What is generative AI? Explanations, types of AI models, applications and impact on work

By: Daan van Beek MSc
Business Partner

Table of Contents

Generative AI | The perfect copycat | Passionned Group

Generative AI creates neat imitations of examples

Generative AI is the field that generates customized content completely independently, based on artificial intelligence techniques such as neural networks. It mimics examples and generates an imitation. Think texts, images, sounds or video. The most well-known example of generative AI is ChatGPT from the company OpenAI. OpenAI’s system (ChatGPT + DALL-E) has “drained” the entire Internet, analyzed all the words, punctuation, phrases, images, videos and associated context, and learned from these to the point where it can write texts or program code or create illustrations on its own. Generative refers to the ability this type of artificial intelligence has to generate, bring forth or produce something independently. Generative AI is going to have a big impact on our work.

What is generative AI?
Applications
Types of AI models
Impact of generative AI
Artificial Intelligence Book

What is Generative AI?

Generative AI uses both unsupervised and semi-supervised machine learning techniques. So it is not something totally new but it uses a combination of multiple types of algorithms. It allows computers to create new content based on existing content, such as text, audio, video, images and code.

Generative AI is the field that, based on machine learning techniques such as neural networks, generates unique content completely independently based on lots of examples and a question (prompt) from a user.

The goal is to generate – based on a question from a user, or a command from a system – completely original artifacts that are indistinguishable from the real thing. Generative AI makes imitations of the originals, it is a good copycat.

AI models for Generative AI

Two common generative AI models are Generative Adversarial Networks (GANs) and transformer-based models, such as Generative Pre-Trained (GPT) language models.

1. Transformer-based models

GPT and LaMDA, are very powerful neural networks (with lots of parameters) that learn context and meaning by following relationships in sequential data, such as those in a sentence, paragraph or chapter. GPT models are so-called Large Language Models (LLMs). When formulating an answer, they predict the best fitting next word or punctuation mark each time. It seems very simple but under the hood of ChatGPT, for example, is an astronomical amount of solidified, highly compressed knowledge that is also very accessible to a large group of people.

It seems very simple but under the hood is an astronomical amount of solidified, highly compressed knowledge

Compare it to a very high-resolution picture (all the knowledge on the Internet) that you compress to the max (ChatGPT). At first glance the compressed looks very similar to the high-resolution photo but when you zoom in you see the loss of quality. When you start comparing the files – the bits and bytes – of both photos they are completely incomparable. All you will ever get from ChatGPT are outlines and thus are “approximate.” Unless you start (extra) training the model behind ChatGPT for a specific domain.

The GPT model is surprisingly more accurate than machine learning models that rely on structured data. GPT models can not only be fed with anything on the Internet, intranet or in a database, but you can thus also train with domain-specific language such as medical practitioners use, for example. For example, a GPT model can learn from all the medical notes stored in a patient’s record and generate a response that includes a prediction of whether the patient is highly likely to die or return to the hospital for the same ailment within 30 days (Jiang, Liu, Nejatian, et al, 2023). The GPT model in this case is more accurate in those predictions than machine learning models.

2. Generative Adversarial Networks (GANs)

GANs are algorithms that pit two neural networks – a generator and a discriminator – against each other in a zero-sum game. The generator creates fake examples, while the discriminator tries to distinguish between real and fake examples. When the discriminator correctly detects a fake example, it wins points and the generator loses points. GANs are often used to generate images and multimedia.

Now you know the main two types of generative AI models. Many companies are currently engaged in developing these types of models and making them public, through APIs or otherwise. Examples include PaLM, Bard, GPT-4, Sparrow and LaMDA.

Applications of Generative AI

Generative AI has applications in various domains, such as:

image generation
image-to-image translation
text-to-image translation
text-to-speech
audio generation
video-generation
image and video enhancement
synthetic data generation

While generative AI has great potential and can produce impressive results, it also brings major challenges, such as the emergence of deepfakes, copyright, fear of job loss and the difficulty with which generated output can be checked (for accuracy or authenticity).

The impact of generative AI on occupations and industries

Generative AI makes it possible for the intelligent, data-driven organization to not only make decisions automatically but can now respond automatically. With that, we can take artificial intelligence all the way to the end of the cycle. This is another indication that AI is a systems technology and not just a toy of tech companies. Many people’s lives as well as the content of their work will change dramatically in the coming years. Researchers predict, for example, that generative AI in particular can achieve annual productivity increases between 1% and 3%. So that’s cumulative, year on year. It’s a challenge for all of us to find a way around that, writers, musicians, actors and other creative and educational professions included.

Anywhere that involves a creative (generative) process, generative AI is going to play a very important role. Consider professions such as:

Policy officers and managers
Consultants and (financial) advisors
Marketers and advertisers
Sales and account managers
Writers, teachers, musicians, filmmakers, screenwriters and actors
Lawyers, brokers and notaries
SEO specialists and web editors
IT architects, programmers and administrators
Accountants and auditors
Data engineers & data analysts

The list can be made as long as you like. After all, creative processes occur in every profession and industry. Would you also like to know the impact of generative AI on your organization or business? Then contact the generative AI specialists.

More information about Generative AI?

Would you like more information or a casual chat about the applications of generative AI? Then leave your contact details here.

About Passionned Group

Passionned Group is the specialist in conceiving and creating innovative artificial intelligence solutions including generative AI. Our skilled consultants help you to tilt your organization to a data-driven company where AI-first can become the rule. Every other year we organize the Dutch BI & Data Science Award™.

Frequently Asked Questions

What is generative AI?

An AI technology that can create imitations that are barely distinguishable from the examples it learned from.

What are Large Language Models?

These are so-called Generative Pre-Trained language models that have an astronomical number of parameters that enable them to recognize and generate multi-level sequences.

Is Generative AI a threat?

Certainly. Every creative process and industry is going to suffer greatly from it. But, of course, generative AI also offers opportunities if you can deploy it effectively in your organization.

Are the results of generative AI always reliable?

No, this form of AI offers "only" an approximation (like all AI) but in many cases it is sufficient. You will always need to check the results (regularly) by an expert. By further specifying your prompts, you can get better results generated.