Position: AI Should Sense Better, Not Just Scale Bigger: Adaptive Sensing as a Paradigm Shift

Eunsu Baek, Keondo Park, Jeonggil Ko, Min-hwan Oh, Taesik Gong, Hyung-Sin Kim

Keywords: adaptive sensing, test-time input adaptation, efficient AI, physical AI

Abstract: Current AI advances largely rely on scaling neural models and expanding training datasets to achieve generalization and robustness. Despite notable successes, this paradigm incurs significant environmental, economic, and ethical costs, limiting sustainability and equitable access. Inspired by biological sensory systems, where adaptation occurs dynamically at the input (e.g., adjusting pupil size, refocusing vision)—we advocate for adaptive sensing as a necessary and foundational shift. Adaptive sensing proactively modulates sensor parameters (e.g., exposure, sensitivity, multimodal configurations) at the input level, significantly mitigating covariate shifts and improving efficiency. Empirical evidence from recent studies demonstrates that adaptive sensing enables small models (e.g., EfficientNet-B0) to surpass substantially larger models (e.g., OpenCLIP-H) trained with significantly more data and compute. We (i) outline a roadmap for broadly integrating adaptive sensing into real-world applications spanning humanoid, healthcare, autonomous systems, agriculture, and environmental monitoring, (ii) critically assess technical and ethical integration challenges, and (iii) propose targeted research directions, such as standardized benchmarks, real-time adaptive algorithms, multimodal integration, and privacy-preserving methods. Collectively, these efforts aim to transition the AI community toward sustainable, robust, and equitable artificial intelligence systems.

Lay Summary:

This paper argues for adaptive sensing for deep neural networks (DNNs) as a first-class design principle, and we advocate it as a necessary, foundational shift. Today’s AI progress largely comes from ever-larger models trained on massive datasets. That path brings strong capabilities but also high energy use, costly hardware, and reduced access—and it still falters when real-world conditions shift.

Inspired by biology—our pupils shrink and eyes refocus—adaptive sensing lets systems adjust how they capture data during recording. Instead of fixed, human-oriented settings, sensors tune parameters in real time to collect inputs the model can interpret most reliably. In practice, systems can adjust measurement settings (e.g., exposure, gain, multimodal configurations) to match the situation. These adjustments capture clearer, more informative inputs, making the AI’s job easier—like cleaning your glasses before reading.

By improving input at its source, adaptive sensing boosts efficiency and robustness without always needing bigger models or more data. Our studies show adaptive sensing can enable small models to match or surpass much larger models that use conventional sensing settings. We outline a roadmap for deployment across healthcare, robotics, autonomous vehicles, agriculture, and environmental monitoring—toward AI that is more sustainable, reliable, and fair.

Submission Number: 235