Generative AI

Introduction

On a rather unceremonious day in June, me and some of my friends were having prosaic musings of our lives when one of my friends shared her experience revolving around Dall-E. She narrated how she had been a part of a university project and how her teammates had come up with a handiwork that required the prowess of generative AI at its core. This one instance piqued my curiosity around the same. Surprisingly, I’m not the only one, many people have been pondering the neologism.

In case you have been living under a rock, The graph below depicts the number of times “Generative AI” was searched on Google.

Gen-AI has been reaching the zenith every day, and its popularity has surged swiftly to a new stardom and has found its home in equivocal debacles in every other room. I am pretty sure innumerable such examples in the real world are enough to make one question the whats, the hows, and the whys of the same.
Time to open the can of worms- What is gen AI?

The Pandora’s Box

But before we get to the essence of our topic, let us gather a bit of context here.
We need to get a gist of a few buzzwords, namely: Artificial intelligence, machine learning, and deep learning, how they are interconnected, and what is the fine line between them.

I used to think that the terms aligned themselves like this:

To my surprise, it actually is like this:

And where does Generative AI lie?

Right there at the core of deep learning.

Artificial Intelligence in simple terms means to make machines that can act like humans. The name states the obvious:

Artificial: An entity that has been created by humans and

Intelligence: The ability to think and apply.

Artificial intelligence is a discipline of development of computer systems to make them able to perform a task that requires human intelligence.

It is an umbrella term for computer systems to mimic human cognition and perform tasks.

One of the main differences between humans and computers is that humans learn from past experiences, but computers or machines need to be told what to do. Computers work on logic not common sense. That means if we want them to do something, we have to provide them with detailed, step-by-step instructions on exactly what to do. That’s where Machine Learning comes in. Machine learning Concept consists of getting computers to learn from experiences-past data. Machine Learning is a sub-field of AI that uses algorithms trained on data and patterns to develop models that can perform complex tasks.

So where do they contrast? Artificial intelligence takes on this monumental idea of making systems that resemble human intelligence. Human cognition is a pretty heavy term: it incorporates the entire human thought process from thought to experience to senses. Machine learning is just (not really that minuscule) about training the machines so that they can perform tasks. It doesn’t have its lofty visions of finding alternatives to human conscience.

Machine learning majorly relies on two types of data: Supervised and Unsupervised. Supervised labels are about predictions. They take examples and information from the past and then predict the future outputs based on the dataset provided. Unsupervised labels on the other hand are about discovery. It takes raw and random data and discovers whatever naturally forms. It understands the relationships between datasets. A semi-supervised label is a label that has features of both supervised and unsupervised data.

And this leads us to deep learning, a part of the broader domain of machine learning which is based on artificial neural networks. Sounds familiar? Yes, our biological brain. Imagine them to be just like neurons in your brain. It has several interconnected nodes, and these neurons are used to process data and make predictions based on semi-supervised labels given to them.

So, how does Generative AI find itself in the vast mural of artificial intelligence?

In simple terms, Generative AI generates content; the content here ranges from images to text, data etc. Specifically, a data instance which is based on learning probability distribution of existing data. Based on Artificial Neural Networks it takes a small amount of labelled data and a large amount of unlabeled data to generate content.

Differing from its counterpart, Discriminative AI, where AI is used to differentiate between content. Where Generative AI is used to generate content, discriminative AI is suited for tasks which require classification.

There are ways to distinguish between what is generative AI and what is not. When the generated content is in the form of Number, discrete, Class, or probability then it is not generative AI but if the output is in the form of Natural Language, Image, or Audio then it is generative AI.
I am pretty sure you must have used generative AI to perform your day-to-day tasks.
Chat-GPT The G in GPT stands for generative. Namely Generative pre-trained transformer.

Dall-E for example is very famous as a Bing image creator.

“A sea otter with a pearl earring” by Johannes Vermeer.

And GitHub copilot is popularly used for code suggestions.

The generated content gets exponentially better if the prompt given to it is appropriate.

The how

The functioning of a generative AI model largely depends on transformers and neural networks. Transformers brought a revolution in natural language processing. A transformer consists of an encoder and a decoder. An encoder is used to encode the input given which is passed to the decoder which is used to decode the representation and carry out the complex task.

It relies on neural networks to identify the pattern and structure within the existing data to generate new content.

This moment in time

But again, why a sudden appalling rise in its use, why is generative AI such a hot topic now? The plausible reasons could be:

Computing power:

According to James Currier, this recent rise is largely parallel to the advances made in algorithms. They have more processing power, speed, bandwidth, and speed of computation. “It didn’t change suddenly, it just changed gradually until the quality of its generation got to where it was meaningful for us,” says Currier.

Accessibility:

Generative AI is more easily accessible now. Recent advancements surround us in every corner of life. There is no escape.

Talks:

As it becomes more accessible, with new technology knocking on our doors now and then, the topic of generative AI can’t steer away from ambivalent discussions. It has slowly seeped into our lives.

The Catch

The convenience of generative AI lies in the fact that we cast any prompt and get content with ease. It is a no-brainer, just a few keyboards tap, and you are done. But just like anything, generative AI also has its own Achilles heels. The chink in the armor is its inaccuracy and lack of transparency. The expected output can easily deviate just because of the nuances in the given prompt. Also, there is a lack of transparency surrounding the same. What do we mean by that? When the internal workings of the model are either not known by the user or are not interpretable by humans then it is said to lack transparency. It is hard to decipher and hard to predict. Another disadvantage is that these models are hard to train and require large expenditures making it difficult to have an establishment in small businesses.

Conclusion

One thing is evident, Generative AI is going to change the world and our lives.

There is great inquisitiveness about what is to come in the future. It looks promising and intriguing. The rise of generative AI will lead to game changer industries and creations which would reduce the labor from our everyday lives and will pave room for more other possibilities for the mind to occupy over. At the same time, there are pitfalls with these advancements. Unemployment and loss of privacy might become pressing issues.

As the advancements progress, we have to be careful in mitigating these risks.

We are at a tipping point right now. Right now we are at that point in time when the world before and the world after would be vastly different. The changes in the future are going to be rapid and extremely dynamic. From economic surges, accessibility, data that we give, and the explanations given to the system to the demands of society; the future of AI lies in our hands.

“We need to insist that the humans will take those decisions who can be held accountable and not some machine who doesn’t have a conscience.”