putilov_denis - stock.adobe.com

Tip

Assessing different types of generative AI applications

Learn how industries use generative AI models, which function on their own to create new content and alongside discriminative models to identify, for example, 'real' vs. 'fake.'

AI encompasses many techniques for developing software models that can accomplish meaningful work, including neural networks, genetic algorithms and reinforcement learning. Previously, only humans could perform this work. Now, these techniques can build different kinds of AI models.

Generative AI models are one of the most important kinds of AI models. A generative model creates things. Any tool that uses AI to generate a new output -- a new picture, a new paragraph or a new machine part design -- incorporates a generative model.

The various applications for generative models

Generative AI functions across a broad spectrum of applications, including the following:

  • Natural language interfaces. In performing both speech and text synthesis, these AI systems power digital assistants such as Amazon's Alexa, Apple's Siri and Google Assistant, as well as tools that auto-summarize text or autogenerate press releases from a set of key facts.
  • Image synthesis. These AI systems create images based on instructions or directions. They will, if told to, create an image of a kiwi bird eating a kiwi fruit while sitting on a big padlock key. They can be used to create ads, fashion designs or movie production storyboards. DALL-E, Midjourney and Wombo Dream are examples of AI image generators.
  • Space synthesis. AI can also create three-dimensional spaces and objects, both real and digital. It can design buildings, rooms and even whole city plans, as well as virtual spaces for gameplay or metaverse-style collaboration. Spacemaker is a real-world architectural program, while Meta's BuilderBot (in development) will focus on virtual spaces.
  • Product design and object synthesis. Now that the public is more aware of 3D printing, it's worth noting that generative AI can design and even create physical objects like machine parts and household goods. AutoCAD and SOL75 are tools using AI to perform or assist in physical object design.

Many tools harness both generative and discriminative AI models. Discriminative models, adversely, identify things. Any tool that uses AI to identify, categorize, tag or assess the authenticity of an artifact (physical or digital) incorporates a discriminative model. A discriminative model typically doesn't say categorically what something is, but rather what it most likely is based on what it sees.

Diagram of GAN training method
GAN training method

How generative and discriminative models function together

A generative adversarial network (GAN) uses a generative model to create outputs and an adversarial discriminative model to evaluate them, with feedback loops between the two. For example, a GAN might be tasked with writing fake restaurant reviews. The generative model would attempt to create seemingly real reviews, then pass them, along with real reviews, through the discriminative model. The discriminator acts as an adversary to the generative model, trying to identify the fakes.

The feedback loops ensure that the exercise trains both models to perform better. The discriminator, which is then told which inputs were real and which were fake after evaluating them, adjusts itself to get better at identifying fakes and not flagging real reviews as fake. The generator gets better at generating undetectable fakes as it learns which fakes the discriminator successfully identified and which authentic reviews it incorrectly tagged.

This phenomenon is applied in the following industries:

  • Finance. AI systems watch transaction streams in real time and analyze them in the context of a person's history to judge whether a transaction is authentic or fraudulent. All major banks and credit card companies use such software now; some develop their own and others use commercially available solutions.
  • Manufacturing. Factory AI systems can watch streams of inputs and outputs using cameras, x-rays, etc. They can flag or deflect parts and products likely to be defective. Kyocera Communications and Foxconn both use AI for visual inspection in their facilities.
  • Film and media. Just as generative tools can create fake images (e.g., a kiwi bird eating kiwi on a key), discriminative AI can identify faked images or audio files. Google's Jigsaw division focuses in part on developing technology to make deepfake detection more reliable and easier.
  • Social media and tech industry. AI systems can look at postings and patterns in postings to help spot fake accounts by disinformation bots or other bad actors. Meta has used AI for years to help find fake accounts and to flag or block COVID misinformation related to the pandemic.

Generative AI may well become a widely known tech buzzword, like automation, and its myriad applications prove that this nascent branch of AI is here to stay. To meet modern challenges facing the tech industry, it only makes sense that this technology will expand and become deeply embedded in more and more enterprises.

Dig Deeper on AI business strategies

SearchBusinessAnalytics
SearchCIO
SearchDataManagement
SearchERP
Close