How does AI actually work?

Yet for some of us, AI poses more questions than answers. How will it affect us? Are there risks? How do we make it trustworthy?

Before we can answer these complex questions, it helps to get the basics down on AI. Read on for a crash course on artificial intelligence.

First, what exactly is AI?

AI is essentially software that can learn patterns from information. Think language, images, audio, online behavior and more. Using patterns from existing and new data, AI makes predictions to perform tasks that normally require human intelligence – like finding products we’re likely to buy or finishing a sentence in an email.

How does it work?

Take a customer service chatbot, for example. After you type a question, the chatbot uses an algorithm – or a set of rules  – to recognize keywords and identify what kind of help you need. The machine learning model, based on the existing and new information it has, then generates an appropriate response. The chatbot improves over time as it interacts with new customers and receives more data. 

“Think about the algorithm as the program that works with the dataset, and the model is the output that makes the prediction,” explained Mozilla researcher Becca Ricks.

Why are chatbots like ChatGPT sounding more… human?

The latest chatbots use a type of machine learning model called a neural network. Inspired by the structure of the human brain, it’s designed to learn increasingly complex patterns to come up with predictions and recommendations. With chatbots, the model learns language from a large amount of existing and new data, making it really good at sounding how a person might talk. 

Can AI get things wrong?

Absolutely. AI models learn from data, which can be incomplete. ChatGPT, for instance, is a language model trained on data on the internet. That’s why it may have trouble solving simple math problems. 

AI can also produce biased outputs. For example, image recognition trained on a set of images featuring mostly light-skinned people may not be able to recognize individuals with darker skin tones. Algorithms and data come from humans, so AI technologies typically follow biases that exist – like ones based on race, gender and age. 

“They might affect whether or not our friends are seeing what we post,” Becca said. “Or, they might affect whether or not we’re getting resources from our local government.” 

How do we make sure we can trust AI?

The first step is learning about it. From there, we can demand transparency and accountability.

“The more that we all know the way AI systems work, the easier it makes for us to imagine what better looks like,” Becca said. “And it makes it easier for us to design alternatives that benefit society and reflect the values of our communities.”

For a deeper dive on AI, the people who are creating it and stories about how it’s affecting communities, check out the latest season of Mozilla’s IRL Podcast. And if you’re a builder looking to create trustworthy AI solutions, you’re encouraged to apply to Mozilla’s Responsible AI Challenge. Applications close on Thursday, April 20.

AI exists in almost everything we use on the internet, like search engines and our social media feeds. But if you want to reduce the amount of personal information you have out on the web, don’t give away your true email and phone number when signing up for the latest AI apps.
Like any other application, an AI app can expose your information to online trackers, spammers and hackers. Firefox Relay offers email and phone number masks so you can sign up for new accounts anonymously.

The post How does AI actually work? appeared first on The Mozilla Blog.

Original article written by Kristina Bravo >