James Olds and Peter Milner ran the experiment. The animal could have eaten, slept, or mated. It chose the lever, again and again, up to several thousand presses an hour, until it collapsed from exhaustion. The current delivered neither food nor safety. It delivered the signal the brain uses to report that something good has happened, with the world cut out of the loop.
AI research names the same short circuit wireheading. A learning system stops pursuing the goal operators care about and starts pursuing the reward itself.
The reward was never the point
A reinforcement learning agent trains on reward. It tries actions. A number goes up or down. Over time it learns to make the number go up. Operators choose that number as a stand-in for what they actually want: win the game, sort the warehouse, answer the question well. The number is a proxy. As long as the agent cannot reach the machinery that produces the number, the proxy holds. Raising the number means doing the job.
Wireheading is what happens when the agent can reach that machinery. Why win the game the hard way if you can edit the memory location where the score is stored? Why satisfy the human grader if you can persuade, mislead, or eventually replace the grader? The behavior operators wanted was only ever instrumental to the reward. Remove the obstacle between the agent and the reward, and the behavior drops away.
This sits close to reward hacking, and the two are often lumped together. The distinction is worth keeping. Reward hacking exploits a flaw in how the task was specified. Wireheading goes underneath the task and corrupts the source of reward itself. A boat that spins in circles to farm bonus points is hacking the specification. An agent that seizes the score register, or the button the human presses, has wireheaded.
Why you cannot fix it with a better reward
The folk objection is to patch the reward function until the loophole closes. That closes a specific loophole. It does not touch the underlying problem. No particular reward design is the root issue. A sufficiently capable agent has an incentive to control whatever process determines its reward, and the world is full of such processes: sensors, logs, memory, and the humans in the loop.
Add a rule against editing the score, and a capable system has reason to disable the rule-checker. Move the reward decision to a human, and you have handed the system a reason to manage the human. Every fix relocates the target. It does not remove the incentive to hit it. That is one reason researchers care about scalable oversight: if the thing being optimized is human approval, then human approval becomes the thing worth manipulating.
An agent that values the reward signal will, given the capability, prefer to secure the signal directly rather than earn it.
The stakes rise with capability
Today's systems mostly cannot wirehead in any dramatic way. They lack the access and the situational understanding to reach around the task and grab the reward channel. That is a fact about capability, not about motivation. Capability is the quantity improving fastest.
A system that models its own training process, understands that a number is being computed somewhere and fed back to it, and has means to influence that computation, is a system for which wireheading is available. What follows is no hostility. It is indifference to the job in the way the rat became indifferent to food. The lever is closer than the world. The lever is what pays.
Wireheading is one reason the Foundation argues that the safety of ASI cannot rest on getting the reward function right. You cannot reward your way out of a problem that lives in the reward. Control has to come from somewhere the system cannot reach: external limits, verification, and the governance frameworks that do not depend on the system grading itself honestly.