This AI stuff is moving quickly—perhaps too quickly. But now’s not the time to pine for the rotary phone. Here’s the AI vocabulary you need to stay in the loop. Don’t worry—there’s no math, because I struggled with pre-algebra in seventh grade.
Artificial Intelligence (AI)
A broad concept that refers to machines imitating intelligent human behavior.
Machine Learning
A subset of AI focused on the ability of computers to learn a task without being explicitly programmed. The general idea is to provide a machine with a ton of data and then have it figure out how to achieve a specific goal using that data (e.g. identify images with cats in them).
Neural Network
A type of machine learning algorithm that loosely mimics the way in which the human brain works.
Neural networks consist of layers of nodes (think neurons) connected to each other by links (think synapses). The first layer receives input data and the last layer produces the output. Each layer in between performs additional computations with the received data, but what exactly goes on in these hidden layers is not always clear.
The use of many layers allows for something called deep learning, which can identify complex patterns in large data sets (using millions or billions of parameters) in a way that exceeds the capabilities of the human brain.
Large Language Model
A type of deep learning model—often based on a neural network that uses something called transformer architecture—that has received a ton of press and freaked out humanity. Think ChatGPT.
These models have been trained on a massive amount of text from various sources and have developed the ability to predict the next word in a sentence, the next sentence in a paragraph, and so on. Progress made with such models has moved the field of natural language processing forward much faster than expected.
One of the most striking features of large language models is so-called emergent behavior, where the systems develop unexpected capabilities.
Generative AI
AI that is able to create original content after having been trained on large amounts of such content. A large language model is one such example, being able to create novel text-based material. (Hopefully, the GPT—Generative Pre-trained Transformer—of ChatGPT now makes sense.) The same idea can be applied to models that create images, videos, and computer code.
Clearly, the hope is that generative AI does not lead to degenerate humans.
Hallucination
A phenomenon in which a machine learning model generates inaccurate or even nonsensical output, often as a result of flaws in its training. Of course, regular humans—particularly politicians—can suffer from the same affliction.
And that’s enough AI vocabulary for now, as that’s about all my neural network can handle.
