AI stands for “artificial intelligence,” but what exactly does that mean?
In science fiction, AI can think, reason, learn, make judgments, and possess cognitive abilities. In reality, however, the narrow AI products we have are trained to do specific tasks. In the case of today’s generative AI, that specific task can be something like text-based chat. (ChatGPT is still not able to do our laundry.)
In short, “AI” is a marketing term used to describe both new and old technologies that are bundled into software applications. Below, we dive into the different types of AI models, examples of products, and concerns and cautions about AI. We also link to some online tutorials and videos to learn more.
Types of AI
Generative AI (GenAI)
Based on statistical likelihoods found in the provided training data. Uses vast amounts of data (text, images, video, audio) scraped from the internet, chat rooms, and forums.
Large language models (LLMs) fall into the generative AI category, and most of what people mean when they say “AI” in casual conversation refers to the generative AI category. Examples include ChatGPT, Claude, Gemini, and Grok.
Physical AI
Systems that use sensors to decide and act in the physical world, such as robots, vehicles, smart infrastructure, and devices.
These are things like self-driving vehicles, house temperature controls that turn on the AC when it is too hot, or a health watch that monitors how you are sleeping.
Agentic AI
Uses one or multiple AI models that can plan and execute tasks with varying autonomy levels and minimal human supervision.
This might be a robot programmed to navigate over rough terrain, or a software that buys more coffee when the sensor in your smart coffee machine tells it the beans are low. It could even be more complex, such as a program that manages an online business, making digital products and selling them, all with minimal input from a human.
Industrial AI
Used for analytics, quality, maintenance, and optimization in manufacturing, energy, and supply chain logistics. Uses other multiple AI models in a pipeline to gather data, assess it, suggest improvements for efficiency, and implement improvements.
This might look like a factory increasing production after using sensors to collect data (physical AI), then using that data in a generative AI model to find out where slowdowns occur, and finally an agentic AI slowing or speeding a conveyor belt somewhere in the factory to keep everything moving smoothly.
Generative AI examples
Below are some of the more commonly mentioned generative AI software, with the company and CEO that owns that AI product.
- ChatGPT: Owned by OpenAI, Sam Altman
- Claude: Owned by Anthropic, Dario Amodei
- Gemini: Owned by Google, Sundar Pichai
- Grok: Owned by xAI (SpaceX), Elon Musk
- Deepseek: Owned by High-Flyer, Liang Wenfeng
- Perplexity: Owned by Perplexity AI, Inc., Aravind Srinivas
Concerns & cautions
Remember: anytime a service is free, the company makes its money by selling your data.
Regulation
Currently, there are no public-facing federal regulations around the creation or labeling of AI products, but local governments and organizations like professional associations, educational institutions, and scientific organizations either have or are in the process of developing AI-related policies and frameworks.
Copyright
Commercially available AI products are often trained on copyrighted material that may have been used without permission from the original creator.
Depending on the product, the content that is “generated” through an AI product is an outcome of data gathered or extracted from sources outside its own infrastructure, including but not limited to information found through searches on the internet, individual blogs, community forums like Reddit, and news sources that may or may not be behind a paywall.
Training models
This data is put through a process called “training,” which involves comparative matching between pieces of data, and these comparisons are performed by human beings who verify the outcome and scope of the algorithms the product’s owning corporation determines.
The global scale that these products reach means the human beings doing that work of comparison are likely to be working under high-pressure, time-limited contracts with little pay.
Data centers, electricity & water
The computational power needed to run algorithms, perform training, and respond to input inquiries (the prompt you put into a chatbot, for example) requires major amounts of electricity, water, and infrastructure through large data center campuses.
Inaccuracies & biases
The content that comes out of AI products may be inaccurate, biased, discriminatory, and/or harmful. AI seems like it can answer any question, but it’s important to remember that AI reproduces already existing information that has been fed to it by human programmers and creators. This means that AI content still has the same biases that human-created content has.
Learn more with online tutorials & videos
LinkedIn Learning is an online learning platform with video tutorials to learn business, software, technology, and creative skills, and you have access to it with your library card. Sign into the platform to learn more about AI with tutorials like these:
And here are two short, easy to understand YouTube videos: