Skip to main content

What is generative AI?

Generative AI is type of AI that can be used to create new text, images, video, audio, code, or synthetic data. 

What is generative AI?

Techopedia editor Margaret Rouse offers a comprehensive explanation of generative AI, describing it as “a broad label that’s used to describe any type of AI that can be used to create new text, images, video, audio, code or synthetic data.  While the term [is] often associated with ChatGPT and deep fakes, the technology was initially used to automate the repetitive processes used in digital image correction and digital audio correction”.1

Generative AI includes learning algorithms that make predictions and algorithms that can leverage prompts to autonomously compose articles and generate images. “Therefore, because machine learning and deep learning are inherently focused on generative processes, they can be considered types of generative AI, too”.1
 

How does generative AI work and what is its history?

George Lawton notes that generative AI first begins with “a prompt in the form of a text, image, video design, [some] musical notes, or any input that the AI system can process, [followed by] various AI algorithms [that] return new content [such as essays, solutions to problems, or realistic fakes created from pictures or audio of a person] in response to the prompt”.2

Rouse states that early generative AI “required submitting data via an API or an otherwise complicated process, [requiring] developers [to] familiarize themselves with special tools and write applications using programming languages such as Python”.1

Modern generative AI has a much more flexible user experience where ender users can input their requests using natural language instead of code. “Generative AI was introduced in the 1960s in chatbots. But it was not until 2014, with the introduction of GANs [that] generative AI could create convincingly authentic images, videos and audio of real people”.2 GANs and variational autoencoders (VAE) are two common generative models for image and text creation.

Random noise can be leveraged by some generative AI models as an input to generate new outputs. To do this, the generative AI model “takes a random noise vector as input, passes it through the network and generates output that is similar to the training data. The new data can then be used as additional, synthetic training data for creative applications in art, music and text generation”.1

Generative AI that is leveraged as a means of enhancing human creativity “can be categorized as a type of augmented artificial intelligence”.1
 

What are the differences between generative adversarial networks and variational autoencoders?

GANs are made up of two ML models being trained simultaneously: one is a generator and the other is the discriminator. The generator creates new outputs resembling training data and the discriminator evaluates the generated data and provides the generator feedback for it to leverage to improve its output.

VAEs are made up of one ML model that is “trained to encode data into a low-dimensional representation that captures the data’s important features, structure and relationships in a smaller number of dimensions [and afterwards] the model decodes the low-dimensional representation back into the original data”.1 The process allows the model to “learn a compact representation of the data distribution, which it can then use to generate new outputs”.1

What are common generative AI applications?

With the immense capabilities that generative AI offers, it’s no surprise that there’s a myriad of different applications for end users looking to create text, images, videos, audio, code, and synthetic data.  Here are some examples of the most popular generative AI applications.

  • ChatGPT – Possibly the most famous or infamous example of a generative AI application. Created by Open AI in December 2022, ChatGPT is an online AI chatbot where users prompt it with questions, and it responds by generating answers to those questions.
  • The Lensa app – This application uses AI to transform your portrait-type photos into dynamic custom portraits. Created by Prisma Labs in 2018, Lensa allows users to create transform their selfies into that of a superhero, a rockstar, or a myriad of other templates.
  • DALL·E 2 – An AI system where users input descriptions using plain language and it creates realistic images and art based on those descriptions.  Created by Open AI in April 2022, DALL·E 2 uses a diffusion model that generates higher quality images than the original Dall-E’s discrete variational autoencoder (dVAE).
  • Copy.ai – An AI writing tool that leverages ML to create various types of text content. Released by Paul Yacoubian in October 2022, Copy.ai offers different tools depending on each users’ copywriting needs and can produce long-form web copy, emails, social media content, and more.
  • Midjourney – An AI-based image generator program and service. First launched on 14 March 2022, Midjourney has been leveraged to generate award winning art, artwork used in children’s books, and images of public figures that have caused a lot of controversy.
     

What is the different between generative AI and traditional AI?

Bernard Marr writes that traditional AI, (aka narrow AI or weak AI) “focuses on performing a specific task intelligently [and] refers to systems designed to respond to a particular set of inputs”.3 These traditional AI systems can process data and make learned choices or predictions from that data.  Some of these systems function similarly to something like the IBM supercomputer Deep Blue.  They’re fed a considerable amount of data, in Deep Blue’s case chess specific data, and use it to either develop a game winning strategy or to respond to an opponent’s strategy. Other traditional AI systems operate similarly to Siri or Alexa, responding to and predicting the needs of a household, while others function more like recommendation engines for Google, Netflix, or Amazon. “AIs [that] have been trained to follow specific rules, do a particular job, and do it well, but they don’t create anything new”.3

Inversely, generative AI can create new things (text, art, music, videos, and more) from the plain language prompts that it receives. “Generative AI models are trained on a set of data and learn the underlying patterns to generate new data that mirrors the training set”.3

 

Resources

  1. Generative AI, Margaret Rouse, Techopedia, 27 June 2023.
  2. What is generative AI? Everything you need to know, George Lawton, TechTarget, 2023.
  3. The Difference Between Generative AI And Traditional AI: An Easy Explanation For Anyone, Bernard Marr, Forbes, 24 July 2023.