Responsible AI · RUDI’s RESPECT framework
What is human-centered AI?
AI should help people do useful work, with someone responsible for how it is used and what it does.
Start with the people doing and receiving the work.
Human-centered AI means designing and managing AI around people’s needs, judgment, and ability to act. That includes the employees using a tool, the people whose information it handles, and anyone affected by its output.
Consider a digital worker that prepares a weekly operations report. It may gather updates, compare figures, and write a draft. People still need to decide which sources it may use, what a good report looks like, who checks it, and how to correct a mistake. The team also needs time and support to learn the new process.
What we mean by responsible AI.
At RUDI, responsible AI means a person is accountable for the AI-assisted work and has the authority, information, and support to manage it. We expect every digital worker to have a human manager, including when routine tasks run automatically. The manager defines its boundaries, monitors results, and handles exceptions and concerns.
RUDI stands for Responsible Use of Digital Intelligence. RESPECT turns that commitment into seven questions people can use when choosing a tool, learning a workflow, or building an agent.
The RESPECT framework
Seven principles for responsible AI at work.
Use these together. A useful system also needs appropriate access, oversight, and a way for people to raise concerns.
Responsible
A person remains accountable for the work.
Every AI workflow needs a named human owner. For a digital worker, that person is its manager: they define the job, approve its access, check the quality of its work, and respond when something goes wrong. An organization remains responsible for the systems it chooses to use. “The AI did it” does not give the affected person a way to resolve a mistake.
Ask your team: Who owns this workflow, and how can someone reach them?
Equitable access
Give people the support they need to participate.
Employees start with different experience, confidence, responsibilities, and access to technology. Some need foundational practice; others need accessible tools, time to learn, or examples that fit their role. Involve the people whose work will change and adapt the support to their needs. Equitable access includes accessibility and a fair opportunity to benefit from the change.
Ask your team: Who might be left out, and what would help them take part?
Safety
Check what could go wrong before expanding use.
Identify foreseeable harm to people and the organization. Test representative cases, difficult exceptions, and unreliable outputs. Define quality checks, limits on actions, and conditions that require a person to intervene. Review what happens during use, too. A successful demonstration cannot establish that a workflow will be safe in every situation.
Ask your team: What could cause harm, and when must this system stop or ask for help?
Privacy
Limit data access to what the work requires.
Decide what information an AI system may use, where that information can go, and who has permission to see the result. A connector should receive only the access needed for its task. Check retention, sharing, and the handling of personal or confidential information before using real workplace data. Permission to read a record does not automatically include permission to change or share it.
Ask your team: What data can this system access, and who approved that access?
Efficient and effective
Improve the whole task, including the work of checking it.
Compare the result with the way the work is done today. Include preparation, review, corrections, and ongoing maintenance when measuring time saved. Check whether quality improves and whether the tool is easy enough for people to use. If an AI-generated draft creates three times as much checking and repair, the workflow needs to change.
Ask your team: Does this produce better work with a reasonable total effort?
Control
Give people authority to question and intervene.
People need usable ways to review, correct, override, or pause AI-assisted work. Define which actions require approval and which can run within agreed limits. Make the route to a human clear, along with any available alternative to using the AI system. Someone affected by an error should be able to challenge it and get a response.
Ask your team: Can the right person stop the work, correct it, and resolve a problem?
Transparency
Tell people when and how AI is involved.
Disclose when someone is interacting with an AI system and when AI is used in the process that produces a service or deliverable. Explain its role, relevant limitations, and where human help is available. A polished response or a convincing voice does not remove that responsibility. People should be able to understand what they received and how to question it.
Ask your team: Would the person receiving this know where AI was involved?
Give the human manager a workable role.
Putting a person’s name on a workflow is a starting point. That person needs time to review the work, enough knowledge to spot problems, access to the relevant records, and authority to pause or change the system. A review step only helps when the reviewer can make a meaningful decision.
The level of review should match the consequences of an error. A routine internal summary and a decision that affects someone’s pay require different checks. Define those checks before launch and revisit them when the task, data, or system changes.
Apply RESPECT to a weekly report.
For the reporting example, the team would name an owner, approve the source folders, and involve the employees who prepare and use the report. The digital worker would flag missing or conflicting figures for review. The owner could pause it, correct its access, and use the existing reporting process if needed.
The finished report would disclose the AI assistance and identify the responsible reviewer. The team would compare total preparation and review time with its baseline, while checking accuracy and whether readers can use the result. That gives the owner evidence for deciding whether to continue, change, or stop the workflow.
Use RESPECT with your team.
Choose one recurring task. Write down who it affects, answer the seven questions, and record the evidence or unresolved issue behind each answer. Assign an owner to any gap before deciding whether to test the workflow or put it into regular use.
RUDI brings this approach into team training, department workflow programs, and the agents and applications we help build.
Further reading.
RESPECT is RUDI’s practical framework. For broader guidance, NIST’s AI Risk Management Framework discusses reliability, safety, accountability, transparency, privacy, and fairness. The OECD AI Principles address human rights, transparency, safety, and accountability. These references provide additional guidance for teams developing their own policies and practices.
Work with RUDI
Bring a workflow and the people who use it.
We can help your team learn to use AI, evaluate a proposed agent, or set practical rules for your tools and data.
Discuss your team’s work