Generative AI is a type of artificial intelligence that uses unstructured deep learning models to produce content based on user input. As part of this process, generative AI uses a foundation of machine learning and deep learning algorithms. The content it creates includes written materials, images, video, audio and music and computer code.
For example, when a human types a question or statement into ChatGPT – a pioneering example of generative AI – it delivers a brief but reasonably detailed written response. A user can also enter follow-up questions and engage in an ongoing conversation with the chatbot, which can remember details from earlier in the conversation.
Generative AI has recently garnered attention because major breakthroughs in the field are accelerating. For example, OpenAI’s ChatGPT can generate grammatically correct text that appears to be written by humans, and its DALL-E tool can produce photorealistic images based on word input. Others companies, including Google, Facebook and Baidu, have also developed sophisticated generative AI tools that can produce authentic-looking text, images or computer code.
For more information, also see: Understanding the ChatGPT AI Chatbot
How Does Generative AI Work?
Generative AI creates new content based on a training set. Researchers feed enormous volumes of data—words, pictures, music or other content—into a deep learning system called a Generative Adversarial Network(GAN) framework. The supervised neural network sifts through the data and uses a system that rewards successes and penalizes errors, mistakes and failures, advances. Over time and with human oversight, it learns how to identify and understand complex relationships.
The original OpenAI Codex used for ChatGPT, for instance, is derived from more than 700 gigabytes of data collected from the Web and other sources. This includes content from books, magazines articles, websites, technical manuals, emails, song lyrics, stage plays, scripts and other publicly available sources. Natural language models like ChatGPT typically rely on mathematical models called word vectors to weight and rank phrases. They also rely on a technique called Recognizing Textual Entitlement (RTE) to better understand word relationships, such as they’re, their and there or whether there are strong winds or the road winds.
Entailment, Contradiction, or Neutrality
As researchers add data to a natural language model like ChatGPT or LaMDA and additional training takes place, the system continues to compare and contrast words through a lens of entailment, contradiction, or neutrality.
For instance, the premise “A dog has legs” entails “legs have feet,” but contradicts “dogs swim under the sea,” while remaining neutral to a statement like “all dogs are good.” As the system runs through millions of combinations, it learns how to build an accurate and contextually correct predictive model.
According to OpenAI, researchers fed more than 300 billion words into the actual ChatGPT model. Initially, human AI trainers provided input for both sides—as a user and as an AI assistant (generator and discriminator). Humans then reviewed randomly selected model-written messages, ranked various completions from the model, and fed them back into the GAN to further train the reward model. The result was a reasonably accurate reinforcement learning algorithm that, with additional training and user input, continues to improve over time.
Researchers use similar techniques to classify pictures of birds, clouds, trees, faces, cars, and millions of other objects. Over time, the reward model is updated and refined—and it continues to produce more realistic details. In fact, models like DALL-E and Google’s MiP-NeRF produce highly detailed effects, including shadows, color gradients and textures. This makes things such as a stone surface or water shimmering on a lake look remarkably realistic.
For more information, also see: AI and Deep Learning
What Does Generative AI Do?
Generative AI and other foundational AI models are dramatically influencing the development of AI, boosting assistive technology and enabling powerful capabilities for nontechnical users. This includes content creation from text to code to images to music.
OpenAI is far from the only company to develop natural language chat capabilities. Google’s LaMDA and Bard, Apple’s Siri, Microsoft’s Cortana, and Amazon’s Alexa all generate written or spoken words through the use of Generative AI models.
Other Generative AI tools, such as DALL-E and Google’s MiP-NeRF, can generate photorealistic images based on word input. For instance, a web designer might type the words “classic Spanish plaza” into the DALL-E engine and view an image that looks incredibly real—though it doesn’t represent any actual place. Likewise, a person might ask DALL-E to produce an image of a woman sitting at a café in the style of Monet and nearly instantly view an image that looks like it was produced by the artist.
Generative AI is also used to produce audio and music—including full-fledged compositions and specialized sound effects. Several companies, including Amper Music, Aiva, Amadeus Code, Google Magenta and MuseNet are capable of generating original music with multiple realistic-sounding instruments. A user can request a genre, artist or style—say jazz, Mozart, the Rolling Stones or upbeat—and listen to the resulting AI generated composition.
Another burgeoning use case for Generative AI is software development. Platforms such as Amazon’s CodeWhisperer and GitHub’s CoPilot introduce natural language-based low-code and no-code platforms for developers. With Generative AI, a software developer can speak or write a request into a platform and view actual lines of software code in Python, R, Java or other major languages. This allows developers to work faster and create reusable modules more easily.
Story and Game Development
More advanced use cases revolve around things like story and game development, robotic designs and even debugging products or operational methods by asking questions and probing a topic. What’s more, by asking the Generative AI tool to provide ideas and concepts, it’s possible to explore themes and even develop new and different digital and physical objects.
How Can Businesses Use Generative AI?
The history of AI and business is filled with innovation, disruption and profound changes. Generative AI promises to lead organizations down the same path. Some of the leading use cases for generative AI in the business world include:
- Marketing and sales. Generative AI systems can produce a variety of written content for emails, website text and images, brochures, eBooks and product guides, product labels and internal documents. Organizations can also use the technology to analyze customer feedback, identify risks and opportunities, and deploy highly usable and functional chatbots.
- Human resources. HR departments can tap generative AI to write an enterprise handbook, job descriptions and interview questions. A chatbot can deliver information and self-help for employees. This might include automating on-boarding or providing options and advice for choosing healthcare insurance or a retirement savings strategy.
- Operations. Customer service chatbots can help companies manage inquiries and direct people to the right information—and hand them off to an agent when it’s beneficial. Generative AI can also identify errors, defects and other problems through comparative images. For example, a company might use generative AI to create an ideal image of a highly technical component and then capture images during manufacturing to ensure they adhere to quality control standards.
- Other business uses: Numerous other business use cases exist. These include modeling systems used for research and development (R&D); reviewing text in documents to ensure they meet legal and regulatory standards; and optimizing and improving general employee communication, including emails and business presentations.
For more information, also see: What is AI?
Generative AI: A Brief History
Artificial intelligence research began to take shape during the 1950s. Alan Turing and other scientists began to explore ways to create computing frameworks that could duplicate human thinking.
By the 1960s, so-called Markov models began to appear. These probability-based algorithms could generate speech or text based on basic mathematical models—though with limited success. By the 1990s, more sophisticated generative models began to appear. Over the last decade, GPUs and advances in deep learning have ushered in far more advanced AI. Today, these recurrent neural networks can generate content in a way that approximates—and in some cases exceeds—human artists, musicians and writers.
At this point, artificial intelligence – particularly generative AI – is fundamentally reshaping the way people and businesses act, interact and process information.
Market research firm Grandview Research projects that the Generative AI market will grow by 34.4% annually through 2030. It says that the technology has value across a wide swath of industries, including finance, healthcare, automotive and transportation, information technology, telecommunications, and media and entertainment. Generative AI can transform tasks as wide ranging as marketing, image classification and quality control.
In fact, Gartner has proclaimed that Generative AI technology will revolutionize digital product development. The consulting firm reports that by 2025 about 10% of all digital content will derive from these algorithms. No less important, McKinsey & Company reported, Generative AI will fundamentally change job roles along with the way people work. It noted: “The rise of generative AI has the potential to be a major game-changer for businesses.”
On a related topic: The Future of Artificial Intelligence
What Ethical Concerns Exist About Generative AI?
Not surprisingly, the rise of Generative AI has unleashed concerns. One fear is that the technology will replace humans for many job functions. However, the technology—at least for the next several years—will more likely serve as a complement to humans.
Lack of Accuracy
For instance, ChatGPT text requires human review because it isn’t always complete and accurate. Blindly plugging in text could lead to a variety of problems, ranging from accusations of bias to legal issues.
Legal Issues and Plagiarism
Likewise, an enterprise must take caution about what types of music, images or other materials derive from Generative AI. Because these models are built from actual content produced by writers, musicians and painters, they can raise questions about ownership, control and copyright.
For this reason, generating a photorealistic image that’s similar to the specific style of an artist could raise questions—and even lead to a lawsuit or public backlash. There are also growing concerns about the technology being used for deepfakes and as a way for students to avoid writing essays and papers.
Privacy and Security
Privacy and security concerns are also at the forefront of Generative AI. Some data used to build models may inadvertently contain private data and information that might later be exposed. Equally concerning: cybergangs and other criminals have already begun to use Generative AI to produce highly convincing documents, software and images that become part of social engineering campaigns.
Another issue is the overall societal impact of Generative AI, particularly tools like ChatGPT and Bing’s AI chat feature (which is built on the ChatGPT framework). Some observers, such as New York Times technology columnist Kevin Roose, have raised concerns about the technology being used to manipulate humans—including in harmful and destructive ways. In addition, they have voiced concerns about the technology carrying out its own dangerous acts.
For more information, also see: AI Software and Tools
Bottom Line: What is the Future of Generative AI?
While nobody can predict the exact trajectory of generative AI, it’s clear that it will make a profound impact on businesses—and society. Clearly, as the technology advances, the capabilities will expand. Within a few years, the technology may well be capable of writing full-fledged reports and scientific papers as well as producing mockups for websites and other design materials.
Years from now, it’s possible that Generative AI will produce better final drafts than professional writers and generate better art and design elements than professional human artists and graphic designers. More advanced Generative AI may also be able to entire computer applications, video games, movies and other complex elements with little or no human supervision. It’s likely that voice assistants like Siri and Alexa will be capable of handling far more sophisticated functions—such as planning a vacation or buying birthday gifts for a family member.
If Generative AI can match or exceed human performance for many tasks, the nature of work—and roles within organizations—will change dramatically. Some roles and job functions will disappear, while new roles will likely replace them. However, this displacement could rival or even exceed past events, such as the Industrial Revolution. In addition, society will have to sort through a variety of issues—including ethical and legal concerns—in order to fully benefit from the technology.
In the end, one thing is certain: Generative AI is here to stay. As neural nets and GPUs continue to advance and AI algorithms become more refined, the ability of machines to perform human tasks will increase. Whether Generative AI will lead to singularity, the hypothetical point where AI exceeds human intelligence, remains to be seen. However, it’s entirely clear that generative AI is poised to change the nature of business—and the world around us.
For more information, also see: The AI Market: An Overview