# Computer Vision Computer vision (CV) is the subfield of [[Artificial Intelligence]] that gives machines the ability to interpret visual information — recognizing objects, faces, scenes, and actions in images and video. It is how a self-driving car sees pedestrians, how a phone unlocks at your face, how a medical AI reads a scan, and how a surveillance camera identifies a person in a crowd. The field was revolutionized by [[Deep Learning]]: the Convolutional Neural Network AlexNet's 2012 win on the ImageNet dataset (assembled under [[Fei-Fei Li]]) is widely treated as the spark of the entire modern AI boom. ## Why it matters here Computer vision is the AI subfield most directly fused to the **surveillance** concerns at the center of this vault — the place where a technical capability becomes an instrument of state and corporate power. **Facial recognition and the surveillance state.** CV is the engine of [[Facial Recognition]], which lets governments and companies identify and track individuals at scale from ordinary camera footage. This powers the [[Surveillance State]] — China's pervasive monitoring is the maximal case, but US law enforcement use of tools like Clearview AI (which scraped billions of social-media photos) raises the same [[Privacy]] and due-process concerns. The technology makes anonymity in public effectively obsolete. **Bias with high stakes.** Facial-recognition systems have documented higher error rates on women and darker-skinned faces, producing wrongful arrests — a concrete case of [[Algorithmic Bias]] with liberty at stake. **Autonomy and weapons.** CV is essential to self-driving vehicles and to [[Autonomous Weapons]] — drones and systems that can identify and engage targets, raising the "killer robot" and accountability questions the vault tracks in the defense-tech cluster. The dual-use character is stark: the same object-recognition advance that helps radiologists spot tumors also enables automated population surveillance. In most AI subfields dual use is a policy caveat appended to a technical description. In computer vision it is the **defining characteristic** — the identical convolutional architecture reads a medical scan and a drone feed, and nothing in the model distinguishes the two. [[Clearview AI]] scraped public photos for law-enforcement facial recognition; [[Project Maven]] applied the same class of technique to Pentagon drone surveillance footage. See [[Dual-Use AI]]. ## The robotics side CV is not only a surveillance technology. It is the sensing layer of autonomy: - **Navigation.** SLAM — simultaneous localization and mapping — is a computer-vision problem. [[Skydio]]'s autonomous flight is fundamentally CV: the drone sees its surroundings through cameras and makes navigation decisions in real time. - **Manipulation.** Object recognition and pose estimation are what let a robot grasp a thing it has not seen before. - **Defense-tech integration.** [[Anduril Lattice]]'s threat detection processes visual feeds from [[Anduril Sentry Tower]]s and drone platforms — the same pipeline, pointed at a border. The technical scope of the field, for reference: image classification, object detection, semantic segmentation, pose estimation, and video analysis. ## See Also - Convolutional Neural Network — the architecture that cracked vision - [[Facial Recognition]] — CV's most politically charged application - [[Surveillance State]] — what large-scale CV enables - [[Autonomous Weapons]] — CV in military systems - [[Algorithmic Bias]] — the disparate-error problem - [[Dual-Use AI]] — the policy frame - [[Sensor]] — the hardware that captures visual data - [[Technology Lexicon]] — the full vocabulary (robotics section) - [[Surveillance Lexicon]] — the monitoring applications