Neural Computers

2026-04-07 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Expert, quick

Summary

Researchers propose "Neural Computers" (NCs), a new computing paradigm that unifies computation, memory, and I/O into a learned runtime state, aiming for a "Completely Neural Computer" (CNC) as a general-purpose realization. As an initial step, the study investigates whether elementary NC primitives can be learned solely from collected I/O traces, without instrumented program state. NCs are instantiated as video models that generate screen frames from instructions, pixels, and user actions in both command-line interface (CLI) and graphical user interface (GUI) settings. The findings indicate that NCs can acquire basic interface primitives, specifically I/O alignment and short-horizon control, though challenges remain in routine reuse, controlled updates, and symbolic stability. This work outlines a roadmap for developing CNCs, positioning them as a distinct computing paradigm beyond current agents and conventional computers.

Key takeaway

For research scientists exploring novel computing architectures, consider the Neural Computer paradigm as a direction for future work. This approach suggests that fundamental computer operations can be learned from I/O traces, offering a path to systems that integrate computation and memory more deeply. You should investigate methods to overcome current challenges in routine reuse and symbolic stability to advance toward a Completely Neural Computer.

Key insights

Neural Computers unify computation, memory, and I/O into a learned runtime state, aiming for a new computing paradigm.

Principles

• I/O traces can teach elementary interface primitives.
• Unifying computer components into a learned state is feasible.

Method

NCs are instantiated as video models that roll out screen frames based on instructions, pixels, and user actions, learning from I/O traces in CLI and GUI environments.

In practice

• Explore learning system primitives from I/O data.
• Focus on I/O alignment for neural interface control.

Topics

• Neural Computers
• Completely Neural Computer
• Learned Runtime State
• I/O Traces
• Video Models

Best for: Research Scientist, AI Scientist

Related on AIssential

• The next paradigm: AI, Neural Networks, the Symbolic Brain — Current neural networks are fundamentally limited; a new symbolic brain paradigm is needed.
• Princeton builds bio-hybrid computer using living neurons — A 3D bio-hybrid computer integrates living neurons with electronics to create energy-efficient, pattern-recognizing neural networks.
• AI Engram: In Search of Memory Traces in Artificial Intelligence — AI Engram identifies and surgically manipulates specific memory traces in neural networks using a geometric framework.
• Separable neural architectures as a primitive for unified predictive and generative intelligence — Separable Neural Architectures unify diverse models by exploiting factorisable structures for predictive and generative intelligence.
• 😺 🎙️ Watch: Is Brain-like Computing What's Next? — Brain-inspired hardware using light and integrated memory-compute can overcome traditional AI bottlenecks.
• How AI Taught Itself to See [DINOv3] — Self-supervised learning with DINOv3 enables robust visual feature extraction without human labels, leveraging distillation and architectural innovations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.