From content creation to fraud detection and identifying errors in lab results, practical applications of GAN AI are many. And businesses aren’t shying away from exploring the technology
In 2018, when Portrait of Edmond Belamy sold for an incredible $432,500, created by an algorithm fed with a data set of 15,000 portraits painted between the 14th century to the 20th, it signalled the arrival of GAN (generative adversarial network) on the world stage.
Since then, a whole new era for GAN AI began sparking innovations, as technologically advanced marketers (marketing has been working with AI for longer than most) started using the latest GAN architectures to generate high-resolution, realistic images for campaigns, saving time and money.
For example, Rosebud.ai can generate model faces for ad campaigns using artificial humans. Marketers can make infinite models to target their customers with various visuals. DataGrid developed AI for automatic whole-body model generation utilising GANs to synthesise images of non-existent people for ad campaigns and online clothing stores.
However, GAN AI capabilities are not limited to visual content. In 2020, the Guardian ran an op-ed titled, “A robot wrote this entire article. Are you scared yet, human?” written by GPT-3, OpenAI’s massive language model.
Now, marketers are using GAN to curate impactful marketing messages – emails, editorials, ads, and social media – that are data-driven and optimised for SEO. Each can be tailored to specific audiences – particular platforms, publications, or even individual consumers. GAN AI can constantly analyse, edit content, remove contradictory or dated information, and keep marketing messaging contemporary.
For example, Persado has an AI-driven solution that allows marketing creative to run multiple experiments by using different combinations of elements such as narrative, emotion, descriptions, and calls-to-action to develop the best-performing message.
Dell claims to get impressive results after leveraging Persado’s technology to harness the power of words for their promotional emails, Facebook ads, display banners, and even radio content and garner data-driven analytics for a customer segment.
Dell noticed a 50 per cent average increase in CTR and a 46 per cent average increase in responses from customers. It also generated a 22 per cent average increase in page visits and a 77 per cent average increase in add-to-carts.
GANs are algorithms that create synthetic datasets indistinguishable from real datasets by having two neural networks that compete against each other.
While one neural network generates the data, the other compares it to an accurate data set in iterative cycles. The degree of error in the data set is decreased.
GANs’ potential is enormous. They are a class of machine learning that creates new data based on what they have observed. They can mimic any data, including images, music, speech, prose. For example, after viewing millions of marketing images for different campaigns, a GAN could begin to create images for a specific ad campaign, using its gained experience and some provided assets.
No wonder companies, no matter the size or industry, are finding ways to bring GAN AI into their organisation.
Last year, Siemens released a generative AI-based web application dedicated to discovering better system architectures. Microsoft released its new product that uses OpenAI’s GPT-3 AI model, allowing non-coders to write code using natural language. BMW projected AI-Generated Art on its cars in one of its recent campaigns. Recently, Swedbank, one of Sweden’s largest banks, trained GANs using Nvidia GPUs as part of its fraud and money-laundering prevention strategy.
Indeed, as you read this, more businesses will be using GAN AI’s unique ability to recreate content with increasingly remarkable accuracy.
Here are a few applications of GAN AI:
Identity Protection, Mitigate Bias
Today, more than ever, we’re spending time online. It’s likely that with the rise of the metaverse, immersive virtual worlds will command an increasing portion of that time. Microsoft Teams is getting new 3D avatars in a push toward a metaverse environment, while Facebook’s metaverse is likely to feature hyper-realistic 3D avatars that use AI.
In these virtual worlds, Generative AI can help maintain the anonymity of individuals through avatars for people who do not want to disclose their identities.
Generative modelling also helps reinforcement machine learning models be less biased and comprehend more abstract concepts in simulation and the real world. Experimental results show that GAN AI can efficiently mitigate different types of biases while at the same time enhancing the prediction accuracy of the underlying machine learning model. The generative model synthetically produces new data, which augment the training set to overcome bias.
While GAN AI is being extensively used in the intelligent processing of realistic images and creating human-like voices for voice-over, narrations, and other audible solutions, it is helping marketers to identify and segment target groups for campaigns. It studies data to predict response to promotions and advertisements. Using text, image, and voice analysis, GAN AI can comprehend customer sentiment. The algorithm studies user data to decipher customer opinion towards products and services.
Just as GAN AI can recognise the style of an art piece and then create new artwork, like the Portrait of Edmond Belamy, it can also generate other forms of non-existent content – from building facades and apparel items to furnished rooms. GANs have already shown their ability in creating and modifying imagery.
Nvidia is using GANs for a variety of tasks. Nvidia’s GAN-based models include the AI painting app GauGAN, the game engine mimicker GameGAN, and the pet photo transformer GANimal.
In 2020, by applying a breakthrough neural network training technique to the popular Nvidia StyleGAN2 model, its researchers reimagined artwork based on fewer than 1,500 images from the Metropolitan Museum of Art. Using its DGX systems to accelerate training, they generated new AI art inspired by historical portraits.
The adaptive discriminator augmentation technique, or ADA, reduces the number of training images by 10-20 times while still getting great results. According to the chipmaker, the same method could impact healthcare, for example, by creating cancer histology images to help train other AI models.
“These results mean people can use GANs to tackle problems where vast quantities of data are too time-consuming or difficult to obtain,” said David Luebke, vice president of graphics research at Nvidia. “I can’t wait to see what artists, medical experts, and researchers use it for.”
GANs can also generate new and novel images just from textual descriptions of an image. Nvidia’s GauGAN2 can convert words to photographic-quality photos that one can then alter. Its deep learning model enables anyone to turn their ideas into photorealistic artworks. GAN can also be used for a quick image search, and when used in conjunction with unstructured data repositories, it can retrieve and identify visually similar images.
In the healthcare industry, from identifying errors in lab results that could lead to a quicker diagnosis and treatment for patients to facilitating drug discovery and novel drug creation, GAN AI is being used as an augmented intelligence for medical professionals. It is also being used to look into medication alterations by aligning treatments with diseases to generate new medications for incurable conditions.
In 2019, Insilico Medicine used AI to design, synthesise and validate a novel drug candidate in 46 days – 15 times faster than the best pharma companies.
A scientific paper published in Nature Biotechnology said that the company used GANs for drug discovery and biomarker development.
GANs also hold great potential promise in quality control, given their ability to quickly and accurately detect anomalies.
Clearly, applications of GAN go beyond creating realistic-looking photos, videos, and works of art. It can help speed research and progress in several areas of AI and be a key component of unsupervised learning. It is the branch of machine learning in which AI creates its data and discovers its own application rules.
Although GAN-generated content requires significant human work, a big budget, time, and technology, it disrupts more industries than we can imagine. Generative AI finds applications in crucial fields such as marketing, healthcare, and security. It is hoped that GAN AI will find more advanced applications as the technology evolves.
If you liked reading this, you might like our other stories