Artificial intelligence: what is it and what isn't?

What exactly does the term "artificial intelligence" refer to, and what does it not?

In a broad sense, artificial intelligence (AI) is a field of study and a type of technology that focuses on making and using machines that can do things that humans would normally have to be smart enough to do.

AI has already changed many industries and parts of society. For example, chatbots for customer service and better GPS and mapping apps are just two examples. But there are some wrong ideas about AI and how it could be used.

In the following Q&A, Kay Firth-Butterfield, who is the Head of Artificial Intelligence and Machine Learning at the World Economic Forum, talks about the different kinds of AI, important developments and applications in the field of machine learning, and, perhaps most importantly, common misconceptions about AI.

How many different kinds of AI are there?

"AI is made up of different types of machine learning models." These include, but aren't limited to, reinforcement learning, supervised and unsupervised learning, computer vision, natural language processing, and deep learning.

"All machine learning models make and improve statistical predictions, but they use and understand data in different ways. For example, ChatGPT is an AI-powered chatbot that can guess what the most likely next word in a sentence will be. ChatGPT can make paragraphs that make sense because it makes a lot of predictions that are mostly right.

What is it that most people get wrong about AI?

"AI is not intelligence; it's prediction. Large language models have made it easier for machines to accurately predict and carry out what we want them to do. But it would be wrong to compare this to how smart people are.

"This is clear when you look at machine learning systems, which can still only do one thing very well most of the time. This doesn't make sense and isn't like the way humans think, which makes it easy to do more than one thing at once. Information from one source can be used in many different ways by people. In other words, our intelligence can be passed on, but machines can't."

Where do you think AI will be most useful?

"AI has a huge amount of potential to do good in many areas, such as education, health care, and fighting climate change. For example, FireAId is an AI-powered computer system that uses wildfire risk maps and seasonal factors to predict how likely it is that a forest fire will happen. It also looks at the risk and severity of wildfires to help decide where to put resources.

"AI is also being used to improve health care by making prevention, diagnosis, and treatment more personalized and effective. Healthcare costs are also going down because of better ways to do things. Also, AI is going to change care for the elderly in a big way, hopefully for the better.

Where do you think the possible effects of AI have been overstated?

"Most of the time, people overestimate AI's potential because they don't understand what it can actually do. There are still a lot of AI-powered machines that hallucinate all the time, which means they make a lot of mistakes. So it's unlikely that this kind of artificial intelligence will replace human intelligence.

"Another thing that makes it hard to use AI is that it gets its information from unreliable sources. The vast majority of data is made by people in North America and Europe, so AI systems tend to reflect their view of the world. ChatGPT, for example, gets most of its written words from these places. Nearly 3 billion people still don't have regular access to the internet and haven't made any data themselves.

What are the most dangerous things about AI?

"Systems that use AI are very new. Because of this, businesses and people need to be careful before using them. Users should always make sure that an AI system has been built and designed in a safe way and has been well tested. Think about other products. A car company would never put out a new car without testing it thoroughly first.

"The risk of using AI systems that haven't been tested and aren't well made doesn't only hurt brand value and reputation; it also puts users at risk of being sued. In the US, for example, the government has made it clear that companies will be held responsible if they use AI-powered hiring tools that are biased.

"There are also major concerns about the future of AI and advanced computer systems, which use a lot of energy to build and run." Already, the carbon footprint of the whole information and communications technology ecosystem is the same as the fuel emissions of the aviation industry.

How can we make sure that AI is built in a responsible way?

"People should think first and foremost about whether or not AI is the best way to solve a problem or make a system better." If AI is the right choice, the system should be built carefully and tested well before it is made available to the public.

"Users should also know about the laws that apply to them, and the public and private sectors should work together to make sure that AI is used in a safe way."

"Finally, users should take advantage of the different tools and resources that have been made to help bring about responsible AI."