Designing AI for Real Users -- Accessibility Gaps in Retail AI Front-End

Dimensions overall score 3.0

• Ethical scrutiny of AI systems has largely focused on models, data, and governance, with far less attention paid to how AI is experienced through user-facing design.limitation 95%
  Evidencepartial

  As AI becomes embedded in customer-facing systems, ethical scrutiny has largely focused on models, data, and governance. Far less attention has been paid to how AI is experienced through user-facing design.

  Implicationpartial

  Explicitly and repeatedly stated in both the abstract and introduction.

  Verificationpartial

  partial

• Many AI front-ends implicitly assume an 'ideal user body and mind', which marginalizes users with vision, hearing, motor, cognitive, speech, and sensory differences, as well as age-related variations.limitation 90%
  Evidencepartial

  This commentary argues that many AI front-ends implicitly assume an 'ideal user body and mind', and that this becomes visible—and ethically consequential—when examined through the experiences of differently abled users.

  Implicationpartial

  Central argument of the paper, directly stated in the abstract and introduction with specific user groups listed.

  Verificationpartial

  partial

• Common accessibility barriers in retail AI chatbots and voice assistants include unlabeled controls, keyboard traps, focus jumping, and screen readers failing to announce new messages.technical 85%
  Evidencepartial

  Common accessibility barriers include unlabeled controls, keyboard traps, focus jumping, and screen readers failing to announce new messages.

  Implicationpartial

  Specific technical barriers are listed, indicating a detailed analysis, though the evidence is presented as a general statement rather than with specific data.

  Verificationpartial

  partial

• Conversational interfaces, often positioned as more accessible, can assume age-specific norms around phrasing, pacing, and turn-taking that are not equally intuitive across generations.limitation 80%
  Evidencepartial

  While conversational interfaces are often positioned as more accessible, they can still assume age-specific norms around phrasing, pacing, and turn-taking that are not equally intuitive across generations.

  Implicationpartial

  Claim is directly stated and provides specific examples (phrasing, pacing, turn-taking), but is presented as a general observation rather than a quantified result.

  Verificationpartial

  partial

• The failures of AI front-ends to be accessible persist not primarily due to technical limits, but due to commercial, organisational, and procurement contexts where accessibility is rarely contractual.limitation 75%
  Evidencepartial

  Drawing on practice-led insights, we argue that these failures persist not primarily due to technical limits, but due to the commercial, organisational, and procurement contexts in which AI front-ends are designed and deployed, where accessibility is rarely contractual.

  Implicationpartial

  Strongly implied in the abstract's conclusion ('due to the commercial, organisational, and procurement contexts...') and aligns with the paper's overall argument, though the full supporting analysis is not fully detailed in the provided excerpts.

  Verificationpartial

  partial

• Front-end assurance should be proposed as a practical complement to AI governance to align claims of intelligence and multimodality with user diversity.method 85%
  Evidencepartial

  We propose front-end assurance as a practical complement to AI governance, aligning claims of intelligence and multimodality with the diversity of real users.

  Implicationpartial

  Directly stated as the proposed solution in the abstract and introduction.

  Verificationpartial

  partial

• Adapting AI fairness metrics to interaction outcomes (e.g., task completion parity) for accessibility requires careful attention to the fluid, intersectional nature of disability categories and representative user testing.method 80%
  Evidencepartial

  However, this adaptation requires careful attention to group taxonomy—disability categories are more fluid and intersectional than typical protected attributes such as gender or origin—and to data collection methodology, as interaction-level metrics require representative user testing rather than computational evaluation on held-out datasets.

  Implicationpartial

  Specific methodological challenges are outlined, showing a considered approach, though the claim is presented as a discussion point rather than a finding from executed research.

  Verificationpartial

  partial

Competitive landscape

This paper critiques the accessibility gaps in retail AI front-ends, arguing for front-end assurance to align AI capabilities with diverse user needs.