Artificial intelligence (AI) refers to a computer's ability to imitate or duplicate human cognition.
After reading this article you will be able to:
Copy article link
Artificial intelligence (AI) is the ability of a constructed machine, such as a computer, to simulate or duplicate human cognitive tasks. A machine with AI can make calculations, analyze data in order to create predictions, identify various types of signs and symbols, converse with humans, and help execute tasks without manual input.
For example, a traditional automobile responds only to inputs from its human driver: it accelerates when the driver sees the stoplight has turned green and presses the accelerator, and it stops at a stop sign when the driver sees the sign and presses the brake pedal. An automobile with AI may be able to identify stop signs and stoplights on its own, and stop or accelerate without inputs from the driver.
AI has its roots in the very beginnings of computers, and mathematician Alan Turing was one of the first to describe how an artificially intelligent machine could function. All computers built since then are artificially intelligent on some level, as they are able to perform computations that could previously only be done by humans. However, in recent decades computers' abilities, speed, and storage capacity have expanded rapidly. Today the term "AI" refers to the more advanced cognitive tasks that computers can do.
Most AI is built on the analysis of big data sets that contain too much information for any human to analyze on their own in a reasonable time. An AI model is built to identify patterns in those data sets, and then use those patterns to predict future or additional patterns. AI models use probability and statistical analysis in order to do so. Some AI models are good enough at this to mimic human behaviors.
Theoretically, AI may go beyond this one day and be able to "think" original thoughts. Determining when that point is reached is, in some respects, a philosophical question more so than a technical one.
Machine learning is a branch of AI, and it refers to the practice of feeding a program structured or labeled data in order to train the program how to identify that data without human intervention. For example, a machine learning model for finding bottles of ketchup in photos of open refrigerators may start out unable to identify any condiments, let alone ketchup. It is then fed millions of images of ketchup bottles in various refrigerators and is told that each one represents a ketchup bottle. Eventually, it is able to automatically identify ketchup bottles even in photos it has never seen before.
Machine learning relies on the use of predefined processes called algorithms. A machine learning program will "learn" slightly differently depending on how the algorithm is set up.
Machine learning also relies on access to large data sets. A machine learning program shown only three or four photos of ketchup will almost certainly fail to accurately identify ketchup bottles on a consistent basis, or will identify ketchup in photos where it is not present. The more data the model gets, the more accurate it is.
A wide range of software and technological solutions use machine learning today. From security solutions that use machine learning to detect fraud and identify bots, to social media platforms that use machine learning to recommend content or accounts to follow, machine learning has proved to be a hugely useful development tool.
Just as machine learning is a type of AI, deep learning is a type of machine learning. Deep learning models are able to use probabilistic analysis to identify differences in raw data. A deep learning model could potentially learn what a bottle of ketchup is and how to distinguish it from other condiments from photos of open refrigerators alone, without being told what a bottle of ketchup is.
Like other types of machine learning, deep learning requires access to large data sets. Even an advanced deep learning model would probably need to analyze millions of photos of open refrigerators to be able to identify ketchup.
Generative AI is a type of AI model that can create content, including text, images, audio, and video. A generative AI model could, for example, receive a photo of an empty refrigerator and populate it with probable contents, based on photos it has been shown in the past. While the content generated by such a model may be considered "new," it is based on content that the model has been previously fed.
Generative AI tools are increasingly popular. In particular, the large language model (LLM) ChatGPT and image generators DALL-E and Midjourney have captured the public's imagination and the business world's attention. Other popular generative AI tools include Bard, Bing Chat, and Llama.
The use cases for AI are still expanding. These are some of the real-world applications already being explored:
Data leaks: AI services use inputs to further train their models; they are not designed to be secure vaults for data. But many people use LLMs in ways that enhance risk of data exposure, including processing confidential information or closed-source code. Such data may be reproduced or imitated in further responses from these LLMs.
Loss of control over data: Data passes outside one's control once it is uploaded to an LLM, and users may not have visibility into what happens to provided inputs. For instance, if a baker puts their new secret recipe for focaccia into an LLM and asks it to write a compelling description for their bakery's website, they may get back such a description — but the baker's recipe is no longer secret, as it may be exposed to other users of the LLM, or the operators of the LLM.
Regulatory violations: Using external AI tools often introduces some degree of data risk. As a result, AI may put an organization out of compliance with regulatory frameworks like the GDPR.
Hallucinations: Generative AI tools may invent information in order to generate responses — the technical term for this phenomenon is "hallucinations." If businesses incorporate such information uncritically into their content, they may damage their brands.
Over-reliance on AI for decision-making: Because the information provided by AI models is not always reliable, over-use of AI in the decision-making process can lead to decisions that negatively impact a business.
Cloudflare Data Loss Prevention (DLP) can help organizations get a handle on how AI is being used by their employees. DLP can restrict uploads, copying and pasting, and keyboard inputs to stop confidential data from leaving secured environments. Learn more about how DLP works.
Cloudflare for AI allows developers to build and deploy new AI applications on the Cloudflare global network. Learn more about Cloudflare for AI.