Star Wars Day | Explainer: Difference between narrow AI and strong artificial intelligence

Artificial General Intelligence is also known as robust AI, complete AI, and general intelligent action. Some academic sources even reserve the term "strong AI" for computer programmes that exhibit sentience or consciousness. (Photo: Lisa Fotios via Pexels)

Generative artificial intelligence (AI) is a subset of AI that can generate new content, whereas artificial general intelligence (AGI) is the more ambitious aim of creating a machine with human-like intelligence across a broad range of domains.

AGI systems use software to reflect generalized human cognitive skills to solve complex problems. A machine with AGI can accomplish any intellectual task that a human can. While "generative AI" refers to a popular category of machine learning tools that can generate new text, images, videos, or sounds, based on a substantial training dataset.

There is no reason and no way that a human mind can keep up with an artificial intelligence machine by 2035.

- Gray Scott

Evolution of AGI

Ray Kurzweil has used the word "narrow AI" to talk about making systems that do certain "intelligent" things in certain situations. For a narrow AI system, if the environment or behaviour is changed even a little bit, the system usually needs to be reprogrammed or reconfigured by a person to keep its level of intelligence. It is very different from natural, generally intelligent systems like people, which have a wide range of ways to adapt to changes in their goals or circumstances, such as by using "transfer learning" to generalize knowledge from one purpose or context to others.

The idea of AGI has come up as an opposite to "narrow AI", to describe systems that can generalize over a wide range of things. The AGI approach sees "general intelligence" as a property fundamentally different from the ability to do a specific job or solve a particular problem. Instead, it focuses on directly understanding this property and making systems that show it.

How is Generative AI different from AGI?

A generative adversarial network (GAN) is a type of system for machine learning and a popular way to approach generative AI. Ian Goodfellow and his friends came up with the idea for the first time in June 2014. A generative model can use what it has learned from the examples it has been shown, to develop something completely new. So comes the word "generative!"

Large language models (LLMs) are generative artificial intelligence because they create new combinations of text that sound like standard language. Generative AI tries to use generative AI models to look at data and make new and original material based on that data. Generative AI tools use complex algorithms to look at data and develop new and unique insights. As a result, it makes it easier to make decisions and speeds up processes.

What is the objective of AGI?

AGI is also known as robust AI, complete AI, and general intelligent action. Some academic sources, however, reserve the term "strong AI" for computer programmes that exhibit sentience or consciousness. In contrast, weak AI (or narrow AI) can address a particular problem but lacks general cognitive abilities. Some academic sources use the term "weak AI" to refer to programmes that do not experience consciousness or have a mind like humans do.

Generative AI systems generate images, audio, writing samples, and anything that can be constructed using computer-controlled systems, such as 3D printers. (Photo: Sanket Mishra via Pexels)

How can we use Generative AI?

Generative AI models utilize neural networks to generate new and original content to identify patterns and structures within existing data. Training generative AI models using unsupervised or semi-supervised learning is one of the advancements made possible by these models.

Real-world examples

AGI: Currently, deep learning and natural language processing are used in most AI examples, including chess-playing computers, self-driving cars, and understanding speech or text delivered in natural language.

Generative AI: Generative AI systems generate images, audio, writing samples, and anything that can be constructed using computer-controlled systems, such as 3D printers. Discerning systems identify things such as people in photographs, words in speech or handwriting, and what is genuine and what is fake.

Conclusion

It will take several decades for artificial intelligence to attain the artificial general intelligence and superintelligence stages. However, most businesses use AI to boost sales, forecasts, and growth. The future of AI is, without a doubt, a fascinating field of work.

In AI, generative models have a lengthy history. In the 1950s, Hidden Markov Models (HMMs) and Gaussian Mixture Models (GMMs) were the first to be created. These models produced sequential data, including speech and time series. According to Mallick, multiple inputs and multimedia output are the next steps for generative AI. For example, Microsoft has recently introduced Kosmos-1, a multimodal large language model (MLLM). Moreover, Generative AI will streamline and accelerate the content creation process. As a result, it will reduce the entry barrier for aspiring creators and increase the content they can produce.