AI has become a household term—everyone seems to be talking about it, yet few understand its nuances. Some equate AI solely with tools like ChatGPT, treating it as a new phenomenon. Others use AI as a blanket term without distinguishing between its different types and capabilities.
While my intention here isn’t to dive deeply into AI’s history, it’s worth mentioning that AI is at least 80 years old. In fact, the very first chatbot, ELIZA, was developed by MIT researcher Joseph Weizenbaum in the mid-1960s. As Narayanan and Kapoor (2024) point out, ELIZA “did not use machine learning but was a rules-based system” designed to mimic superficial conversations by paraphrasing user input.
Although this seems basic compared to today’s advanced chatbots, it was groundbreaking at the time—similar to how ChatGPT’s Deep Search impresses us today. ELIZA’s impact was so significant that it even led to what’s known as the ‘Eliza Effect'(Narayanan & Kapoor, 2024) where users attribute deeper understanding to AI systems than they actually possess.
AI’s development hasn’t been a straightforward journey; instead, it has experienced cycles of winters (periods of reduced interest and funding) and springs (periods of rapid advancement). Over time, multiple factors gradually led us to today’s sophisticated state of AI, including increases in computing power, the development of powerful GPUs, widespread availability of vast datasets for training, and substantial financial investment.
However, while these factors collectively contributed to AI’s advancement, Narayanan and Kapoor (2024) remind us to avoid the common pitfall of viewing AI as a single, unified technology. AI, in reality, is an umbrella term encompassing several distinct types, notably generative AI, predictive AI, and content moderation AI—each with unique purposes, functions, and implications.
AI Types
Let’s quickly unpack three key AI categories that educators need to know: generative AI, predictive AI, and content moderation AI.
1. Generative AI
Let’s start with generative AI! These are tools that, as their name indicates, help you generate (not create) content such as text and images. These tools include popular large language models such as: ChatGPT, Claude, Copilot, Perplexity AI, and Gemini.
These tools are interactive and respond in real-time based on you prompt. Their output might sometimes sound eerily human-like but it is actually built from patterns they’ve learned from vast datasets.
I intentionally emphasize “generate” rather than “create” because these models don’t genuinely ‘create’ in the human sense, they don’t understand meaning or intention. Instead, they predict and assemble text or images based on patterns learned from massive datasets.
2. Predictive AI
Predictive AI involves models that analyze data to forecast outcomes or behaviors. In their book, Narayanan and Kapoor (2024) clearly illustrate predictive AI’s purpose as making predictions about the future to inform present-day decisions.
For example, predictive AI may try to answer questions such as, “How many crimes will occur tomorrow in this area?” or “How well will this candidate perform if hired for this job?” However, they also caution, “It’s hard to predict the future, and AI doesn’t change this fact.”
The authors also emphasize that while predictive AI can identify broad statistical patterns, it’s often portrayed as far more accurate and reliable than it actually is, especially when making significant decisions affecting people’s lives and careers.
3. Content moderation
The third type, content moderation AI, is widely used on social media to detect problematic content. Narayanan and Kapoor (2024) highlight two methods: fingerprint matching, which flags known harmful content (e.g., child abuse imagery), and machine learning, which attempts to detect new or evolving threats.
However, as language continually changes—such as users adopting “algospeak” (e.g., “unalive” for “dead”)—these systems can easily miss problematic posts, emphasizing the critical role of human oversight in content moderation.
Final thoughts
If you want to explore these AI types in greater depth, I highly recommend reading Narayanan and Kapoor’s book, AI Snake Oil. It offers valuable insights, real-world examples, and a thorough discussion of the limitations of each AI category.
Next time you mention AI, make an effort to clarify what you really mean—if you’re referring to ChatGPT or similar tools that produce human-like language, say “generative AI,” or simply “GenAI.” I understand these distinctions might be challenging to communicate clearly to a general audience, but as educators, we have a responsibility to know and use precise terminology with our students. Remember, our role is not just teaching with AI; it’s also teaching about AI.
References
Narayanan, A., & Kapoor, S. (2024). AI snake oil: What artificial intelligence can do, what it can’t, and how to tell the difference. Princeton University Press.
