Cognitive steering improves communication accuracy using multiple AI controls
Multi-System Cognitive Steering for Generative Communication
Artificial Intelligence
Summary
Talking or typing takes effort, so people often use short abbreviations to save time. But computers can misunderstand these abbreviations without enough clues, leading to wrong messages. The authors created a method called Cognitive Steering that helps AI understand what someone really means by using extra information, like objects around them or their feelings. By combining different ways to guide the AI, their method makes it much better at figuring out the intended meaning behind brief inputs.
What this means in practice
- •For assistive technology developers: Design communication aids that decode abbreviated user inputs accurately by integrating contextual cues and AI control techniques.$Commercial implications: Enables new assistive devices for faster, more accurate speech generation benefiting users with limited motor abilities.
- •For virtual assistant engineers: Improve virtual assistants’ understanding of brief or unclear commands by incorporating multimodal context and adaptive AI controls.
Authors
Eric Lacosse, Pablo Tano, Daniel C. McNamee
Abstract
A fundamental bottleneck for translating thoughts to speech is the high physical cost of motor input forcing a strict tradeoff between communication speed and expressivity. Generative AI offers a promising solution by expanding highly abbreviated text input into fluent sentences, significantly reducing the motor burden on the user. However, sparse inputs are inherently ambiguous; without sufficient context, AI models frequently misinterpret these abbreviations and generate outputs that fail to align with the user's true cognitive intent. We propose \textit{Cognitive Steering}, a framework that can be leveraged to reconstruct user intent by grounding Large Language Models (LLMs) in multimodal context and applying AI control methods. We demonstrate that aligning frontier models with selected environmental referents, emotional states, and communication motives on simulated abbreviation benchmarks achieves up to $\sim$88\% top-5 expansion accuracy, doubling no-context baselines. We also reveal that different types of context require distinct AI control techniques: while physical objects in a scene are best provided through explicit text prompts, abstract factors like emotion may be more effectively managed by directly altering the AI's internal activations. These findings suggest a hybrid architecture for future interfaces that leverage Cognitive Steering to help users communicate both quickly and accurately, recruiting multiple AI control methods simultaneously to perfectly capture their intended communication goals.