Table 1.

The role-based EthAI-SB framework for use in small business

CharacteristicsAI as an internGenAI as an assistantGenAI as a manager
Level of technologyPassiveInteractiveProactive
Level of control: belief in the ability to influence and have agency (Letheren et al., 2019)NoneLimitedHigh autonomy but business is in control. They set the rules of engagement
Engagement preference: cognitive, affective and behavioural interactions (Letheren et al., 2019)Cognitive engagement, controlled (task oriented, informational)Behavioural engagement, delegated, (suggestions)Emotional engagement, outsourced
Trust and perceived risk: the belief that an entity will act in a predictable and beneficial way, possibility for loss or harm (Letheren et al., 2019)Low trust and high perceived riskModerate trust and riskHigh trust and low perceived risk
Fairness: avoiding discriminatory or unjust outcomes. (Shin and Park, 2019)Ensure a lack of bias in audits of AI toolsReview AI-assisted outputs for potential biases before finalisation. Ensure training data for any custom AI assistant tools is diverse and representative; provide clear instructions to the AI to avoid biased language or outputsConduct regular, independent audits of the AI manager’s decision-making algorithms and data inputs for bias; establish appeal mechanisms for decisions made by the AI manager; ensure diverse human oversight of the AI manager’s functions
Accountability: being answerable for decisions and for addressing risks (Shin and Park, 2019)Define roles for AI oversight; document AI decision-making processes; establish protocols for error reporting and remediation; ensure human supervisionClearly define the scope of tasks delegated to the AI assistant; maintain logs of AI usage and outputs; ensure the supervising professional reviews and takes responsibility for all work produced with AI assistanceClearly define the AI manager’s scope of authority and decision-making limits; assign ultimate accountability to senior human leadership; implement robust logging and audit trails for all AI managerial actions and decisions
Transparency: Reasoning and data management is understandable to users (Shin and Park, 2019)Train staff to interpret AI explanations; develop clear client communication protocols regarding AI useUnderstand the AI assistant’s capabilities and how it generates its contributions; document the extent of AI assistance in work papers; communicate the use of AI to clients if its role in service delivery is significantUse well-developed and proven AI systems with strong AI capabilities; ensure clear documentation and communication of the AI manager’s functions and logic; disclose the use of an AI manager to all relevant stakeholders
Accuracy: correctness and reliability of information or outcomes (Rana et al., 2024)Implement data validation for inputs; conduct rigorous human review of outputs; cross-reference AI outputs with other sources; monitor AI performanceImplement rigorous human review and validation of all AI-assisted outputs; cross-reference AI-generated information with reliable sources; do not solely rely on the AI assistant for critical informationImplement stringent data validation processes for all data informing the AI manager; regularly verify the accuracy of the AI manager’s analyses and outputs; ensure human review of critical managerial decisions based on AI outputs
Autonomy: maintaining decision-making control, rather than being overruled by automated systems. (Rana et al., 2024)Define tasks suitable for AI autonomy; establish human intervention points and override capabilities; regularly review the appropriateness of AI autonomyDefine clear boundaries for the AI assistant’s tasks; ensure human professionals actively guide and oversee the AI assistant’s work; maintain the ability to intervene and correct AI-assisted outputs at all stagesStrictly define and limit the AI manager’s autonomous decision-making authority; require human approval for significant managerial decisions; ensure robust human oversight and the capability to intervene or override the AI manager at any point
Source(s): Adapted from the ten-step process proposed by Caldwell and Peters, 2017 

or Create an Account

Close Modal
Close Modal