We're all using AI to speed things up, but Advait Sarkar warns that outsourcing our thinking to chatbots might be costing us more than we realise. He suggests that instead of just using AI as a quick assistant, we should be using it as a tool to actually make us think better.
Key Takeaways
- AI can make us lazy thinkers, reducing our creativity, critical thinking, and memory.
- We risk becoming "middle managers for our own thoughts" if we rely too heavily on AI.
- AI can be designed as a "thinking tool" that challenges us and helps us learn.
- This shift requires conscious design choices to protect and enhance human cognition.
The Rise of Outsourced Thinking
Picture a typical day for a knowledge worker. Your inbox is overflowing, so you ask AI to draft a reply. You need to write a report, but the blank page is daunting, so you get an AI draft. You're suddenly a "professional robot opinion checker," wondering if you even agree with the AI's output. You have data to analyse, so AI does it. You need to make a presentation, so AI helps. You need to code something, AI assists. This isn't some far-off future; it's a pretty accurate picture of knowledge work today.
We've entered an era of "outsourced thinking." We're not really engaging with our work materials anymore; we're like tourists in our own ideas. Our relationship with our work is completely mediated by AI. Some might call it efficient, but what's the real cost?
The Cognitive Trade-Offs of AI Assistance
Using AI primarily as an assistant can have serious effects on our thinking. Let's break it down:
- Creativity: While we might think AI boosts creativity by giving us quick access to new ideas, studies show that groups using AI assistants actually produce a smaller range of ideas compared to those working manually. We're creating a "hollowed-out mind" that keeps suggesting the same few things.
- Critical Thinking: When knowledge workers use AI, they report putting less effort into critical thinking. This effect is even stronger when people trust the AI more and themselves less.
- Memory: If AI writes for you, you'll remember less of what you wrote. Similarly, reading AI-generated summaries means you remember less than if you'd read the original document.
- Metacognition: This is the ability to think about your own thinking process. Working with AI requires significant metacognition about task goals, analysis, AI output, and your ability to evaluate it. These things get lost when AI mediates the process.
Essentially, we're becoming "middle managers for our own thoughts." The result? Fewer ideas, less critical engagement with them, poorer recall, and difficulty in thinking about our own thinking.
AI as a Tool for Thought
This doesn't have to be the way it is. Instead of just an assistant, AI should be a tool for thought. It should challenge us, not just obey us. We're at a critical point where generative AI is reshaping the workplace, and we need to steer this change towards human values.
A thinking tool, unlike a simple assistant, does more than just get the job done faster. It helps us understand the job better, do it better, ask the right questions, and explore the unknown.
A Prototype in Action
Researchers at Microsoft Research have developed a prototype called a "thought tool." Let's look at Clara, who runs a beverage bottling company. She needs to write a proposal responding to a new industry report on sustainable packaging.
- Loading Information: Clara uploads relevant documents: meeting notes, an internal report, and the industry report.
- Using "Lenses": Instead of just summaries, the tool offers "lenses" – customisable representations of the text that highlight what's most relevant to her task. She chooses a "consumer" lens and dives deeper into a specific section.
- AI "Provocations": As she reads and takes notes, the AI offers "provocations" – comments and critiques that highlight potential opportunities or challenges. These aren't meant to be blindly accepted but to stimulate her thinking.
- Building an Outline: Clara manually builds an outline for her argument, linking her thoughts directly to the source documents. The AI can then help generate draft paragraphs based on this outline.
- Iterative Refinement: Clara can easily adjust the length of paragraphs or test different tones. The AI can even offer alternatives, identify fallacies, or provide counter-arguments to help her strengthen her own points.
Crucially, there's no chat box. Clara isn't chatting with the AI; she's interacting with her work, with the AI providing silent, appropriate assistance. The output is still hers, deeply rooted in her cognitive effort, decisions, and expertise.
The Promise of Thinking Tools
Studies on tools like this show promising results. We can reintroduce critical thinking into AI-powered workflows, reverse the loss of creativity, and build powerful memory aids. The key is design principles:
- Maintain physical engagement: Keep users actively involved with the material.
- Provide productive resistance: Challenge users rather than just agreeing.
- Support metacognition: Help users think about their own thinking process.
These principles can apply not just to professional knowledge workers but also to our daily lives, hobbies, and education.
Why Protect Human Thought?
Even if AI eventually surpasses human thinking capabilities, why should we bother protecting and enhancing human cognition? There are two main reasons:
- Unique Human Strengths: There may always be ways of thinking that remain uniquely human, strengths we haven't even discovered yet.
- Human Agency: The ability to think well is fundamental to human agency, empowerment, and flourishing. If machines can think, grieve, pray, and love for us, does it matter if we can't? For Advait Sarkar, the answer is a clear yes.
He asks: Would you rather have a tool that thinks for you, or a tool that makes you think?