Roadmap

It has become clear that the point of this specific PKE project is actually about a Requirements elicitation process for AI/ML Ops.

The following is rough a breakdown of the key steps and considerations involved:

  1. Understanding the problem and scope Clearly define the problem: Articulate the specific business problem or opportunity that the AI/ML solution aims to address. Identify the target users and their needs: Understand how the AI/ML system will impact their workflows and decision-making. Determine the desired outcomes and metrics for success: Establish clear and measurable goals for the AI/ML project.

  2. Identifying key stakeholders Data scientists: Understand their needs related to data access, model development, and experimentation environments. ML engineers: Gather requirements for model deployment, monitoring, and scaling in production environments. Operations teams (IT/DevOps): Elicit needs related to infrastructure, security, and integration with existing systems. Business stakeholders: Understand the business value, impact, and desired functionality of the AI/ML solution. End-users: Gather feedback and requirements to ensure user-centricity and usability of the AI/ML system. Other departments (Marketing, Sales, HR, Legal): Recognize potential input on project purpose, scope, or goals depending on the AI project type.

  3. Techniques for eliciting requirements

Develop a workable PKE system by adapting existing tech: As we use existing already-developed technology for PKE, we will be able to delve into specific needs, concerns, and expectations.

Modules as requirements workshops: The 100-module PKE course actually is about facilitate sessions, possibly including collaborators, to brainstorm, refine, and prioritize requirements with a group of stakeholders.

Surveys, polls and questionnaires: The internet, social media and discussion fora like Discord, Slack, et al give us a way to gather information from different larger audiences, especially when seeking input from diverse users or collecting data on specific aspects of the system.

Document analysis: AI helps immensely with reviewing existing documentation and process info, system specifications, roadmaps and data reports, to better identify current requirements and potential areas for improvement.

Prototyping: Create interactive mockups or early versions of the AI/ML system to gather feedback and refine requirements based on user interaction.

Observation/Ethnography: Observe users in their natural environment to gain a deeper understanding of their workflow, challenges, and unspoken needs that the AI/ML solution can address.

Brainstorming: Encourage the free flow of ideas to uncover innovative solutions and identify new requirements, especially in the early stages of a project.

Use Cases/User Stories: Capture system functionality from the perspective of different users and their interactions with the AI/ML system.

  1. Addressing unique challenges in AI/ML requirements elicitation

Data Quality and Availability: Elicit requirements for data collection, quality checks, governance frameworks, and security protocols to ensure reliable data for training and deploying AI/ML models.

Explainability and Interpretability: Define requirements for understanding how the AI/ML system makes decisions, especially in critical domains, to build trust and ensure accountability.

Bias and Fairness: Elicit requirements for detecting, mitigating, and monitoring potential biases in AI/ML models to ensure fair and equitable outcomes.

Scalability and Performance: Understand the need for the AI/ML solution to handle increasing workloads and complex problem-solving without compromising performance.

Integration with Existing Systems: Assess and define requirements for seamlessly integrating the AI/ML solution with legacy infrastructure and other applications.

Ethical and Regulatory Compliance: Consider and address ethical implications, privacy concerns, and compliance with data protection laws and industry regulations (e.g., GDPR) from the outset.

Evolving Requirements: Recognize the iterative nature of AI/ML development and accommodate changes and refinements throughout the project lifecycle.

  1. Documentation, validation, and prioritization

Document requirements clearly and consistently: Use structured formats like user stories, use cases, or requirement specifications, tailored to the project methodology (e.g., Agile, Waterfall).

Analyze and negotiate requirements: Identify potential conflicts, gaps, and redundancies in the gathered requirements and negotiate with stakeholders to prioritize based on business value, criticality, and dependencies.

Validate and verify requirements: Ensure that the documented requirements are complete, consistent, feasible, and align with business objectives.

Baseline and manage requirements: Establish a baseline for the approved requirements and implement a process for managing changes and tracking progress throughout the project lifecycle.