[Paper Review] Programmatic Manipulation of Common Lisp Type Specifiers

Jim E. Newton, Didier Verna+1 more|arXiv (Cornell University)|Apr 3, 2017

Advanced Data Storage Technologies103 citations

TL;DR

This paper proposes a framework for human-AI integration through direct neural feedback loops, where AI systems learn from human motor and sensory inputs (e.g., silent speech, hand gestures) and provide real-time output to close the loop, enabling skill acquisition and cognitive enhancement. The key contribution is redefining AI not as a tool but as an integrated extension of human capability, exemplified by voice generation from intraoral movements and gesture-based speech synthesis.

ABSTRACT

In this article we contrast the use of the s-expression with the BDD<br> (Binary Decision Diagram) as a data structure for programmatically<br> manipulating Common Lisp type specifiers. The s-expression is the<br> extit{de facto} standard surface syntax and also programmatic<br> representation of the type specifier, but the BDD data structure<br> offers advantages: most notably, type equivalence checks using<br> s-expressions can be computationally intensive, whereas the type<br> equivalence check using BDDs is a check for object identity. As an<br> implementation and performance experiment, we define the notion of<br> maximal disjoint type decomposition, and discuss implementations of<br> algorithms to compute it: a brute force iteration, and as a tree<br> reduction. The experimental implementations represent type<br> specifiers by both aforementioned data structures, and we compare<br> the performance observed in each approach.

Motivation & Objective

• To reframe AI not as a separate tool but as an integrated part of human cognition and motor control.
• To explore how AI can extend human sensory and motor capabilities through real-time feedback loops.
• To demonstrate that AI systems can become part of a closed-loop learning process akin to natural human skill development.
• To investigate the potential of invasive brain-AI interfaces and networked human-AI systems for future human augmentation.
• To classify human-AI relationships into four quadrants based on visibility and agency, emphasizing 'human-AI integration' as a new paradigm.

Proposed method

• Uses deep neural networks to map intraoral ultrasound images to voice output, enabling silent speech generation.
• Employs neural networks to link hand gestures with voice synthesizer parameters for gesture-based vocalization.
• Establishes a feedback loop where users adjust motor patterns based on AI-generated audio output, mimicking natural speech learning.
• Applies principles from HCI and cognitive science to design systems that integrate organically with human neural processes.
• Classifies human-AI interaction into four quadrants based on visibility (visible/invisible) and agency (self/other), positioning 'human-AI integration' as a core future direction.
• Draws analogies to historical human augmentation (e.g., Hooke’s microscope, Engelbart’s mouse) to frame AI as the next evolution of human extension.

Experimental results

Research questions

• RQ1 How can AI be integrated into human motor and sensory systems to enable new forms of human augmentation?
• RQ2 What role do closed-loop feedback mechanisms play in enabling humans to learn and refine AI-mediated skills?
• RQ3 How does the integration of artificial neural networks with biological neural networks differ from traditional human-computer interaction?
• RQ4 What are the implications of treating AI not as a tool but as a functional extension of human cognition and perception?
• RQ5 How can future human-AI systems be designed to support self-directed, embodied learning and capability re-design?

Key findings

• AI systems that provide real-time feedback on motor-generated outputs (e.g., voice from silent speech) enable users to refine their movements and improve performance, similar to natural skill acquisition.
• Neural network-based systems can successfully translate intraoral ultrasound images into intelligible speech, demonstrating feasibility for communication in speech-impaired individuals.
• Gesture-based voice synthesis systems allow users to produce expressive vocalizations through hand movements, showing potential for non-verbal communication.
• The integration of human motor control with AI creates a feedback loop that mirrors natural human learning, suggesting that AI can become a functional extension of human neural systems.
• The framework of four quadrants (visible/invisible, self/other) provides a useful taxonomy for classifying human-AI relationships, with 'human-AI integration' emerging as a distinct and promising category.
• The paper argues that future HCI should focus on re-designing humans through AI integration, rather than merely improving tools, marking a paradigm shift from tool-based to organism-based interaction design.