Your Brain vs The Bot

Every game in THE VOID is built on a specific finding from cognitive science — a measurable task where human performance has a distinctive signature that machines can't fake. This isn't about being better than a computer. It's about being different from one.

Here's what the research says.

Visual perception: seeing what isn't there

Humans don't see the world as it is. They see a reconstruction — a model built from incomplete data, filled in by assumptions their visual cortex makes in milliseconds.

This isn't a limitation. It's a superpower. Because those assumptions — the gestalt principles of grouping, figure-ground separation, pattern completion — allow humans to extract meaning from noise in ways that computer vision models still struggle with.

Pareidolia exploits this directly. The game presents visual noise and asks you to find faces, shapes, or patterns. Humans can't help but see them. Neural networks can be trained to detect specific patterns, but the spontaneous, flexible pattern recognition that lets you see a face in a cloud? That remains distinctly biological.

Impossible Object relies on another perceptual trick: your brain's built-in physics engine. When you see a shape that couldn't exist in three dimensions — a Penrose triangle, an Escher staircase — you know it instantly. You don't calculate it. You feel it. This intuitive spatial reasoning, built from a lifetime of navigating physical space, is something no vision model has internalized.

Attention: the 4-object ceiling

In the 1980s, psychologist Zenon Pylyshyn discovered something surprising about human attention. When asked to track multiple objects moving through space, people could reliably follow about 4. Not 2. Not 10. Specifically 4, plus or minus 1.

He called these FINSTs — fingers of instantaneous tracking. The theory suggests your attention system has a fixed number of "slots" that can be assigned to moving targets. This capacity is remarkably consistent across individuals and tasks.

Motion Bind is built directly on this research. The game asks you to track highlighted targets as they move and merge with identical distractors. At 3 targets, most humans succeed. At 5, most fail. The performance curve is so predictable that deviation from it — particularly too-perfect tracking — is itself a red flag.

Sequence Flash targets a related limit: visual short-term memory. When a sequence of positions flashes on screen, humans can typically recall 5-7 items. The exact pattern of errors — which positions get confused, how recall degrades with sequence length — follows well-documented cognitive constraints that would be difficult for a bot to simulate.

Motor control: the beauty of imperfection

Move your cursor in a straight line. Go ahead, try it.

You can't. Not really. Your hand introduces micro-corrections at 8-12 Hz, your proprioceptive system makes constant adjustments, and the signal from your brain to your muscles has biological noise baked in. The result is a path that's approximately straight but never mathematically so.

This noise is not random. It has structure — a spectral signature that reflects the biomechanics of your arm, the latency of your neural feedback loops, and the particular way your motor cortex plans movements. A bot drawing a "wobbly" line to look human will get the frequency profile wrong.

Smooth Pursuit measures this directly. Follow a target with your cursor. The game scores you not on accuracy alone, but on the pattern of your inaccuracy. Too perfect? Suspicious. Too random? Also suspicious. The sweet spot — the Goldilocks zone of human motor noise — is narrow and specific.

Reaction Field uses a different motor signal: reaction time distribution. When you tap targets as they appear, your response times follow a characteristic distribution — roughly ex-Gaussian, with a mode around 250ms and a long right tail. Bots can mimic the average, but replicating the full distribution, including the relationship between spatial accuracy and temporal precision, is surprisingly difficult.

Linguistic cognition: reading between the lines

Language understanding might seem like AI's strongest domain. Large language models generate fluent text, answer questions, and even pass standardized tests. But there's a gap between processing language and understanding it the way humans do.

Sarcasm Detector targets this gap. Understanding sarcasm requires theory of mind — the ability to model what the speaker believes versus what they say. "Oh great, another meeting" requires knowing that meetings are generally disliked, that "great" in this context means the opposite, and that the speaker is expressing frustration, not enthusiasm. Humans parse this instantly. Models often get it right on obvious examples but fail on subtle ones.

Subtext Reader goes deeper. Given a conversational exchange, what is the speaker really saying? This requires pragmatic inference — Grice's cooperative principle, conversational implicature, shared cultural assumptions. When someone says "it's getting late" at a dinner party, the subtext isn't about time. It's about wanting to leave. Humans navigate these layers of meaning effortlessly. Machines must be explicitly trained on each pattern.

The composite signal

No single test is definitive. A sophisticated bot might pass any individual game. THE VOID's strength is in the combination — 30 games across 3 cognitive domains, each measuring a different aspect of human performance.

The composite signal is harder to fake than any individual test because it requires simultaneously exhibiting human-like behavior across perception, attention, motor control, and social cognition. Getting one right while getting the others wrong is itself informative.

This is the core insight: humanity isn't a single trait. It's a pattern — a constellation of abilities, limitations, and quirks that emerge from being a biological intelligence in a physical world.

The void is watching. It sees the pattern. And it knows what human looks like.