What is Generative AI? Overview in Simple Language for Non-Experts
Beginner friendly overview of what Generative AI is and what the hype nowadays is all about
News about Large Language Models (LLMs) are hot potatoes in any field. Even in fields that people would view as irrelevant from technology, various ways to incorporate LLMs into their workflow or operations are being actively discussed.
LLMs such as ChatGPT from OpenAI is one example of Generative Artificial Intelligence (AI). What is Generative AI? What is AI in the first place? What do related terms including Machine Learning (ML) and Deep Learning (DL) mean? These different sets of jargon often throw people from non-technical backgrounds or those who have no domain knowledge in these topics at all. However, understanding the basics of Generative AI and its potential applications has now become necessary for anyone to just engage in scholarly conversations about technology. This post will enable you to broadly understand what Generative AI is and give you at least some basic ingredients of knowledge for you to utilize during your conversations on technology and AI.
Artificial Intelligence (AI)
Before we dive into digesting what Generative AI is, we need to first understand what AI means. In my opinion, the term AI is grossly overused in many areas and its definition varies across industries and organizations. Let us take a look at one of those definitions from Google based on its most recently released online learning path on Generative AI.
AI is the theory and development of computer systems able to perform tasks normally requiring human intelligence.
The following another definition from Mckinsey, one of the leading management consulting firms in the U.S.
Artificial intelligence is a machine’s ability to perform the cognitive functions we usually associate with human minds.
A common aspect that these definitions share is some allusion to the system’s ability to perform tasks that require human intelligence, cognition, and minds.
Taking this aspect into consideration, you will be able to quickly list many activities and operations that require human intelligence, cognition, and mind but are actually conducted by some system — prediction of sales for the following, automatic sentence completion and grammar checks, automated summaries of articles and so many more. Yes, AI is embedded everywhere in our lives now.
ML, DL, LLM and Generative AI
Even as a non-technology person, you probably have heard the terms machine learning and deep learning before simply because they are mentioned frequently in any domain that makes data-driven decisions. First take a look at the following diagram that shows the relationship among AI, ML, and DL.
As you can see above, AI is the umbrella term that encompasses all other terms. ML is the second broadest term followed by DL. Do not be intimidated by these seemingly complicated concepts! As for ML, even the simplest statistical model you learned in your undergraduate course falls under it. Linear Regression, for example, is a classic example of a machine learning algorithm that helps us predict numerical values of a target variable. DL uses Artificial Neural Networks (ANN), allowing them to process more complex patterns than traditional ML. Both Generative AI and LLMs are subsets of DL.
Generative AI
Now, let us move onto Generative AI. What is Generative AI?
Generative AI is a type of AI that creates new content based on what it has learned from existing content.
The process of learning from existing content is called training and results in the creation of a statistical model.
When given a prompt, Generative AI uses this statistical model to predict what an expected response might be and this generates new content.
Modalities
Modality is a fancy term for referring to the type of data that the AI is dealing with as inputs and outputs. Are we talking about text, image, audio or tabular data? Note that Generative AI is not just confined to text data. Any AI that generated new “content” counts as Generative AI. Hence, people sometimes make distinctions between different Generative AI by including specific information about their modalities. For instance, Generative AI models that learn patterns in languages through training data and predicts which sequences of text follow after a given text are called Generative Language (AI) models. Similarly, Generative AI models that produce new images using techniques including stable diffusion are called Generative Image (AI) models.
Discriminative v.s. Generative
To be able to tell whether a model falls under the category of Generative AI or not, you need to tell whether that model is discriminative or generative in nature!
As the names suggest, discriminative models are used to classify or predict new unseen data based the knowledge on the relationship between features of the data points and the labels while generative models generate new data that are similar to data it was trained on by understanding the distribution of data and how likely a given example is. For instance, a model that learns how to classify a given image as either a cat or dog would count as a discriminative model while a model that generates new dog images after learning the traits of dog images would be considered a generative model.
Issues
Generative AI are not without issues. Hallucination, referring to words or phrases generated by the model that are often nonsensical or grammatically incorrect, is a classic example of misbehavior of Generative AI. In addition, Generative AI can be extremely sensitive in its behavior and quality of the content generated depending on the user’s prompting. This makes sense — the quality of the input should naturally determine the quality of the output. In ChatGPT, for example, the question or instruction you type in to “prompt” ChatGPT to produce a response would be considered “prompting”. Because of this, many online courses, tutorials, and blog posts are focusing on prompt engineering, effective methods for meticulously leading Generative AI to produce the best quality responses in the most desirable manner possible. The following blog posts are some tutorials that discuss prompt engineering strategies.
Implications for different fields
Generative AI has implications in so many different fields and industries but I briefly touch upon three of them — healthcare and business.
Health
Stanford University’s Human-Centered Artificial Intelligence (HAI) Center released a report titled Generative AI: Perspectives from Stanford HAI in March of 2023, containing interesting insights on the potential applications of Generative AI and its future prospects in several fields including health.
According to this report, Generative AI can be used to assist healthcare providers in seeing through potential issues that can lead to deaths due to medical error, a significant problem in the U.S. as the third leading cause of death. Moreover, Generative AI can make clinical trials more efficient by generating synthetic control patients (e.g., fake patients that are realistic). This would be extremely useful for clinical trials that often suffer from low sample size in randomized control trials for testing new approaches and medicine.
Generative AI is expected to be of help for medical education as well. It can potentially simulate realistic medical conditions or create synthetic patients that can be provided to medical students so that they can learn about those conditions and patients with those conditions. This practice would come in handy especially when that medical condition is rare and therefore difficult for students to experience or witness themselves during their training as interns and residents. Generative AI can also act as additional resources for training and education by giving them practice eliciting signs and symptoms through conversational interaction as physicians who are capable of training them are in constant shortage.
Business
For those who are curious about the business implications of Generative AI, the report titled A new frontier in artificial intelligence — Implications of Generative AI for businesses would catch their interest.
You may view the full report here. This report introduces in-depth case studies of Generative AI applications in business, provides an overview of the competitive landscape of businesses using this technology, and explains various issues surrounding the adoption and commercialization of Generative AI.
In a nutshell, the report argues that Generative AI has the potential to add contextual awareness and human-like decision-making to enterprise workflows, and could radically change how people do business. It poses a question, however, on how much of the potential of Generative AI will be captured by individuals and enterprises to deliver efficiency gains, product improvements, new experiences and operational chance. Furthermore, it mentions the risk of the use of Generative AI in the business domain ranging from privacy and security issues, bias to transparency and equal access.
Conclusion
In this post, I gave you a brief overview of what Generative AI is in simple and laymen expressions for those who have a hard time picking up cues from technology and Generative AI related conversations. If you found this post helpful, consider supporting me by signing up on medium via the following link : )
You will have access to so many useful and interesting articles and posts from not only me but also other authors!
About the Author
Data Scientist. 2nd Year PhD student in Informatics at UC Irvine.
Former research area specialist at the Criminal Justice Administrative Records System (CJARS) economics lab at the University of Michigan, working on statistical report generation, automated data quality review, building data pipelines and data standardization & harmonization. Former Data Science Intern at Spotify. Inc. (NYC).
He loves sports (tennis and surfing nowadays!), working-out, cooking good Asian food, watching k-dramas and making / performing music and most importantly worshiping Jesus Christ, our Lord. Checkout his website!
