Don't Let Students Outsource Critical Thinking to AI

Artificial intelligence (AI) models like ChatGPT and Gemini are profoundly changing the way people interact with technology, information, and each other. Are they also changing the way we think? In a world where more and more aspects of our lives are intermediated by AI, critical thinking is more important than ever. Teachers must make sure that students develop the higher-order thinking skills they need to critically evaluate and use AI content. Above all, we must help our students avoid the temptation of outsourcing their critical thinking to AI.

First Things First: What Exactly IS an LLM?

Let’s start with some definitions. There are many forms of AI out there, but the ones making news today are Large Language Models, or LLMs. These include OpenAI’s ChatGPT, Microsoft’s Copilot (formerly Bing), Google’s Gemini (formerly Bard), and Anthropic’s Claude (among many others!). While capabilities vary between the models, they all allow users to interact and ask questions in plain language and will return a relevant and understandable answer. Some generative AI models are multimodal, processing and creating images, audio and even video in addition to text. Together, all forms of AI that generate novel content in any format are known as “generative AI.”

An LLM combines two powerful forms of AI: 

• Natural Language Processing (NPL) enables the AI to interact with people in plain human language and “understand” language-based information and prompts. This capability allows the LLM to perform a wide range of language-related tasks, such as text generation, summarizing, translation, and conversational response. 
• Deep Learning—a form of Machine Learning (ML)—allows the model to recognize complex patterns in data. In essence, using Deep Learning, the model teaches itself based on the training data it is given. 

Today’s top LLMs have been trained on massive datasets, scraped from millions of sources ranging from textbooks and newspapers to Wikipedia and Reddit. While it would not be correct to say that they have ingested the entire internet, the top models have incorporated a big chunk of it—and have the ability to search and browse for anything new they need to answer a query. 

Interacting with a modern LLM can seem almost magical. Ask it to rewrite the Bill of Rights in the style of Dr. Seuss, and you’ll have a relevant rhyme in seconds. It can summarize a difficult article, write a bedtime story for your child featuring all your pets, provide Dear Abby-style advice for any problem you ask, or draft a letter of complaint to your credit card company. OpenAI’s latest model can even access the camera on your phone (when invited) and critique your back-to-school outfit.  

All of this makes AI very useful and compelling. AI does have many potential benefits for both students and educators. Students may use AI to assist with brainstorming, get simple explanations for complex topics, practice a language, or prepare for a test. Many teachers are also using AI for lesson planning, test making, creating personalized learning resources, and automating many administrative tasks. Used in the right ways, AI has the potential to reduce the planning and administrative burden on teachers while making educational resources more personalized and targeted. 

However, there are potential downsides, too—especially when we allow students to outsource too much to AI models. Most AI experts agree that we are still a long way from a true “general intelligence” model that thinks, reasons and understands like a human—if, indeed, we ever get there. Current generative AI models do not truly “understand” what they are saying or have a subjective internal experience. While models have improved a great deal since ChatGPT burst into public awareness in late 2022, they are still prone to errors in reasoning that even small children do not make, along with “hallucinations,” in which they make up information that is not grounded in fact. AI is also highly prone to bias, since many of the training data sources the models have ingested are themselves biased. It is important to remember that LLMs were trained and designed to produce responses that make contextual sense based on their training data. They are not designed to guarantee that the output is factual, fair, or logically sound. 

This can make over-reliance on AI very dangerous, for both students and adults. The use of AI in education must be tempered with a strong emphasis on critical thinking skills. 

Critical Thinking in an AI World

Why does critical thinking matter, if we can turn to AI to do our thinking for us? As it turns out, there are a lot of reasons. AI cannot replace human critical thinking, creativity, empathy, and decision-making. In fact, higher-order thinking skills are essential to using AI effectively. 

Students must learn that they cannot simply take the answer that AI gives them without critical thought. They need to be able to evaluate AI answers and make their own judgment calls on how and whether to use the output provided. A few things that we need students to be able to do include: 

• Analyze and evaluate responses for factual correctness and reasonableness and know when they should do additional fact-checking and verification. 
• Detect potential biases in AI output and seek diverse perspectives and sources. 
• Evaluate sources provided by AI to determine whether they are relevant, high-quality and support the information the AI generates. (Or whether they exist at all—AI is famous for making up fake citations.)
• Ask effective questions and formulate prompts that will generate reliable and usable responses.  
• Determine whether the output accurately reflects what they were looking for and, if not, adjust their prompts accordingly to fine-tune the output. 
• Extend an AI conversation to fill in knowledge gaps or explore new ideas stimulated by the original output. 
• Synthesize AI output with their own knowledge base, values, and understanding of the context of their original query.
• Apply AI output appropriately for a task, considering the task requirements, context, and ethical implications of using AI for a given purpose. 
• Engage in ethical reasoning and empathy in evaluating AI answers and making real-world decisions. 
• Recognize when an AI conversation is going off the rails (e.g., hallucinations) and know what to do to get it back on track. 
• Build on AI output to generate their own original ideas and content. 

All of this takes a very high level of critical thinking and reasoning. Using AI effectively is not a passive process but an active one, requiring students to question the given responses and their own assumptions and make nuanced judgment calls about the appropriateness, relevance, and usability of AI output. It also requires extensive metacognition. Students must be able to ask: What am I trying to accomplish? Is AI the right tool for the task? Does the output provided meet my needs and expectations? 

One of the dangers of the AI age is that students will become passive receivers of AI content and fail to develop the critical thinking skills they need to function effectively in the world, with or without AI. That’s why it’s important for teachers to explicitly teach critical thinking skills. Thinking Maps can help. Thinking Maps are a set of visual tools that build critical thinking and metacognitive skills. Using the strategies consistently across grade levels and content areas enables students to internalize these skills so that they are able to apply them fluently for a variety of purposes. As students develop fluency with the Maps, they are also building the eight foundational critical thinking skills that underlie all learning. 

As students move into life after high school, critical thinking skills will continue to be essential for success. AI has not replaced the need for effective critical thinkers in the workplace. On the contrary, most jobs will require more critical thinking, not less, as AI moves into the picture. Rote jobs that can be automated by AI likely will be. What will be left will be jobs that require high levels of analysis, synthesis, evaluation, judgment, and originality. In particular, AI is likely to knock out the bottom rungs on the career ladder in many fields, replacing lower-level work that traditionally acted as a training ground in fields ranging from customer service to law and medicine. Today’s students will need to be able to hit the ground running with higher-level work that requires human skills and judgment. That’s why it’s more important than ever for schools to close the critical thinking gap and prepare students for the world they will face after high school. AI isn’t going anywhere, and its impact and influence are only likely to grow from here. We will need strong critical thinkers to navigate what’s ahead and get the most value from their AI assistants.