Generative AI (artificial intelligence) for text, images, and videos has been all over the business press over the past few months. The technology is transforming how businesses create content. We should know. Persado Motivation AI is a kind of Generative AI for text, so it’s definitely no surprise to us that enterprises are increasingly finding ways to embed this technology into their platforms and workflows.
But we also wanted to hear what an outside expert thinks of this burgeoning world of Generative AI. Persado VP of Go-to-Market Strategy Vipul Vyas recently hosted a webinar featuring Forrester analyst and guest speaker Rowan Curran on the future of Generative AI technology. Rowan covers AI, machine learning, and data analytics for Forrester.
His take? There’s an evolutionary explosion of possible use cases for Generative AI for text, images, and a range of other front- and back-end use cases in the next five years. In fact, he predicts that 10% of Fortune 500 organizations will generate content using AI in 2023 alone. Read on for a full summary of the conversation (including some of our perspectives that reach beyond) and to hear answers to the following questions:
· What is Generative AI?
· How does it work?
· What are its business use cases?
· How will Generative AI evolve over the next five years?
What is Generative AI and how does it work?
Generative AI uses models that have been trained on large volumes of “language data”. This includes written text or audio files that train the AI to understand the mechanics of human speech and what we mean, as humans, when we say or write certain things. This area of AI is known as “natural language processing”.
Building on this foundation, Generative AI models then have an additional application layer that adds capabilities. For example, a Generative AI for text leverages the AI model’s knowledge of human language patterns to create fluent, human-sounding outputs. This is also part of what Persado does). A Generative AI for images, in contrast, associates words with their physical equivalents (dog, rose, supernova) and with images (photos, paintings, drawings). The AI can then generate new images based on natural language prompts coupled with image recognition or computer vision capabilities.
The evolution of Generative AI
As Rowan notes on the webinar, there has been an explosion in Generative AI adoption in the past few years. The technology itself is not entirely new, however. One of the catalysts for this explosion was the discovery by Google researchers of a technique that led to “Generative Pre-trained Transformers” (or GPTs).
Transformer models are a type of neural network — which are themselves a subset of machine learning. Neural networks mimic the thought patterns of the human brain. These models have an architecture that allows them to evaluate words both individually and in the context of a sequence, as in a sentence. This allows transformer-based language models to quickly ingest and process huge bodies of training data. Developers have leveraged the vast corpus of language data housed on the Internet to train transformers and create the “large language models” we’re talking about in this post. Beyond the inputs, the unique methods used by transformers produce outputs predicted as the logical words or sentences that follow relatively short text-based prompts.
There are a number of generative pretrained transformers in the market today. They include GPT-3 by OpenAI, the Megatron-Turing Generative Pretrained Transformer developed by Nvidia and Microsoft, Google’s BERT model, and others. These models have a powerful predictive capacity that enable them to generate content at scale.
Yet these models don’t have true “natural language understanding” the way humans do. That can create problems.
The risks of using Generative AI
During the webinar, Rowan highlighted the recent release by Meta of a transformer named Galactica to illustrate the dangers of large language models. Galactica is a large language model for science, but it was pulled down after three days. Early users found that it generated text riddled with falsehoods and inaccuracies, delivered with the authoritative tone of valid scientific writing. It even attributed false findings to real-world researchers.
The Persado view is that Galactica offers a case in point for why enterprises must proceed with caution when leveraging Generative AI for text. Transformer models generate content by predicting the logical continuation of a given language sample. Its strength is volume, not quality, accuracy, or context awareness. Alone, these platforms will not (and cannot) adapt the text to the reader or the medium. To do that, a language model must be “fine tuned” — the technical term for filtering the training data and adding capabilities to improve accuracy and specificity for a specific use case.
Alternatively, a language model could from the outset leverage training data that is fit for purpose. In the case of Persado, we began training our proprietary language model a decade ago using enterprise language. That’s just one way we ensure that our Motivation AI Platform fits the needs of our customer.
What are some front-end business use cases for Generative AI?
The possibilities for Generative AI are vast. Consequently, experimentation and testing is rampant in five key areas that will change how we work.
Generative AI for text-to-image generation
This quickly-evolving technology has the potential to democratize human visual expression by making it more accessible to everyone. Many text-to-image Generative AI programs have been built on large language models and transformer networks created by organizations like OpenAI (the creator of the GPT-3 large language model), Google, and others.
On the webinar, Rowan shared the use case of creating AI-generated images for business presentations. The goal is not so much a gimmick as a fast and convenient way to adapt a stock photo or graphic for a different context.
Using AI-generated images may be a relatively new phenomenon, but it is accelerating fast. Rowan shared that Shutterstock, the visual content platform, recently announced it is integrating Generative AI into its platform so customers can AI-generate their own stock images. In addition, Shutterstock established a contributor fund to pay the photographers whose images were used to inform newly generated assets.
There are also a collection of new platforms emerging to fill demand for AI-generated visual content. Craiyon and Dall-e are two examples. Both of them are free to start, and their licenses allow for commercial use of generated images.
Generative AI for text and language
Generative AI for text relies on large language models that can naturally generate text in response to language prompts. On the webinar, Rowan shared that a number of case studies for human scalability exist within this framework. For example, a marketer may need to write 20-50 variations of the same call to action for an email campaign. She can plug an initial statement into an AI platform and ask it to create dozens of variations. Generative AI produces rote content efficiently, thus freeing up human time for more demanding work.
Improved summarization is another benefit of Generative AI for text.
“This technology can summarize complex documents into a human-readable format that is actually useful,” reported Rowan. “And for me, this is one of the places where Generative AI is really going to come into our daily enterprise lives as a trusted and supportive coworker.”
The benefits of AI text-generation
Generative AI for text won’t replace humans in the workplace, Curran predicts. It will help us be more productive, however. Instead of looking at various network folders and printed material, an employee can send a query to a large language model with an index of organizational data. If prompted, the model can then produce a summary with the relevant information. Parts of the business that could benefit immediately include call centers, claims specialists at an insurance company, and compliance professionals, just to name a few.
The Persado view on Generative AI for text — the space in which Motivation AI operates — is grounded in our experience generating marketing and customer service messages that motivate customers to engage and act. Persado customers in retail, financial service, travel, telecom, and other areas are already leveraging our Motivation AI platform to customize language for acquisition campaigns, customer service, payments, loyalty, and other use cases.
For example, Persado has worked with retailers like Marks & Spencer and Michaels to personalize customer messages. Across email, text, social channels, and other areas of engagement, these brands are driving incremental revenue, loyalty, and customer engagement.
For code generation
One of the big surprises to come out of the Generative AI space is that large language models can write computer code. It was a surprise because the large language model developers hadn’t intended to train their transformers to write computer code. That came about as a result of the fact that web sites contain programs that the models “read”. Upon discovering that, developers at OpenAI went on to train a model on purpose. The result is Generative AI that can address many tedious and time-consuming tasks that application developers must fulfill when drafting and debugging code, or translating it from one platform to another. Imagine how much of a non-issue Y2K might have been if we’d had this tool in our pocket?
Digital customer service agents use Generative AI for text to create a more natural conversational experience. The technology is evolving so that customers can write almost anything into a chatbot and receive understandable responses. Large language models provide a fundamental knowledge base to improve the chatbot experience.
For synthetic video and audio
This evolving use of Generative AI is on the horizon. Several companies are developing technology to allow you to train an AI to produce a synthetic version of your own voice (or another person’s voice) that will then feed into text, which a video version of a person can speak. These synthetic voices embody semantic and emotional flexibility based upon the text, especially if a text model tags different parts of speech and inflection points. Building on this Generative AI for language, creating synthetic humans isn’t far behind.
“There’s actually huge potential for this to be very positive for individuals and the enterprise, especially for training and onboarding,” says Rowan on the webinar.
Synthetic video and audio can transform traditional video production. With AI and automation, organizations can produce more for less. More importantly, advances in synthetic video and audio have the capacity and scale to make video more engaging and easier to edit. Companies don’t have to reshoot a video to make updates. Customization is also easier. For example, a narrator can have a British accent or a southern US accent to better connect with specific audiences.
Backend uses for Generative AI
According to Rowan’s presentation, the term “Turingbots” is a Forrester catch-all term for the broad category of using large language models and other types of Generative AI to create code for software applications. Examples are Google’s CoDeX and GitHub’s Copilot, which have been around for a few years. A user gives the model a prompt such as asking for an application that has a blue button on the left side and a text entry on the right side. The program will code the instructions into a prototype.
This advancement has primarily appealed to individuals, but it’s catching on in enterprises. A layering effect on top of the large language models is occurring, which provides a protocol for creating enterprise apps.
“There’s an increasing focus on how Turingbots can be used to automate not just the initial coding of an app and its wire framing, but how can we use Turningbots to take away the really challenging but necessary coding demands around integrating with security systems, connecting to the broader enterprise databases and software support that often is very specific,” said Rowan. “Turingbots can be used to code sections that enterprise developers often don’t want to do.”
Generate synthetic data
Enterprises also are using large language models and transformer networks to generate synthetic data to further train AI models. For example, enterprises can generate synthetic data to construct a data set that may not exist in the real world.
Where will Generative AI take us in the next five years?
Rowan sees this evolution bubbling up from the bottom as well as from the top down.
“From the bottom, it’s going to come up through the integration of text-to-image generators such as Shutterstock,” said Rowan. “I expect it won’t be very long until PowerPoint and other presentation tools have text-to-image generators directly integrated into them. For the Generative AI text piece, I think this technology will very quickly become part of how we generate marketing emails at a large scale.”
Generative AI also harnesses the potential to unleash human creativity. So many people have unique ideas but struggle to translate them onto paper or digital art. AI art generators will be game changers for untrained artists, as well as for people with disabilities or cognitive impairment. Someone who could never manipulate a paintbrush can use Generative AI for language and text to produce images or art.
“This enablement of a broader swath of humanity to participate in the creation of content and art showcases the potential for Generative AI beyond the enterprise,” said Rowan.
At the enterprise level, Generative AI will complete tasks that many workers don’t enjoy. It will vary from company to company but can include stock image generation and writing product descriptions naturally at scale.
The list of rote activities that are time consuming, but don’t necessarily require an expert to produce, are many and varied.
Ethics around Generative AI—emerging questions
How do we store and manage AI-generated content? Rowan posed this question and posits that digital asset management vendors will need to tackle this challenge soon.
Generative AI could also create trouble for individuals and enterprises, specifically with real-time deep fakes. Bad actors can create videos using avatars without consent from the real person. People who are filmed often, such as politicians and actors, could be targets.
Technology providers are already developing solutions to address this challenge. For example, Intel recently launched a real-time deep fake detector. This technology employs a novel approach that compares one frame to another to determine whether the skin tone has changed from blood pumping under the skin. No blood means it’s not a real person.
Questions are also beginning to circle around the use of Generative AI for text in academic settings. For example, is it plagiarism? A set of tools is emerging to detect text that is created by large language models. If a student includes significant amounts of unsourced text in a paper, is it cheating? The ethical questions around Generative AI are only just surfacing.
Another set of negative use cases are emerging related to websites that are expressly developed to drive search engine optimization and different types of tagging. These websites exist today and are programmed using basic AI. Because AI makes it fast and easy to create these fake sites, it would take very little time to construct a vast, fake Internet, adding to the storage and validity issues.
New apps are emerging every day that allow people to work with Generative AI for text or for images. Having many active, early users will enhance creativity and idea generation. Anyone who can speak, type, or express a thought can create. An abstraction layer is completely removed, replaced by something extremely human, which expands the power to communicate.
In some respects, Generative AI will train us to more clearly articulate what we want. As the technology evolves, human brains will be rewired to better communicate with it.
Moving forward, users must adapt how they communicate to get the right response from Generative AI. A mini-culture will take hold to evolve language. Similar to the way that texting became a primary mode of communication and led to a parallel language comprised of acronyms and icons, something similar will evolve as Generative AI takes hold.
Generative AI is on the cusp of changing enterprise workflows. Watch the entire hour-long Persado webinar to learn more about this exciting topic.