Deaf Led AI Goverance April 21, 2026

NCG Position | SLxAI, “Avatars,” and the Governance Gap in AI Representation

info@novaracg.com Novara Consulting Group
Promotional graphic for Novara Consulting Group titled “Deaf-Led AI Governance.” The scene shows a laptop on a rock in a forested mountain landscape at sunrise, with a holographic interface displaying a person using sign language at the center. Surrounding icons reference data, leadership, systems, and accessibility. Text emphasizes Deaf leadership in AI decision-making, governance that reduces risk, and systems that are fair, accurate, accountable, and culturally aligned. Additional phrases include “Accessibility by design, not an afterthought” and “Control creates change.”

The term “avatar” is currently being applied inconsistently across AI systems, and that lack of precision is beginning to introduce material risk into the field.

At SLxAI, multiple demonstrations fell under a single label despite representing fundamentally different technical approaches. From a governance standpoint, this is not a semantic issue. It is a classification failure.

There is a clear distinction between:

  • Systems that generate sign language from learned data
  • Systems that transform existing human video using overlays or style transfer

These are not interchangeable. They differ in:

  • Underlying model architecture
  • Data provenance
  • Authenticity of output
  • Ethical and representational implications

Conflating them under the term “avatar” creates three immediate risks.

First, representation risk. When video-to-video systems apply overlays, including skin tone modification or stylistic transformation, without disclosure, the output may misrepresent the identity and authorship of the signer. This raises concerns related to transparency, consent, and cultural integrity.

Second, trust and procurement risk. Buyers, institutions, and stakeholders cannot accurately evaluate solutions if categories are blurred. A system that transforms existing footage is not equivalent to one that generates language independently. Without clear labeling, decision-making becomes distorted.

Third, field-level credibility risk. Conferences like SLxAI function as signaling mechanisms for the direction of the industry. When distinctions are not enforced, it sets a precedent where marketing language overrides technical accuracy.

From an NCG perspective, this is a governance failure, not a technology failure.

The issue is not whether these tools exist. It is whether they are:

  • Properly classified
  • Transparently presented
  • Ethically deployed

A minimal standard moving forward should include:

  • Explicit labeling of system type (generative vs. video-to-video vs. 3D model)
  • Disclosure of transformation methods applied to human subjects
  • Clear differentiation between synthetic generation and modified recordings

Without these controls, the industry risks normalizing ambiguity at the exact moment it needs precision.

The question is not “what is an avatar.”
The question is: what standards are we willing to enforce when defining one.