## TL;DR
AI enables space exploration beyond what humans can directly control -- from Mars rovers that drive themselves to satellites that process images in orbit to telescopes that autonomously detect interesting astronomical events. The fundamental constraint of space (communication delays, bandwidth limits, hostile environments) makes AI not just useful but essential for deep-space missions.

## Core Explanation
Space AI is driven by unique constraints: (1) Communication latency -- 4-24 minutes to Mars, hours to outer planets. Real-time teleoperation is impossible; autonomy is mandatory; (2) Limited bandwidth -- deep-space data rates are extremely constrained (Mars rovers: ~2 Mbps). Onboard AI must filter and compress data; (3) Radiation and reliability -- hardware must survive cosmic radiation; radiation-hardened processors lag consumer hardware by 5-10 years; (4) Power constraints -- solar or RTG power budgets are limited, AI inference must be efficient; (5) Unpredictable environments -- alien terrain, novel situations not seen in training data, requiring robust generalization.

## Detailed Analysis
Mars rover autonomy: Perseverance's AutoNav uses stereo cameras to create 3D terrain maps, a neural network classifies terrain types and hazards, and a path planner computes safe routes. The rover evaluates multiple paths and selects the best, executing without waiting for Earth. The AEGIS system autonomously selects interesting scientific targets (rocks with unusual composition) and aims instruments (SuperCam laser spectrometer). Orbital AI: ESA Phi-Sat missions demonstrate edge AI in space. Phi-Sat-1 (2020): Intel Movidius Myriad 2 vision processing unit runs a convolutional neural network to detect and discard cloudy images -- only cloud-free, useful images are downlinked. Phi-Sat-2 (2024): more advanced onboard AI for ship detection, wildfire monitoring, and urban change detection. NASA-IBM Prithvi: a foundation model for Earth observation, pretrained on Harmonized Landsat-Sentinel data, adaptable to multiple downstream tasks (flood mapping, crop classification, burn scar detection). Spacecraft autonomy: NASA's Deep Space 1 (1998) was the first spacecraft with AI-based autonomous navigation (AutoNav) using optical navigation -- comparing star tracker images against onboard star catalogs. Modern missions (OSIRIS-REx, DART) use AI for autonomous guidance during critical phases (asteroid sampling, kinetic impact). The Artemis program (lunar return) requires AI for autonomous lunar landing, surface navigation, and habitat management. An emerging frontier: AI for SETI and exoplanet detection -- machine learning models processing petabytes of telescope data to find signals and planets that humans might miss.