Generative artificial intelligence (AI) is best known for providing answers quickly and creating AI art on the fly, but business use cases abound. Here's a primer on generative AI and its application.
What is generative AI?
Generative AI is a type of artificial intelligence that can produce content such as text, imagery, audio and data based on what it has learned from a massive training set of data. Generative AI has reached a tipping point as technologies such as OpenAI's ChatGPT and DALL-E have become popular.
In addition, technology vendors are racing to include generative AI into products and services. Businesses are also exploring how to integrate generative AI into multiple use cases.
Generative AI takes inputs from training data and produces similar outputs with a unique spin. As a result, generative AI can seem creative. More specifically, generative AI relies on a few different models.
Here’s a look:
- Generative Adversarial Networks (GANs), which have two neural networks. One is a generator and the other is a discriminator. The two networks compete as the generator creates new data instances and the discriminator values the quality. The generator improves its output based on feedback from the discriminator, which aims to distinguish between generated data and the real training data. You could think of a GAN as a model that somewhat replicates the writer and editor relationship.
- Variational Autoencoders (VAEs), which introduces a probabilistic component to create diverse and realistic outputs.
- Transformer models, which include GPT (Generative Pre-trained Transformer). Transformer models are a type of deep learning architecture used for natural language processing tasks. These models generate text with context based on patterns from training data.
Github, which offers Github CoPilot generative AI, has more on the various models as does Nvidia.Generative AI models have limitations including the need to compute at scale and outputs that are only as good as the quality of training data. Nevertheless, generative AI is showing it can create new content such as marketing content, social media posts, scripts and books to name a few. Beyond content, generative AI can create new data to train other AI systems, compress data by removing redundant information and create new data as well as programming code.
Where is generative AI being used?
The bigger question is where generative AI isn't being used. Technology companies are moving quickly to integrate generative AI into productivity applications. For instance, Microsoft is integrating ChatGPT throughout its applications, Salesforce is doing the same and Slack has plans to use generative AI and consumer apps from the likes of Redfin and Zillow are doing the same. When you consider search engines such as Microsoft's Bing and Google have generative AI plans it's likely that most of the software you touch will have ChatGPT or a similar technology embedded.
Simply put, technology vendors are embedding generative AI and related tools everywhere. However, enterprise buyers are wary of AI and its implications for compliance, first party data and security even as boardrooms push for rapid adoption.
The list of companies leveraging generative AI is expanding:
- Shutterstock's generative AI way forward: 6-year training data deal with OpenAI
- JPMorgan Chase: Digital transformation, AI and data strategy sets up generative AI
- Rivian: AI, data power customer experiences
How is generative AI different than AI?
Generative AI creates new content, chat responses, designs, images and programming code. Traditional AI has been used for detecting patterns, making decisions, surfacing and classifying data and detecting anomalies to produce a simple result.
Machine learning is a type of artificial intelligence. Machine learning is used to learn from data patterns without human support. Given the scale of data, machine learning enables the models used for AI.
What's a Large Language Model (LLM)?
A LLM is a machine learning model that's trained on text data from multiple sources at scale. LLMs can complete natural language processing tasks and answer questions in a conversational way. Vendors are creating proprietary LLMs as well as tuning for specific use cases and behaviors.
LLMs are trained with books, articles, code and forms of text. This training data is then used to generate text, translate languages and answer questions via natural language processing (NLP). While LLMs are still being developed, CXOs have noted that they can be used for code generation, technical document creation, marketing and data analysis.
How will generative AI impact work?
For now, generative AI is being seen as a grand experiment rolling out in real time. However, as numerous companies--Microsoft, Google, Salesforce to name a few--look to embed generative AI in productivity tools the technology's reach will be broad.
In a research note, Constellation Research analyst Dion Hinchcliffe said the impact of generative AI will advance industry use cases as well as work in general.
"The broadest and most impactful area of AI will be in general purpose capabilities that quickly enable the average professional to get their work done better and faster. In short, helping knowledge workers work more effectively to achieve meaningful outcomes to the business. It's in this horizontal domain that generative AI has dramatically raised the stakes in the last six months."
Companies are likely to put resources behind creating generative AI models, algorithms and tools for competitive advantage. CXOs should spend time exploring OpenAI's ChatGPT 4 subscription service as well as Google's Bard to find use cases.
What are the use cases for a generative AI model?
Constellation Research CEO Ray Wang recently outlined five emerging use cases. They are:
- Marketing. Diffusion models will dynamically generate content, provide translation capability, and run A/B and experimentation tests for user experiences. Personalization models will gain greater context, enabling hyper targeting for campaigns, ad networks, and polling with ChatGPT.
- Sales. Sales specific tasks such as pipeline reviews, scheduling meetings, install base analysis, and forecasting will move from manual to automated.
- Service. Crawlers inside one’s internal systems can scan knowledge bases, augment case history, and hasten issue resolution. The AI can create new case tickets, augment missing information, and predict customer satisfaction.
- Commerce. Speed of product catalog creation will improve as diffusion models will take prompts from regulatory requirements enabling faster global rollouts of new products and services content.
- Customer success. Generative AI will identify accounts with low adoption and automatically identify at risk customers based on their level of interaction to increase the frequency of engagement.
There will be other personal use cases too. For starters, generative AI tools can be entertaining. They can also write thank you letters, email responses and data profiles.
How are companies using generative AI?
In earnings conference calls, executives typically get a question or two about generative AI. The answers fall into a few key categories.
- Technology vendors are integrating generative AI rapidly. Whether it's ServiceNow, SAP, C3 AI or Google there's a generative AI story that revolves around integration with their respective platforms. If these tech vendors hold their timelines, you'll be using generative AI indirectly through your personal and business productivity apps.
- Broad industry usage. Generative AI is turning up in financial services, insurance, healthcare, hospitality, fast food and a bunch of industries. CEOs are watching generative AI closely, pondering integration scenarios, new experiences and compliance issues. These CEOs are very interested in generative AI but do want more transparency into the models.
What are the benefits of generative AI?
Generative AI is likely to have a bevy of benefits including automating manual tasks, augmented writing, increased productivity and summarizing information and data.
The technology can also be used to explore new markets, enhance products, personalize experiences, create new knowledge, educate, boost decision making, gather information and optimize processes.
These benefits may become more evident as technology vendors embed generative AI into their applications. Amazon CEO Andy Jassy said on the company's first quarter earnings conference call:
"I think if you look at what’s happened over the last 9 months or so is that these Large Language Models and generative AI capabilities, they’ve been around for a while, but frankly, the models were not that compelling before about 6, 9 months ago. And they have gotten so much bigger and so much better, much more quickly that it really presents a remarkable opportunity to transform virtually every customer experience that exists."
What are the risks of generative AI?
There are also risks to balance out the benefits. For starters, generative AI may replace human workers to some extent. Workers will also have to upskill and reskill due to automation and generative AI, according to Coursera. Other risks include:
- Data biases. Generative AI algorithms can only be as good as the data set it is being trained on. If generative AI is being trained on a flawed model it'll only scale mistakes.
- Transparency. Generative AI models are complex, and it will be hard for businesses and consumers to understand how an answer was generated. This problem will become more important as various generative AI technologies and algorithms are integrated.
- Ethics. Generative AI applications are trained by data provided by humans. There's the potential to scale unethical behavior and bias.
- Business models. Sourcing of material has been abstracted in more popular generative AI technologies. In other words, it's unclear how intellectual property owners will get paid.
- Black box thinking. Humans will still need to offer expertise for decision making.
- Model sprawl. It's increasingly clear that enterprises will use multiple generative AI technologies. At some point (probably soon), these early adopters will have to wrangle these tools and make them work together.
- Security. Enterprises are concerned about sharing first party data with LLMs. Speaking at Domino Data Lab's Rev 4 conference in New York City, Jan Zirnstein, Director of Data Science at Honeywell Connected Enterprise, said the company has been looking at generative AI use cases but questions remain. "Generative AI has tipped the public perception of what AI is, but tipped it a little too far," said Zirnstein. "There's nothing in the actual training model and architecture that's tied to truth and factual correctness. We're looking at use cases tied to where factualness isn't imperative like saving time on the creative side. There are also use cases on the summarization side."