Generative AI Mixes Promise With Pitfalls

When used wisely, the tool can be powerful, effective and time saving

Daniel Reichman headshot
Daniel Reichman, Ph.D., is the CEO and chief scientist at Ai-RGUS.

Artificial intelligence (AI) and, more specifically, generative AI (gen-AI) is in the news on an almost daily basis. The technology captivated society’s collective attention approximately 18 months ago with the release of ChatGPT, which garnered 100 million users in just two months, a feat that took social media giant Facebook two and a half years to accomplish.

Gen-AI creates new content automatically using AI (hence the modifier “generative”). The applicable uses range from jobs as simple as generating text and images to more complex tasks like creating music, videos and code. A gen-AI model that can mimic human language, art, mannerisms and thought processes is a powerful tool. Ultimately, this unleashes a new realm that can enable businesses to transform their work.

To build a Gen-AI system, the model needs to review enormous amounts of data. Thanks to the Internet, there is a seemingly infinite number of data sources (which is both a blessing and a curse). The AI must be fed this data so it can learn to identify patterns in and structures of the source material. In this way, it learns both about language and the world around us (e.g., what is a correctly constructed sentence or image), as well as how to converse.

Gen-AI uses statistical methods to learn how to predict (or generate) a response to a prompt. For example, a user can ask a gen-AI system to write creatively about a given topic. This would be an example of a large language model. A user could also use a diffusion model of gen-AI to generate an image based on a piece of text. Other models of Gen-AI can predict the word that comes next in a block of text or fill in a missing part of an image. Because it has reviewed a tremendous amount of data, it can use the information it has seen to answer questions with relevant content and correct-sounding prose and context.

When new technology such as this is rolled out, it is an exciting time for experimentation, and with that come both success stories and tales of failure. Laborious tasks that would take employees hundreds of hours to complete can now be done in minutes – or even seconds. For example, people have used this tool as a way to get summaries of a topic that would otherwise require extensive research. It can automatically generate documentation and how-to guides, and provide suggestions based on an explanation of a specific situation or project. It has the ability to generate photo-realistic images from just a short textual prompt.

The applications of gen-AI include creating more engaging marketing campaigns, using predictive analytics to review video footage, and developing advanced products based on the current research in a given field. It has spurred a wave of new companies that can customize and tailor the tool to fit specific use cases.

Some of the failures, meanwhile, stem from the way in which gen-AI learns about the world. If the information it has seen on a topic is sparse, wrong or outdated, its knowledge will be limited, biased or corrupted. Another issue is that it learns how to complete sentences based on what is most likely, which, contrary to claims of being “predictive,” can lead to unpredictable outcomes. One case that led to difficulty for its user was when the gen-AI engine cited sources that were not real. The format and placement of the citations were correct because the gen-AI system had seen many such examples, but the “sources” were entirely made up. This issue in automatic content production is referred to as “hallucination,” and avoiding or mitigating it is an active topic of research. To say the least, it is highly advisable to verify the content that gen-AI produces before using it.

Finally, an important aspect of utilizing gen-AI models is the quality of the training material. As more people and businesses publish AI-created content online, the pool of information can be polluted. If someone does not check for accuracy and uploads to the Internet incorrect information (that is stated as fact) from an AI model, this can lead to gen-AI systems regurgitating incorrect information repeatedly. At some point, this would evolve into a feedback loop of AI being trained on AI-generated material, further taking the “human” out of the process with each pass. With all of this in mind, the most successful way to use gen-AI, to date, is as a starting point and not as a final product.

Notwithstanding these challenges, gen-AI remains a powerful tool that has major potential to relieve people of repetitive and tedious work. One way that gen-AI is being deployed in the security industry, for example, is through video search and querying. To achieve useful results with video, like identifying a problem in real time or putting together a timeline of events after an incident, the video footage needs to be reviewed frame-by-frame. This is an immense undertaking for a human operator. Gen-AI opens the possibility of an interface in which a user can describe in plain English what content is of interest and then have AI produce the images that are related to the query. This can save time, resources and money. In the future, a conversational engine will likely be a standard feature of user interfaces, allowing someone to retrieve specific information with ease.