Exploring AI use cases and applications

Question 1

When are AI/ML solutions appropriate?

Answer 1

1. Coding the rules is challenging - e.g. determining if an email is spam or not.
2. Scale of the task is challenging - e.g. scanning millions of emails to determine which one is spam.

Question 2

What are the different types of ML model learning techniques?

Answer 2

1. Supervised Learning - use of labeled data
2. Unsupervised - unlabeled data, machines recognize patterns inherent in the data
3. Reinforcement - given performance score; Reinforcement learning is broadly useful when the reward of a desired outcome is known, but the path to achieving it isn’t—and that path requires a lot of trial and error to discover

Question 3

What are the two sub-categories of supervised learning?

Answer 3

1. Classification - model trained on a labeled dataset and then used to predict unforeseen data. - e.g. recognize a new car
2. Regression - supervised learning, model the relationship between dependent and independent variables and then use them to predict the dependent variable - e.g. what I did in South Carolina with the powerplant sensor readings.

Question 4

What are the two types of unsupervised learning?

Answer 4

1. Clustering - A kind of algorithm that groups data into different clusters based on similar features or distances between the data point to better understand the attributes of a specific cluster
2. Dimensionality Reduction - reduce the number of features or dimensions in a dataset while preserving the most important information or patterns

Question 5

What are the capabilities of Gen AI?

Answer 5

1. Adaptability - to various tasks and domains
2. Responsiveness - generate content in real-time; useful in chatbots
3. Simplicity - simplifies complex tasks
4. Creativity and Exploration - generates novel ideas
5. Data efficiency - train on a small amount of data and generate additional synthetic data
6. Personalization - to suit individual preferences
7. Scalability - producing content at scale

Question 6

Challenges of Gen AI

Answer 6

1. Regulatory risks - e.g. disclosure of PII; mitigate through strict control of training data
2. Social risk - unwanted content; mitigate via testing and evaluation prior to deployment
3. Data security and privacy concerns - mitigate via encryption/firewalls
4. Toxicity - inflammatory content; mitigate via guardrails.
5. Hallucinations - inaccurate responses not consistent with training data ; mitigation - users much check output and must not over-rely on the model
6. Interpretability - being able to explain how a model arrived at the decision that it did.
7. Non-determinism - model generates different outputs for the same input.

Question 7

What factors are in play when choosing a Gen AI model?

Answer 7

1. Model Types - different models have different capabilities - e.g. text generation, summarization, chat, images, etc. - which may suit different use cases.
2. Performance Requirement - accuracy, reliability
3. Constraints - computational power, data availability, onprem vs. cloud
4. Capabilities - text vs. multi-modal vs. text-to-images etc.
5. Compliance - biases, privacy issues, potential for abuse
6. Cost - larger models are precise, but cost more. Smaller models are cheaper and faster.

Question 8

What are the key business metrics for Gen AI?

Answer 8

It depends on the use case, but, in general:
* User Satisfaction - with AI generated content
* Average revenue per user - generated by use of Gen AI
* Cross-domain performance - ability to perform across domains and industries
* Conversion rate - e.g. rate of visitors to a website that turn into buyers
* Efficiency - how much gen AI improves resource utilization, computation time, scalability - e.g. on a manufacturing site.