AI-assisted development introduces questions about privacy, fairness, security and accountability. Teams need clear review responsibilities and appropriate safeguards throughout design, implementation and operation.
A practical example
An assistant drafting internal notes creates different risks from a system influencing an important customer decision. Identify who is affected by errors and what review is required. Document which information may be processed and who handles problems. Evaluate the system as it is actually used, including its integrations and the decisions made from its outputs.
Responsible AI review questions
| Focus | Practical approach | Check before proceeding |
|---|---|---|
| Data | Identify sensitive information and permitted uses. | Apply appropriate access and retention controls. |
| Impact | Assess who could be harmed by an incorrect output. | Increase human review as consequences increase. |
| Accountability | Assign a person to review and resolve problems. | An automated answer does not remove organisational responsibility. |
Can a disclaimer replace human oversight?
No. A warning does not correct an unsafe workflow or an unreliable output. Review and escalation should be designed around the potential consequences.
Useful next steps
For a related perspective, read AI and Machine Learning: Start With a Measurable Product Problem, Using ChatGPT at Work: Drafting, Summarising and Planning, Choosing AI Tools: Inputs, Outputs and Human Review.
If you need help applying this to your business, explore our AI development and automation or discuss your project.
Sources and further reading
By Tech Me Today. Originally published November 7, 2023. Expanded on .
Explore all articles →
