A Beginner's Guide to Gen AI

We've all heard the term Generative AI. Many of us have experimented with it or used it. BUT, what exactly is Generative AI? In this brief guide, I will introduce you to what it is and why it matters.

Generative AI uses algorithms, specifically machine learning algorithms, to review large amounts of data and then receive outputs based on this large input of data. The "Generative" part of Generative AI refers to these tools' ability to create new content based on the data it has been trained on.

Before getting into use cases for Generative AI, let me distinguish Generative AI from "traditional" AI. Traditional AI uses predefined rules and code to perform specific tasks pre-defined for it. Generative AI has no predefined tasks. You can define what Generative AI does based on the data it has been trained on.

There are different types of Generative AI models. I want to briefly touch upon each type:

1) Generative Adversarial Networks: Known as GANs for short, these are used to create realistic images and videos.

To illustrate, consider this. GANs behave like two people competing against one another. One looks at something and decides if it is real or fake. The other provides feedback. The game continues until the evaluator is able to more accurate distinguish between the two.

2) Transformers: I'm not talking about the movies or the toys, though I did love both. These are models used for creating predictive text and recognizing speech.

For example, you are writing a text and your phone starts to finish your message. This is an example of a transformer.

3) Variational Autoencoders: All you need to know is that these models are used to create artificial images and text.

For example, you give your friend a picture. The friend takes the picture and makes it smaller to fit into a certain frame. But then you decide you want the original photo back. Your friend...who has been watching Star Wars and somehow understands The Force, is able to turn the photo back into its original size. This is what this type of model does.

4) Flow-Based Models: These are also used to create text and images, but in a specific way.

For example, consider you have a certain number of legos. No more and no less. You can create whatever you want with that number of legos, but cannot add more or use less. This is how a flow-based model works to create things.

5) Recurrent Neural Networks: These are used for speech recognition and natural language processing.

Say two friends are playing catch. One of your friends is an excellent athlete and is able to always catch the ball seemingly without moving an inch no matter how it is thrown. This type of model behaves like this friend. It can do something consistently based on all previous activities of that type.

Okay, so now that you know the types of models thet exist, it's time to touch upon the concerns and limitations of these models:

• Biased, Outdated, or Unreliable Information: Since generative AI systems are trained on existing data, they may replicate biases or use outdated or unreliable information. It's crucial to vet and validate data sources, create processes to remove biased data, and regularly monitor and review content to ensure it's factual and unbiased.
• Generative AI Hallucinations: Generative AI may produce incorrect or irrelevant information, known as "hallucinations." These occur when the AI creates new content based on the facts it has learned but adds its own interpretation, leading to distorted information. These instances can result in misinformation or insensitive content.
• Replacement of Humans: While the goal of generative AI is to enhance productivity and skills, it also is changing what work is done by humans and, subsequently, the types of jobs humans are needed for. This does not necessarily mean the loss of a job, but it does mean that it is incumbent upon us all to adapt to the rise of Generative AI and be prepared to augment our existing capabilities with its own.

While AI itself is not a new concept or technology, Generative AI is. The growing impact of Generative AI cannot be overstated or underestimated. How the world works is changing. The way forward requires three things from each of us - a) A willingness to learn and to adapt, b) Being comfortable being uncomfortable at times as Generative AI evolves, and c) Recognizing that technology is here to augment our own abilities, but to do so effectively requires experimentation and iterating.

This article is just a high-level and brief introduction to this exciting space. To learn more, I strongly encourage you to check out these resources:

1) Josh Kubicki's excellent newsletter, The Brainyacts.

2) This article from ZenDesk (be mindful of its customer-service focused bent due to Zendesk being in that space).

3) This excellent resource from the Michigan Institute for Data Science at the University of Michigan.

4) This interview with the Dean of Suffolk University Law School, known for its legal tech prowess.