Choosing the Right AI Approach

Choosing an AI approach begins with understanding the problem, not selecting the most advanced technology.

A simple rule-based solution may be more appropriate than machine learning when the requirements are clear and predictable. A machine learning model may be more suitable when the problem depends on complex patterns in data.

Factors to Consider

1. Problem type

Determine whether the application needs to classify information, make predictions, generate content, follow rules, or plan actions.

2. Data availability

Machine learning usually requires suitable examples or other relevant data. If reliable data is unavailable, a rule-based approach may be a practical starting point.

3. Explainability

In applications involving finance, healthcare, compliance, or other high-impact decisions, it may be important to explain how decisions are made.

4. Cost and performance

Consider development effort, computing requirements, response time, maintenance, and infrastructure costs.

5. Reliability and risk

Evaluate what happens when the system makes an incorrect decision. Important decisions may require human review, additional validation, or a fallback process.

6. Maintenance

Consider how frequently the system's rules, data, or models will need to change.

Approach Selection Guide

These are starting points rather than universal rules. Real applications may combine several approaches.

Practical Activity: Design an AI Solution

Choose one of the following scenarios:

• A customer-support chatbot.
• A product recommendation system.
• A traffic prediction application.
• A document-processing tool.
• A warehouse navigation robot.

Write a short design proposal covering:

1. Problem: What issue should the system solve?
2. Input: What information will it receive?
3. Output: What result should it produce?
4. AI approach: Which approach would you select, and why?
5. Data and rules: What data or predefined rules would be needed?
6. Evaluation: How would you determine whether the system works well?
7. Risk: What could go wrong, and how would you reduce that risk?

For example, a customer-support chatbot could combine predefined rules for account verification with a language model for understanding questions and generating responses. Sensitive actions would require appropriate authorization and validation.