When planning to build a new AI system, it is essential to gather a wide range of information to ensure the system’s effectiveness, accuracy, and relevance. Here are the key types of information we should collect initially.
Problem Definition
To define the problem, we need to ask following questions. What is the primary purpose of the AI system? What problem is the AI system intended to solve? What are the boundaries of the problem? What will the AI system focus on, and what will it exclude? How will we measure the success of the AI system?
Stakeholder Requirements
To collect stakeholders’ requirements, we should understand user needs, business requirements, and regulatory and compliance requirements. We then need to find answers to following questions. Who are the end-users, and what do they expect from the AI system? What are the business objectives, and how should the AI system align with the business objectives? Are there any legal or regulatory constraints that the system must adhere to?
Data Collection
Data collection includes data sources, data quality, data volumeand variety, and data privacy. Here we have these questions need to be answered. What data sources will the AI system use? Are those data sources internal, external, structured, or unstructured? What is the quality of the data? Is the data clean, accurate, and up-to-date? How much data is available, and what are the different types of data(e.g., text, images, audio)? What are the data privacy concerns? How will sensitive data be handled?
Technical Requirements
We should figure out infrastructure needs, scalability, integration, and security. For infrastructure needs, what kind of hardware and software infrastructure is required? How will the system scale as the data or user base grows? How will the AI system integrate with existing systems and workflows? What are the security requirements for the system?
Algorithm and Model Selection
First of all, what is the type of AI model? Will the AI model be a supervised, unsupervised, reinforcement learning, or a combination of these? The second is the model complexity. What level of complexity is required for the models to meet the system’s objectives? The next is evaluation metrics. What metrics will be used to evaluate the performance of the AI models.
Ethical Considerations
We need to consider bias and fairness, transparency, and accountability. How will we ensure that the AI system is fair and unbiased? How transparent does the AI system need to be in its decision-making processes? Who is responsible for the decisions made by the AI system?
Project Management and Timeline
Just like all projects, we should have project management. The project management covers milestones, resources, and risks and mitigation. we should work out the key milestoners for the project. We need to make sure the availability of resources (team, budget, and tool). We should know potential risks and how to mitigate.
Feedback and Iteration
Feedback and iteration involve continuous improvement, monitoring, and maintenance. We will find out how the system to be updated and improved based on user feedback, and what processes monitor the system’s performance and maintain it over time.
Collecting and analyzing the information will help guide the development of the AI system, ensuring the AI system meets the intended objectives and functions effectively within its intended environment.