top of page
Modern Workspace


Explaining the Concept of Generative Artificial Intelligence (Generative AI)

Unpacking the Headlines


A quick scan of today's headlines might give the impression that generative artificial intelligence (AI) is omnipresent. Notably, some of these headlines might even be crafted by generative AI, such as OpenAI’s ChatGPT—an advanced chatbot demonstrating remarkable human-like text generation.


Defining Generative AI


But what exactly is meant by "generative AI"? In the pre-generative AI era,

discussions about AI primarily revolved around machine-learning models

making predictions based on data. In contrast, generative AI focuses on creating

new data rather than predicting from existing datasets. It involves training a model

to generate objects resembling the data it was trained on.


The Evolution of AI Technology


While the buzz around ChatGPT and its counterparts might suggest a recent breakthrough, the technology itself isn't entirely novel. These advanced machine-learning models build upon over 50 years of research and computational progress.


Early Examples: Markov Chains


A rudimentary form of generative AI, exemplified by simpler models like Markov chains, dates back to early statistical methods introduced by Russian mathematician Andrey Markov in 1906. These models were initially used for tasks like next-word prediction, akin to today's email autocomplete function.


Complexity and Scale: ChatGPT's Foundation


The underlying models of ChatGPT, similar to Markov models, generate text based

on learned patterns. However, the key distinction lies in the enormity of these

models. ChatGPT is vastly more extensive and complex, boasting billions of

parameters and training on extensive datasets, including much of the publicly

available internet text.


Major Advances: GANs, Diffusion Models, and Transformers


In the past decade, generative AI has seen significant advancements. Generative adversarial networks (GANs), introduced in 2014, utilize two models—one to generate an output and another to discriminate real data from generated data. Diffusion models, emerging in 2015, iteratively refine output to generate new data samples resembling the training dataset. Transformers, introduced by Google in 2017, encode words in a corpus and generate attention maps, enabling context-aware text generation.


Applications of Generative AI


Generative AI opens up a vast array of applications. For example, it can be employed to create synthetic image data for training computer vision models or design novel protein and crystal structures.


Challenges and Concerns


While generative AI exhibits incredible capabilities, it is not a one-size-fits-all solution. For structured data prediction tasks, traditional machine-learning methods may outperform generative AI models. Additionally, the implementation of generative AI in call centers raises concerns about worker displacement. The technology can also inherit biases from training data, potentially propagating hate speech and false information.


Positive Outlook: Empowering Creativity and Changing Economics


Despite challenges, generative AI has positive potential. It can empower artists by assisting in creative content creation that might be otherwise challenging. Furthermore, it has the potential to change the economics in various disciplines.


The Future of Generative AI


Looking ahead, generative AI could find applications in fabrication, moving

beyond image generation to producing plans or blueprints for various objects.

It also holds promise in enhancing the intelligence of AI agents, enabling them to

think and plan more autonomously, akin to the human brain.


In conclusion, while generative AI has its challenges and concerns, its

transformative potential across various domains suggests a promising future for this evolving technology.

machine learning.jpg
Modern Workspace
bottom of page