We tackle the concept of ‘self-recognition’ in a simulated setting. We propose an experiment where two simultaneous reinforcement learning environments are controlled by two agents. Although each agent is given the control of its own environment, both agents receive the visual input of the same environment. The success threshold depends on self-recognition by definition as the agent must answer: am I seeing a mirror, or am I seeing a camera? We show that this experiment can be posed as an optimisation problem, solvable via evolutionary computation.