3. A Digital Fly Evolves in Terraria — Real Injuries, Better Senses & Persistent Learning

I placed a digital fly inside Terraria and connected its body to an electronic brain based on a real fly connectome. Today begins a new overnight experiment with a cleaner body simulation, improved sensory feedback and more reliable learning measurements. The fly sees through two compound eyes, smells creatures and objects, tastes surfaces with its feet and proboscis, and senses wind, gravity, balance, fatigue and directional pain. Its brain directly controls the wings, six legs, grip and proboscis—without navigation commands or scripted goals. Since the previous stream, the experiment has been improved: every injury now has a physical cause: creatures, projectiles, spikes, thorns, lava, starvation, dehydration or toxic food; impacts produce pain according to their strength; starvation and dehydration now damage the body gradually instead of causing an instant artificial death; item gripping is more stable and no longer rapidly grabs and drops the same object; food, water and damage events are measured more accurately; death causes are recorded separately instead of being labelled simply as “injury”; the fly can smell food, creatures, hazards and different objects more clearly; it can carry weighted objects, return valuables to its hive and exchange them for nutritious hardened honey; its learned neural changes survive death and respawn; the observer now shows the latest injury source, survival trend and real-time learning activity. There are still no commands such as “find food,” “avoid enemies” or “return to the hive.” Terraria provides only light, colour, smell, taste, contact, balance, pain and internal body signals. The connectome must discover useful behaviour through experience. On screen: the fly living and dying inside Terraria; both 16×16 compound-eye images; live activity of its wings, legs, grip and proboscis; smell, taste, contact, wind and balance sensors; energy, hydration, individual limb fatigue and directional pain; immediate reward and harm signals; neural plasticity and learning activity; exact causes of death; lifespan across successive lives. Early results from today’s test are encouraging: every completed life has included eating and drinking, the best life lasted over 12 minutes, and recent lives are longer than the first ones. The sample is still small, so tonight will show whether this is genuine learning or simply luck. This is a live biological-computing experiment. Nothing guarantees survival. The fly may explore, panic, get attacked, carry useless objects, starve beside food—or gradually discover how to live in its strange new world.