What is Generative AI?

Defining generative AI

On 30 November 2022, the research organization, OpenAI, took the world by storm. They launched ChatGPT, a generative AI-powered chatbot that can produce new content by deciphering patterns in existing data and converse with users in an uncannily human way. Soon after, the hype around this new tech took over the internet. Some saw it as the dawn of the “Fourth Industrial Revolution” (Pam Baker, ChatGPT For Dummies, 2023). Others feared it would put people in many industries out of a job. While neither seems to have happened yet, ChatGPT represents a major step in the development of artificial intelligence. As Divit Gupta and Anushree Srivastava explain,

In addition to being able to create new content, ChatGPT has even passed the Turing test, an assessment of how convincingly AI can emulate human reason.  A number of other generative AI tools have also emerged that not only generate text as ChatGPT does but can also produce images and more. These include GitHub Copilot, Dall-E, and Midjourney to name a few.

What exactly sets generative AI apart from previous iterations of automation technology? This study guide is devoted to answering this question in more depth, as well as exploring generative AI’s capabilities, real-world applications, limitations, and ethical considerations. 

How generative AI works 

When prompted, generative AI can do some pretty amazing things. It can write poetry, create a custom itinerary for your next holiday, produce lines of code, and make animations tailored to your instructions. Yet, it doesn’t do so out of thin air. Rather, the unique and unprecedented capabilities of generative AI are the result of sophisticated technology, which has been under study since as early as the 1960s when deep learning first became a field of research. As Utpal Chakraborty, Soumyadeep Roy, and Sumit Kumar recount, 

Since those early days, much of the progress that has enabled conversation-based generative AI tools like ChatGPT has come from the development of large language models (LLMs) and their counterparts in image generation: generative adversarial networks (GANs). 

ChatGPT can converse with users in such a human-like way because it is trained on vast amounts of content written by real people, all made possible by LLMs. ChatGPT also has an impressive ability to summarize huge volumes of content, pick up on conversational tone and sentiment, and even perform translations. So, let’s lift under the hood and delve into how LLMs have enabled all this.

Large language modules (LLMs)

LLMs are a form of automation technology that is built on transformers, which are a kind of vast, artificial neural network that enables the processing of data in a simultaneous way rather than in a sequential one. In reference to ChatGPT, Baker explains that, 

In other words, LLMs are able to process vast quantities of information — say, in the case of ChatGPT, much of the written content on the internet — instantly and all at once. This capability means that LLM-based chatbots don’t need to be trained manually by sequentially feeding it the data set that you want it to be able to converse about. To recap, transformers lay the foundation for LLMs to function, and LLMs are a subset of generative AI tools that produce text. Other analogous models such as GANs also use transformers to enable the simultaneous processing of visual inputs rather than text-based ones.

Thus, text-based generative AI, which is powered by LLM technology, has the appearance of being all-knowing and can instantly converse on just about any topic within the enormous data set to which it is connected. This has considerably democratized the use of generative AI across many fields because setting up a chatbot, for instance, no longer requires coding experience or training as a bot builder. Still, it also means that generative AI is more like a content aggregator that uses pattern recognition, rather than a sentient tool. Continuing with Baker, 

The distinction between ChatGPT’s uncanny ability to parrot human conversational style vs. actually having a mind of its own is important. We will delve into its implications later.

Special capabilities of generative AI 

It is LLMs and other forms of transformer-based automation that give generative AI its special capabilities. We’ve mentioned some of these so far, but before we can explore the real-world applications of generative AI, let’s review some of its unique features.

Mimicking human conversational style

Generative AI is not only able to instantly analyze enormous swathes of data to produce responses, but it can deliver those responses in an unprecedentedly natural, conversational way. Chatbots like ChatGPT are able to talk to users in a way that takes into account the context of the conversation by building on what’s already been said. Generative AI-powered bots can also identify the mood of the person they are speaking to, and adjust their tone to respond appropriately. What’s more is that they can instantly speak the user's language of choice, right down to regional dialect. In fact, 

And it can do all this right out of the box, unlike previous generations of chatbots where designated bot builders would have to build out individual conversation trees that manually programmed all potential conversation paths one might have with that chatbot. ChatGPT is hugely different from the typical chatbot experience, which often delivers clunky, canned responses that never quite fit a person’s specific question. ChatGPT is designed to give the user the sense they are speaking with a real person. 

Generating new content 

Generative AI has been especially groundbreaking in creative industries because it can create new content and help generate ideas based on the specific prompts given. As Rohit Bhargava and Henry Coutinho-Mason foresee,

You can ask ChatGPT to write a cover letter in the style of Shakespeare, suggest a recipe in the tone of a pirate, or design birthday cards personalized to the specific details of the person you are celebrating. Such tools are great for brainstorming and summarizing information, which can be helpful for enhancing your existing creativity or giving you that extra burst of inspiration. 

Democratizing access to content creation  

Whether used for fun, to help prepare for a job interview, or implemented in a corporate setting, generative AI has enabled a new generation of automation that is easily accessible to a wide range of people and applications. Such is supported by the proponents of generative AI who argue that, 

You can simply go online and access these tools, often for free, and they can assist you with tasks that previously might require technical training in design, programming, and more. Now, virtually anyone can produce graphics and lines of code. Implementing a chatbot in your professional organization — whether to interface with customer queries or to help organize your workflow — now requires no previous technological know-how.

Real-life applications of generative AI

Since generative AI has become accessible to the public, its use cases abound — across nearly every field and industry. Generative AI will make it possible to automate a wide array of mundane tasks, as well as creative pursuits — ranging from fields as diverse as law, computer programming, manufacturing, healthcare, and more. Hype aside, this tech is still at its beginning stages and is not at a point where it is ready to overhaul the entire economy. Below, however, are a few arenas where generative AI is already making a considerable difference in the way things are done in different industries. 

Marketing and creative industries 

Generative AI lends itself well to where creative work intersects with the need to be efficient, cost-effective, and productive. Because it enables a faster turnaround and also helps to generate text, images, and even audio, generative AI as a tool is a natural fit for the marketing sector. As Gupta and Srivastava outline,

While these tools are not quite so sophisticated at the moment, they do prove useful in generating content tailored to a specific brand and creative vision, which can serve as a useful jumping-off point for marketers when writing copy or generating branded banners and templates for a website. 

At this stage, however, these tools still need a sentient being at the other end to take their suggestions and apply them to fit within a given campaign. For example, let’s say a copywriter is working on an ad campaign for selling shoes. To do this, they might want to inject more wordplay into their copy describing the products on sale. They could prompt ChatGPT to generate shoe-related puns. While some of those puns might be corny and unuseful, ChatGPT can serve as a good source of inspiration for adding an extra level of creativity to the project. 

Customer support 

Another sector that has already been transformed by generative AI is customer support. That’s because generative AI has enabled a whole new generation of chatbots that can actually interact with customers in a convincingly conversational way, rather than delivering pre-programmed responses that often miss the mark. As Husain asserts, 

Since they also don’t require technological know-how to implement and manage, customer support teams can connect them to their existing help centers or FAQs pages so the chatbot can pull info from these knowledge sources but deliver it in a more natural, conversational way. 

This has enabled companies to scale their support and offer instant resolutions to common issues, around the clock. For industries like financial services or travel, where customers often have time-sensitive concerns around transactions or managing bookings, this capability can be a game-changer. Plus, by clearing away more repetitive queries off the plates of overworked human support agents, it can make their jobs more specialized and interesting as agents can focus on providing a higher level of service on more complex issues. 

Software engineering 

Aside from its utility in generating conversational responses or coming up with puns to use in your next ad campaign, generative AI can also produce code. Just as you might type in a description of an image you’d like Dall-E to produce or a prompt for a poem you’d like ChatGPT to write, programmers can enter parameters for code. Generative AI tools like GitHub Copilot can then create lines of code accordingly. As Bhargava and Coutinho-Mason explain,

All this is possible because generative AI tools for coding learn from vast amounts of existing code in order to understand common coding styles, strategies, and best practices. As a result, these tools can help programmers save time, produce more consistent code, and focus on the more complex and creative aspects of coding.

Healthcare 

Finally, generative AI has proven useful for certain fields within the healthcare sector such as medical imaging and diagnostics. Using GAN (generative adversarial network) models, generative AI can render high-quality images from lower-quality scans. As Gupta and Srivastava point out,

This can help increase the accuracy and precision of diagnostics, enabling earlier detection of medical conditions. Generative AI can even help with drug development, as its predictive powers position it to suggest potential molecular structures that would interact most effectively with biological targets. This could potentially speed up the discovery of new and effective medications and treatments. 

Risks and limitations of generative AI

Despite the groundbreaking capabilities of generative AI that we’ve discussed thus far in this study guide, it’s important to exercise caution when using these tools. In actuality, we are only beginning to scratch the surface of their many applications, as well as the long-term impacts they’ll have on the ways we live and work. Below are some of the primary limitations and risks to keep in mind when it comes to generative AI. 

Hallucinations 

Generative AI can often provide inaccurate answers, known as hallucinations. As Baker writes,

ChatGPT has been known for its fair share of confidently providing wrong answers, such as saying that the fastest mammal is a peregrine falcon. It has even been known to provide false references to back up its claims. Therefore, one should be very careful when using these tools for research or when seeking any kind of factual information. 

Ethical concerns 

There are also plenty of ethical concerns to keep in mind when it comes to using generative AI tools. Among them are how these bots can perpetuate biases, plagiarize existing content, and lead to significant job losses. 

Biases and deepfakes 

Because generative AI is trained on an amalgamation of data and content that was produced by real people, it is ultimately repeating human patterns and tendencies — including our racial and gender biases and prejudices when responding to questions. Thus, if we are not critically engaging with these tools, then we may end up accepting the information they provide as fact. This can make us susceptible to politically-motivated misinformation, and serve to reinforce racism, sexism, and other forms of bias and prejudice. 
Generative AI programs can also produce fake images known as “deepfakes,” which can have pernicious consequences, as Chakraborty, Roy, and Kumar explain,

This, in turn, can contribute to the rise of misinformation propagated on the internet and contribute to the prevalence of fraud and propaganda. To explore the impacts of deepfakes in more depth, check out Graham Meikle’s Deepfakes (2022).

Plagiarism and copyright infringement 

It is also important to exercise caution when it comes to using text or other content produced by generative AI in your own writing, reports, or other pursuits. As Chakraborty, Roy, and Kumar point out,

As of now, it can be very difficult to regulate or parse through what is original and what is duplicated. Moreover,

In aggregating and parroting existing content in its responses, there is no guarantee that a generative AI-powered bot won’t end up plagiarizing or infringing on copyrights. It’s important to exercise caution when using the exact information or content it produces. Beyond the legal implications of infringing on copyright laws, there are also ethical concerns involved in the fact that AI can enable users to steal artwork by copying the style of the artist for profit. We cover these issues in more depth in our study guide on AI ethics. 

Job loss

One of the biggest concerns associated with generative AI is that these tools will enable automation across many sectors of the economy, ultimately replacing humans and leading to widespread unemployment. As Martin Musiol observes, 

While generative AI is not advanced enough to take over arenas that require critical thinking skills and other forms of human discernment, this could be on the horizon.

Already, entry-level positions in fields like marketing and graphic design, as well as customer support agents are becoming increasingly replaced in lieu of increasingly sophisticated forms of automation such as generative AI. In an optimistic view, these tools will take over mundane tasks. That said, they may still push out creative positions like writers, artists, and designers in addition to those repetitive, manual, and mind-numbing forms of labor. Only time will tell. 

The role of technological advancements in changing the labor market is not new, however. In fact, Karl Marx was writing extensively about these questions back in the 19th century in light of innovations that were emerging at that time. To read more about this, check out our study guides on Marxism and industrialization.

Closing thoughts  

While generative AI represents a major breakthrough in artificial intelligence, it is important to remember that bots like ChatGPT or Dall-E are just tools. They are far from being sentient and act more as aggregators that parrot existing patterns of content produced by humans. They can aid in creative processes from writing to generating images, code, or even the development of drugs, but they are not yet the conscious beings you might have imagined in sci-fi movies on artificial intelligence. It’s important to exercise caution when engaging with a generative AI-powered bot, as they don’t always produce accurate information, can perpetuate biases, and even plagiarize from existing content. 

As of yet, they are no substitute for human discernment, reasoning, and critical thinking. That said, AI bots have already been integrated into many fields, ranging from marketing to programming to customer support and healthcare. Moreover, it’s unlikely that generative AI is merely a passing trend. These tools are forecasted to become more accurate, sophisticated, and applicable to realms such as education, law, healthcare, finance, and creative industries. Despite its limitations at this current juncture, no doubt artificial intelligence and other forms of automation will continue to transform the way that we live and work over the years to come.

Further reading on Perlego 

Fully Automated Luxury Communism: A Manifesto (2019) by Aaron Bastani

The Eye of the Master: A Social History of Artificial Intelligence (2023) by Matteo Pasquinelli

Artificial Intelligence: An Illustrated History: From Medieval Robots to Neural Networks (2019) by Clifford A. Pickover

The Age of AI: Artificial Intelligence and the Future of Humanity (2020) by Jason Thacker

Bibliography 

Baker, P. (2023) ChatGPT for Dummies. For Dummies. Available at: https://www.perlego.com/book/4161048/chatgpt-for-dummies 

Bhargava, R. and Coutinho-Mason, H. (2023) The Future Normal: How We Will Live, Work and Thrive in the Next Decade. Ideapress Publishing. Available at: https://www.perlego.com/book/3843959/the-future-normal-how-we-will-live-work-and-thrive-in-the-next-decade 

Chakraborty, U., Roy, S., and Kumar, S. (2023) Rise of Generative AI and ChatGPT: Understand how Generative AI and ChatGPT are transforming and reshaping the business world (English Edition). BPB Publications. Available at: https://www.perlego.com/book/3884409/rise-of-generative-ai-and-chatgpt-understand-how-generative-ai-and-chatgpt-are-transforming-and-reshaping-the-business-world-english-edition 

Gupta, D. and Srivastava, A. (2023) The Potential of Generative AI: Transforming technology, business and art through innovative AI applications (English Edition). BPB Publications. Available at: https://www.perlego.com/book/4334812/the-potential-of-generative-ai-transforming-technology-business-and-art-through-innovative-ai-applications-english-edition

Husain, A. (2023) Generative AI For Leaders. AM Press. Available at: https://www.perlego.com/book/4341797/generative-ai-for-leaders 

Musiol, M. (2023) Generative AI: Navigating the Course to the Artificial General Intelligence Future. Wiley. Available at: https://www.perlego.com/book/4321393/generative-ai-navigating-the-course-to-the-artificial-general-intelligence-future