What AI opposition cannot offer is genuine adaptation between runs. A bot will not learn your habits across a week, so a repeatable exploit stays repeatable — a strength for consistency and a limit on depth.
A screen full of small moving shapes with names over them and a live top-ten board looks like multiplayer. Frequently it is not. A large share of games built in the io arena style populate the field with AI rivals, and the tell is usually the thing players like most about them: the run starts the instant you tap, with no queue, no lobby and no waiting for a match to fill.
That is not a criticism. It is a design decision with specific consequences, and knowing which side of it you are on changes how you should play.
Why the question is hard to answer from the screen
The genre borrows its entire visual grammar from real-time multiplayer. Bodies of varying size, a leaderboard updating as you play, a crown on whoever is currently largest — none of that requires another human being, and all of it signals one.
The honest answer for any individual game comes from the store listing and the in-app description. Jellyfish Arena Survivor io, which is our own app, states it plainly: the rivals are AI-controlled, so it has the feel of a multiplayer arena while staying a solo experience that starts immediately. Up to a hundred rivals fill a standard run, and they hunt food and each other rather than converging exclusively on you.
The practical consequence of that architecture is the one players notice first. A real-time match cannot begin before enough players are matched and connected, which puts a floor under how short a session can be. An AI-populated arena has no such floor, and it also has no dependence on someone else's connection quality for how the opposition moves. The related question of whether a game works with no connection at all is a separate one, and we covered it in what bots change about playing io games offline.
What researchers say gives a bot away
The clearest inventory of bot tells comes, unexpectedly, from the people trying to eliminate them. A recent survey of the challenges in building human-like agents for game environments sets out thirteen obstacles to making an agent indistinguishable from a person, and read backwards the list is a field guide.
Bots typically show precision no hand produces — aim or movement that is consistently exact rather than approximately right. They lack the idle, non-goal-directed actions humans generate constantly: drifting, hesitating, circling something with no plan, abandoning an approach halfway through for no reason. They follow repetitive patterns, because a behaviour that works in a situation will be selected again in the same situation. They operate without reaction-time, vision or memory constraints unless those are deliberately simulated. And they do not tilt — a bot that loses a contest does not then play recklessly for thirty seconds, which is one of the most reliable human behaviours there is.
The survey traces the formal version of this question to the 2K BotPrize in 2008, which adapted the Turing test to a first-person shooter and asked judges to distinguish humans from bots by watching them play rather than by conversing with them. The competition existed because the distinction is genuinely difficult at a glance and genuinely obvious over a few minutes, which matches most players' experience.
Authored behaviour is learnable behaviour
The most useful thing about facing AI is not that it is weaker. It is that its behaviour was written down by somebody, which means it is consistent enough to learn.
The canonical demonstration is nearly half a century old. The four ghosts in Pac-Man share one maze and run four different targeting rules, documented in detail in Jamey Pittman's dossier on the game. Blinky uses Pac-Man's current tile as his target — the simplest and most direct of the four. Pinky aims at an offset four tiles ahead of Pac-Man in the direction he is travelling. Inky runs the most complex scheme, requiring both Pac-Man's position and Blinky's to compute a target by doubling a vector. Clyde switches behaviour on proximity, heading for Pac-Man when far away and retreating to his own corner when close. On top of that, all four alternate between scatter and chase modes at set intervals, with scatter occurring four times per level before chase becomes permanent.
The reason that example still matters is what it proves about the category. "AI opponent" does not mean one behaviour repeated. It can mean several distinct agents with different rules producing emergent pressure that feels coordinated and is not. And every one of those rules is, in principle, discoverable by a player paying attention.
Craig Reynolds' steering behaviours supply the modern vocabulary for the same idea: seek, flee, pursue, evade, wander, arrival, obstacle avoidance and separation are composable building blocks, and Reynolds notes that combinations of them can be used to achieve higher-level goals. Wander in particular is what produces the plausible aimless motion whose absence the survey above flags as a tell — which is a neat illustration that the gap between "obviously a bot" and "possibly a person" is often one behaviour wide.
What this changes tactically
Three things follow if the field around you is AI.
The field has its own business. Where rivals hunt food and each other, contests happen whether or not you participate. That gives you a real option — let two larger rivals resolve a confrontation and collect afterwards — that simply does not exist in an arena where every AI converges on the player. It also means the field thins and refills continuously, so the composition of threats near you changes without anyone reacting to you at all.
Consistency is exploitable, once. Patterns that repeat can be planned around within a run. A rival that always commits to the nearest food cluster is a rival whose destination you know, and interception works on destinations. This is the flip side of the point made in how momentum steering punishes straight lines: predictability loses, and it loses for bots too.
Nobody is adapting to you between runs. This is the real limit of AI opposition, and it deserves stating honestly. A bot will not notice that you always approach from the same angle and start punishing it next Tuesday. That makes difficulty stable and sessions repeatable, which is exactly what a short-session game wants — and it caps how deep the opposition can get. Consistency is a feature and a ceiling at the same time.
Choosing on the right criteria
If you want opposition that learns your habits, argues with you and occasionally does something no model would predict, you want people, and you should accept the queue and the connection dependency that come with them. If you want a run that begins the moment you tap, behaves the same at midnight as at midday, and never requires a lobby, AI opposition is not a compromise version of that — it is the thing that makes it possible.
Jellyfish Arena Survivor io is built by Reign Creative on the second premise: an ocean arena of AI rivals, a live top-ten board, a minimap for reading the field, and pearls from each run that unlock skins whether the run went well or badly. The app page has the full feature list. For the genre in general, what the name "io game" actually means is the place to start, defending a lead once you reach the top is the late-run problem, and the rest sits under action and arcade games.
Good to know
Frequently asked questions
Am I actually playing against other people in an io game?
How can you tell whether an opponent is a bot?
Why do AI arenas start instantly when multiplayer games make you wait?
Do AI rivals fight each other or only the player?
Sources
- The Many Challenges of Human-Like Agents in Virtual Game Environments (opens in a new tab)arXiv (preprint) — accessed
- The Pac-Man Dossier (opens in a new tab)Jamey Pittman — accessed
- Steering Behaviors For Autonomous Characters (opens in a new tab)Craig W. Reynolds — accessed
About the author
Reign Creative LLC — The independent app studio that builds the apps covered on this site and publishes every article here. On Google Play it publishes as Reign Collective Apps.
We use AI tools to help research, draft and edit our articles. Each article lists the sources its factual claims rely on, and details about our apps come from their live Google Play listings. No article on this site has been reviewed by an independent expert unless the reviewer is named on it.
Spotted an error? Tell us — how corrections are handled is set out in our editorial policy.