Skip to main content
Top

22-01-2024 | Automotive Electronics + Software | In the Spotlight | Article

Why Not All AI is the Same

Author: Christiane Köllner

4 min reading time

Activate our intelligent search to find suitable subject content or patents.

search-config
print
PRINT
insite
SEARCH
loading …

ChatGPT, GenAI and the power of algorithms are making headlines and fueling the AI debate. But what is behind these technologies? An overview. 

ChatGPT from US software company OpenAI has triggered a worldwide hype around artificial intelligence (AI). When the AI text robot was launched in the US a year ago, it took just five days to reach the one million user mark. Since then, many people and companies have gained practical experience with AI applications. Amazon is also joining the chatbot trend. The cloud division AWS recently introduced an AI chatbot for companies, a program called Q, which is intended for business customers. The AI chatbot can, for example, create summaries of documents or drafts of texts. Amazon is thus competing with similar products from Microsoft and Google.

AI chatbots such as ChatGPT can hold human-like conversations with users in natural language and formulate texts at the linguistic level of a human. The principle behind this: They are trained in advance using extensive text datasets to learn grammar, context and semantics. When prompted, they predict the next word or sentence based on the patterns they have learned. The "GPT" in ChatGPT stands for "Generative Pre-trained Transformer". "A 'Transformer' here represents an architecture for machine learning. This is already pre-trained so that, unlike many chatbots, training by the user is no longer necessary. In addition, GPT is creative, hence the adjective 'generative'," explains Springer author Ralf T. Kreutzer in the German book chapter "What is artificial intelligence and how can it be used?". ChatGPT is therefore a major step forward in AI development - and a current example of generative AI, also known as "GenAI".

Generative versus Predictive AI

A basic distinction is made between two types of AI: Generative AI and "traditional", predictive AI. GenAI is a subgroup of deep learning, which in turn is a type of machine learning. GenAI is concerned with generating new data by learning from large data sets and recognizing patterns in them - the capabilities include text, image and sound. GenAI works with unstructured data, is open-ended and creative. Examples of GenAI include the aforementioned ChatGPT (text), StyleGAN from Nvidia (image) and NSynth from Google (sound).

"Traditional" AI or predictive AI, on the other hand, is concerned with solving specific tasks by making predictions based on previously analyzed data sets and predefined rules. This type of AI requires structured data, is goal-oriented and specific. Examples of "traditional" AI include route optimization, sales forecasting and sentiment detection.

Creative AI

AI applications are based on the knowledge gained from neural networks. Machine learning and deep learning concepts are used. Commonly used neural networks include the Generative Adversarial Network (GAN). A GAN is a system consisting of two coupled neural networks. It is used, for example, to generate deceptively real images or videos, so-called deep fakes, as Springer author Patrick Krauss explains in the German book chapter Creativity: Generative Artificial Intelligence.

It consists of a generator network and a discriminator network. "The generator always generates new candidate images or videos, while the discriminator simultaneously attempts to distinguish real images and videos from artificially generated ones. Over the course of training, both networks iteratively improve in their respective tasks," continues Krauss. The deep fakes generated in this way can then usually no longer be distinguished from real images and videos.

GenAI uses various machine learning methods "to recognize what certain data structures look like and how they can be generated," explains author Kreutzer. "The algorithms are trained by analyzing a large amount of data and trying to find patterns in this data. Once the model is trained, it can be used to generate 'new', as yet unseen data," says the author.

Components of AI

According to the Springer authors led by Akshay Kulkarni, this results in the following overview of the performance components of AI, which they outline in the book chapter "Introduction to Generative AI": According to this, GenAI is a component of artificial intelligence and can be differentiated from machine learning and deep learning as follows.

Artificial intelligence (AI): This is the broader discipline of machine learning to perform tasks that would normally require human intelligence. Authors such as Markus H. Dahm and Valentin Zehnder differentiate between strong and weak AI in the German book chapter Fundamentals of AI. The differentiation refers to the degree of intelligence and autonomy that an AI exhibits. Weak AI (also narrow AI = Artificial Narrow Intelligence) refers to AIs that have been trained for a specific task and only have limited capabilities, while strong AI (also general AI) refers to systems that have general intelligence and are able to solve a variety of tasks.

Machine Learning (ML): As a branch of AI, ML comprises algorithms that enable computers to learn from data without being explicitly programmed to do so. According to Springer author Kreutzer, there are different types of learning in ML: Supervised Learning, Unsupervised Learning, Reinforcement Learning and Self-supervised Learning.

Deep Learning (DL): A special subgroup of machine learning. Deep learning comprises neural networks with three or more layers that can analyze different factors of a data set.

Generative AI: An advanced subset of AI and DL, generative AI focuses on producing new and unique results. It goes beyond the simple analysis of data and creates new creations based on learned patterns.

print
PRINT

Related topics

Background information for this content

Premium Partner