Ever since OpenAI released the newest version of ChatGPT in late 2022, AI jargon has flooded the internet. This whole new vocabulary has caused a lot of confusion. But we can’t afford to stay confused for much longer as more businesses incorporate AI into their offerings for employees and consumers alike.
Here, we’ll lay out what the different terms mean, how they relate, and why understanding generative AI helps clarify what the tool does (and doesn’t) hold for the future.
AI is like an onion
Just like ogres, AI is like an onion. It doesn’t smell, but it has many layers.1 The phrases thrown around today interchangeably are actually subsets of each other. Here are some simple definitions:
- Artificial intelligence (AI): machines that mimic human cognitive functions.
- Machine learning (ML): AI learning from data patterns.
- Machine deep learning: AI learning from unstructured data without the help of humans.
- Neural networks: Enable machine deep learning by making associations through a mechanical memory, like the neural networks within a human brain.
- Large language model (LLM): AI trained on huge amounts of text using neural networks.
- Generative AI: A type of LLM that “creates” by making highly accurate predictions for what comes next in art, music, writing, and more.
- Artificial general intelligence (AGI): AI with the flexibility of human thought and the processing power of a computer.
Artificial intelligence is the overall term for machines that can perform human-like cognitive acts. Underneath AI, you have machine learning, machine deep learning (made possible through neural networks), large language models, and generative AI (Figure 1).
Figure 1. The AI onion
Generative AI versus AGI
Generative AI mimics human creativity by predicting patterns. For example, ChatGPT uses a complicated framework of word vectors to predict which words come next in a sentence. If you ask ChatGPT to write a poem about cats, it will mimic a poem based on the unstructured data it has processed.
Here’s an excerpt of a poem ChatGPT wrote for me based on the prompt “Please write a poem about a kitten playing with a ball of yarn:”
Amid a world of wonder spun,
A kitten danced, its play begun.
A ball of yarn, a treasure rare,
Unraveled joy in sunlit air.
Oh, playful sprite in whiskered grace,
With every bound, a smile’s embrace.
A dance of innocence and light,
A kitten’s heart takes flight in flight.2
Sounds pretty good except for that last sentence, right? ChatGPT wasn’t inspired to create this poem after watching a kitten play with a ball of yarn on a sunny day. In fact, ChatGPT doesn’t know the meaning of “kitten” or “yarn.” ChatGPT used its artificial neural network (inputs, outputs, and hidden layers) to find and place related words in a poem structure.
AGI is more like Haley Joel Osment’s character from A.I. Artificial Intelligence than my prose-generating software. In theory, an AGI is a machine able to think on par with humans. According to Britannica, AGI generated a lot of interest during the 1950s and 1960s but has made little progress over the last few decades, with modern-day researchers questioning whether it is worth pursuing.
Understanding the intangible brings perspective
I mentioned AI during a recent conversation, and the person I was talking to said they “didn’t want to go there.” I have a feeling what I was thinking about AI (software that can be a nifty tool) was different than what they were thinking (humanoid robots and the end of the world as we know it).
Understanding AI buzzwords brings perspective to this quickly evolving world. Generative AI is possible because people have found ways to create an AI that reads unstructured data, not because people have created a machine capable of independent thought. We are a long way from that.
1Shrek, 2001. If you’ve never heard of this line, please watch the movie. You’ll love it!
2“Please write a poem about a kitten playing with a ball of yarn.” ChatGPT, default 3.5 version, OpenAI, 24 Aug. 2023, https://chat.openai.com/share/b19e80fb-9bf4-4521-803b-ded9b9748f91.