Will enterprises soon keep their best gen AI use cases under wraps?

The retail industry has no shortage of cases on display where generative AI has shown tangible benefits. Take for example French multinational Carrefour, who used it to make digital avatars and videos. They had ChatGPT write the script, and other gen AI tools to create a digital person who reads the script, a scalable process with at least one measurable benefit: speed.

“Suddenly, you can create engaging customer-facing videos at the click of a button,” says Oliver Banks, retail consultant and author of Driving Retail Transformation: How to navigate disruption and change.

Similarly, clothing brand Under Armour recently produced an ad that used AI-generated 3D models of the British boxer Anthony Joshua, based on videos they took of him in the past. “The new ads were created very quickly,” says Banks. “And the celebrity boxer wasn’t even included in the process of generating the new AI content.”

Oliver Banks

But according to Banks, one of the most impressive cases he’s seen in retail involves a company and source he can’t disclose. A chief commercial officer of a clothing company learned to use Midjourney to create images of new clothing products. Normally, a CCO develops ideas about what the market needs and communicates them to a design team, which produces sketches to then be reviewed by the CCO. It takes several iterations over weeks to get the drawings right. Now using Gen AI, the whole process takes only one person and it’s finished in a matter of minutes.

But what we’re learning from public announcements like these might just scratch the surface of gen AI use cases for the enterprise. As the technology exhausts its honeymoon period, and companies in all industries find new ways of using it to get ahead of the competition, a new race is on. The victors may not be the greatest innovators, but they’ll be the best at protecting their innovations.

Helping software developers write and test code

Similarly in tech, companies are currently open about some of their use cases, but protective of others. Gen AI has become useful in software development, and Planview, an Austin-based software company that helps project planning and execution — anything from digital transformation initiatives to construction projects — is a clear example. “The best uses of generative AI are linguistic in nature,” says Richard Sonnenblick, the company’s chief data scientist. “And software code is a language.”

Planview development teams use gen AI coding assistants, including GitHub Copilot, that help produce routine and tedious code. The tools also help junior developers by suggesting common code patterns they might not think of. “A lot of this was incremental innovation around the user interface of coding assistants that developers have been using for 25 years,” says Sonnenblick.

He reckons his company has seen a nearly 20% boost in productivity over the last year, and points to two particularly interesting things about the latest generation of coding assistants. One is putting in secure chat, so developers can converse freely with the LLM about their specific coding issues, as opposed to just going to OpenAI where there isn’t necessarily sufficient trust to share issues about proprietary code. The second is the tools go beyond coding assistance. They now use what they learn about a program to help build unit tests.

Planview

“Achieving complete coverage of your code during testing is a massive challenge in itself,” says Sonnenblick. “And unit tests are too tedious for humans to build reliably. But it’s a perfect job for LLMs that collect information as you write your program.”

While this example offers a peek at how productivity gains can be achieved in software development, what’s even more interesting is how tech companies, including Planview, are jockeying for competitive advantage in what they bring to market. According to IFI Claims, an organization that tracks patent data, one way of finding out where competition is heating up is to look at what kinds of patents people are applying for. IFI Claims writes that over the last five years, applications for gen AI patents have grown at a compounded annual rate of 31%.

Gen AI helps scientists develop new proteins

Another interesting set of use cases can be found, for the time being, in biotech. Sandra Castillo, senior scientist and computational biologist at Finland research organization VTT, is using gen AI to design new protein sequences based on what can be learned from nature. The new sequences are then tested at the VTT lab by using E. coli or other bacterial hosts to express the proteins.

In 2018, DeepMind, now a subsidiary of Alphabet, developed AlphaFold, a deep learning system that learns from a database of existing proteins and predicts their 3D structures. The database now consists of more than 200 million entries and the latest generation of gen AI has enabled improvements to the way that data is used.

“About a year ago, LLMs were developed for creating proteins,” says Castillo. “You can treat protein sequences as sentences and amino acids as words. The LLMs are giving us much better results than we had before.”

VTT

One model Castillo uses at VTT is a variational autoencoder to generate new polyhydroxyalkanoates (PHA), proteins that can be used to create biodegradable plastics. “You can give the plastics different properties by designing enzymes that use different monomers,” she says. “The intention is that in the future, we’ll replace the plastics we used today with biodegradable versions.”

An additional use of new proteins is to detect chemicals in the body very quickly. “If you want to test for testosterone or other hormones, we can design proteins that bind specifically to what you’re looking for,” she adds. “Or you might use new proteins to test for other compounds, such as antibodies. In the future, we could use generative AI to quickly make a test for diseases like Covid.”

So far, Castillo has been publishing her work with open-source licenses. But discussions are underway within VTT about how to protect and commercialize the AI models they develop. “I know that some companies are selling generative AI models based on the public models,” she says. “Some will probably sell a service where they adapt open models to specific problems.”

While it isn’t yet clear how the line will be drawn between what is public domain and what can be sold, scientists and biotech firms are certainly on their way to finding out—and once they do, they won’t be saying much publicly.

Helping service providers share knowledge

IT services provider Cognizant has every interest in sharing best practices across its 9,000 projects running in 50 different countries at any given moment — and they’re relying on knowledge management to do just that. Last year the company launched a grassroots innovation program called Bluebolt to empower its global employees to dream up and submit ideas, large and small, that can be used in projects around the world.

“We encourage all of our 350,000 associates to come up with ideas,” says Alexis Samuel, SVP, who as global head of delivery excellence with Cognizant, is responsible for the governance and oversight of all IT service projects the company delivers globally. About 20% of them have been implemented and described in innovation libraries, he says, so they can be re-used elsewhere.

“We started experimenting with gen AI about six months back to help our associates ask the right questions and get the right answers,” he adds. “We uploaded the library of innovative content from the last seven years across Cognizant, ring-fenced the whole thing, put ChatGPT 3.5 Turbo on it, and created a virtual innovation assistant that associates can use to ask questions and get relevant answers.”

Cognizant

So if somebody is implementing an app modernization program, moving from on-prem to cloud, and they want to know what problems might occur, the assistant can give a list based on what’s in the library, and then the associate can ask for solutions. “If they use any of them, they’re encouraged to update the library to include anything new they did,” he says.

Cognizant curates the information to make sure the tool doesn’t reveal details that shouldn’t be shared, and the engine returns an abstract version of ideas that can be re-used in other projects, without showing private client data or personally identifiable information.

They’re now feeding the model new information with a three-week lag. To date, about 7,000 associates have used the tool, around 19,000 transactions have been recorded, and 60% of the feedback is positive. “So far, the virtual innovation assistant has been very valuable to those who use it,” says Samuel. “The biggest challenge is that in an organization our size, some employees may not know the tool exists, despite all the announcements we’ve made.”

Far flung teams in large companies may have trouble hearing about specific tools, but the broader market is getting the message loud and clear that gen AI is revolutionizing industries. But the most important use cases are probably the ones we don’t know about.

“While companies have proudly boasted about what amazing things they’re doing with this new technology, they’re starting to find ways of using it for competitive advantage,” says Banks. “There’s no way they’ll let the rest of the world in on that.”