Our next installation for our StructuredWeb 2023 Predictions Series is all about Generative AI. You may have heard one of the latest buzzwords going around: generative AI. The process uses artificial intelligence (AI) and machine learning (ML) algorithms to generate new content based on AI training data. Although deep fake technology receives the most news, this AI process actually lets us create all kinds of useful content, including material for channel partner marketing. From simple (and not so simple) prompts, the AI creates text, photorealistic images, audio files, and videos. Sometimes that output is obviously computer created — but output can be startlingly realistic.
In 2022, AI modeling made rapid advances. Machine-generated content has grown into a regular business asset. As channel partners apply AI and automation to customize and personalize assets for end customers, all parties will benefit. Such integration will help vendors create and share marketing assets with partners faster and at a lower cost. It also will further enable partners to apply automation to customize and personalize assets for their end customers.
Using generative AI applications, smart content curation can identify and assemble campaign material based on known, existing content. Using textual descriptions and material from products, marketing activities, and branding assets, these neural networks can build new marketing material.
And similar AI capabilities could be applied to content distribution. Instead of throwing everything at everyone, AI could select and deliver the appropriate assets to the appropriate clients. All automatically, based on real data and unsupervised learning, and using defined generative models based on channel requirements. It’s exciting stuff. Let’s look at some of the ways deep learning and generative modeling are already producing creative work in the marketplace.
AI models are becoming better at generating text from prompts. Although good creative writing is still beyond the technology right now, creating marketing content is well within its ability. All it takes is to refer the app to your product descriptions, features, and specifications. Then you ask for a sales pitch, press release, or perhaps just a better product description. The AI condenses the source material and creates something similar to the underlying pattern of ad copy it trained on. The output might be good to go as is. But if not, it might require only minor tweaks to be useful.
Prompts for writers, designers
One major use of generative AI is creating starting points for writing or design. Using an online app like Sudowrite, you can enter 20 words or more to describe a story or scene. The app then returns several versions of a fleshed-out scene, ready for expansion. If you want more, you can ask it to create characters, sights, sounds, smells, and sense of touch. All of this is highly useful when building up a scene or story.
Other writing apps, like Ryter, Jasper, and Grammarly, offer similar functions, creating new content from general prompts. Some mainly focus on creative writing (fiction) but others lean more toward marketing and blog content. The writing software market has exploded over the past decade. Increased demand is fueled by improvements in neural network hardware and software as well as better models for applying AI.
Artwork and graphic design
On the graphic side, apps like MidJourney, Dall-E 2, Google’s Imagen, and the open source Stable Diffusion produce incredible content. From simple text prompts, the software builds realistic images that you can continue to tweak and expand upon. Realistic photographs of human faces are possible, with some work on the user’s part. Every new version of these apps shows improved quality of image generation.
Generating photographs or artwork of cityscapes or landscapes is easy. Old images can be colorized, increased in resolution, or even modified for items or parts they contain. Don’t like that beachball in a stock photo? Ask the AI to remove it. Realistic images of other objects, real or unreal, can be created from any starting source. A sketch to image function can convert a simple stick figure drawing into a beautiful painting or ad banner.
Using diffusion models, these systems generate new images by adding random noise to existing images. Over many passes, the app learns how to remove the noise that it doesn’t think fits the prompt. In a somewhat backward-to-forward method, output emerges from chaos into a final, polished image.
What are generative adversarial networks?
Generative AI employs various techniques to work its magic. One of those techniques employs a generative adversarial network, or GAN. The term describes the way two neural networks can compete against one another for the greater good. One network (the generator) creates new content, the other network (the discriminator) compares that content to the original source material. When the underlying pattern related to the source material satisfies both generative adversarial AIs, the new content is output.
What is GPT-3?
The AI process can take different angles to produce content. GPT-3, called a transformer, is often used to generate text. GPT-3 can give responses based on your prompt or question. Its language models include Wikipedia and much of the internet. When such large models are employed, it improves the results.
The output from GPT-3 ranges from human-like chatbots, autocompletion, website design, software coding, database statements, and document interpretation. Customer service, sales communications, and marketing copy are well within its capabilities, and large corporations already are using it. Earlier in this article we mentioned the writing app Sudowrite. Under its hood, it uses GPT-3 power. Other writing apps, using natural language understanding and machine learning models of their own, take different approaches. But the end results are usually similar.
What advances do we expect soon?
AI is growing daily as companies polish their modeling systems and software. Current buzz is mostly about text-to-image generative AI models. But the real impact and value is likely to be in AI text generation. As large language models capture more and more modern language concepts, the AI tools using them could improve exponentially.
In the not-distant future, look for AI products to:
- Automate outgoing emails from sales development reps.
- Answer buyers’ questions — accurately — about products.
- Be the main email handler for prospective customers in the sales process.
- Offer immediate guidance and feedback to sales agents while they speak to customers.
- Summarize sales discussions and create next-step marketing suggestions.
And it’s not all about sales and marketing. AI will likely:
- Transform legal research and discovery by summarizing and answering case questions during litigation.
- Help clinicians diagnose problems, compose medical reports, and offer treatment suggestions.
- Create polished news articles from reporters’ notes, videos, and audio recordings.
- Become a solution for help desks, call centers, and customer service as machine learning models become proficient in problem solving.
Generative AI and beyond
As AI grows, it will likely become the first-line interaction consumers have with companies, on any subject or topic. Companies that embrace the new and growing technology of generative AI will benefit in many ways. The businesses that don’t embrace AI might go the way of floppy disks and CRT monitors.
With custom AI-enabled workflows and automation of marketing and support, StructuredWeb provides companies with the technology of the future, today. Extend your platform business logic by using StructuredWeb to automate labor-intensive tasks and processes — and enjoy AI’s benefits.
Additionally, look out for our next piece in our predictions series about High Touch Customer Service – a must-have in today’s channel marketing ecosystem.