Discriminative AI vs. Generative AIHuman

Discriminative AI vs. Generative AI

Human beings can discriminate between win/loss, faulty/non-faulty, herb/tree, living/non-living, etc. They have natural intelligence (Discriminative Intelligence) to do so.  

Inspired by the human behavior, many artificial intelligence algorithms have been developed that mimic this human intelligence. 

AI systems also perform tasks like classification, clustering, and regression analysis.  This capability is termed as Discriminative AI. 

 

 

Figure 2: Discriminative AI

 

In addition to Discriminative Intelligence, humans also possess Generative Intelligence. They exhibit it while creating a work of art, creating music, creating new original designs, writing a story or poem and more. 

Today, there are state of the art AI algorithms which can also create – art, text and more! We term this as ‘Generative AI’. 

Consider data with two different classes – A and B represented by a scatter plot such as one given here. 

Figure 3: A scatter plot representing two classes

A simple way to separate one type of points from another, would be to draw a boundary separating the classes.  Once this is done, a new point such as Point P (See Figure 2), can be easily classified as Class A based on its position. 

Figure 4: Classification using a discriminative model

 

This is how the Discriminative models work.

Generative models on the other hand understand the data in such as way that they can generate new samples belonging to the same type(s).

 

Figure 5: A Generative Model 

 

New samples can be generated using the Generative Model as shown below. 

Figure 6: Generating new data using a Generative Model

 

OpenAI provides two of the most popular generative models – GPT-3 for text and code tasks and DALL-E2 for image generation tasks.