How to Stop AI from Killing Your Critical Thinking | Advait Sarkar | TED — Transcript
Full transcript
- 0:04I'm here today to talk about thinking for yourself.
- 0:08And I must admit, I did use AI to help me think about it.
- 0:11(Laughter)
- 0:12The irony is not lost on me.
- 0:15But the way I did so is not by using AI as an assistant
- 0:19to help me prepare this talk faster.
- 0:21Rather, I use AI as a tool for thought.
- 0:25And by the end of this talk,
- 0:26I will have explained what I mean by that,
- 0:29why it's important,
- 0:30and given you a glimpse of how it might work.
- 0:33But first I need to set the scene.
- 0:36Let's look at a day in the life of a 21st-century knowledge worker.
- 0:40I arrive at my office and look at my inbox full of emails.
- 0:45Oh.
- 0:47Let's summarize it.
- 0:49OK, I'm struggling to figure out how to respond here,
- 0:52so let's get AI to write a response.
- 0:56Next, I need to write a report.
- 0:59But I'm struck by the blank-page problem.
- 1:02I know, I'll drop in some resources and get an AI draft.
- 1:05Looks good to me.
- 1:07By the way, a writer's block used to be staring at a blank page.
- 1:11Now it's staring at a page that AI filled out for me
- 1:13and wondering if I agree with it.
- 1:15I've become a professional validator of a robot's opinions.
- 1:19(Laughter)
- 1:22I've got some data to analyze.
- 1:23Maybe AI can analyze this data for me.
- 1:27Probably correct.
- 1:29OK, I've got to make a deck as well.
- 1:32You know the drill.
- 1:34Alright.
- 1:35Oh, I was supposed to prototype something as well.
- 1:38OK, let me vibe code something.
- 1:40Alright, all this looks good, let's go.
- 1:44This isn't a vision of the future.
- 1:47This is a completely plausible,
- 1:49if slightly exaggerated, picture of the world of knowledge work today.
- 1:54Welcome to the age of outsourced reason.
- 1:57Where the knowledge worker no longer engages with the materials of their craft.
- 2:01We've become intellectual tourists.
- 2:03In our own work, we visit ideas.
- 2:06We don't inhabit them.
- 2:09Our relationship to our work
- 2:12is entirely intermediated by AI.
- 2:16Some might say alienated.
- 2:17We've heard that story before.
- 2:20What's wrong with this picture?
- 2:23For one thing, it's only one step removed from this,
- 2:26which is important.
- 2:27But that's a different talk.
- 2:29What I want to focus on today
- 2:32is that using AI in this way
- 2:34can have profound implications on human thought.
- 2:39Consider creativity.
- 2:40On an individual level,
- 2:41we might think that AI is a creativity boost,
- 2:44giving us rapid access to new ideas.
- 2:47But numerous studies have shown that on a collective level,
- 2:51knowledge workers using AI assistants
- 2:53produce a smaller range of ideas than a group working manually.
- 2:58We've created a hive mind.
- 3:01Except the hive is really boring and keeps suggesting the same five ideas.
- 3:06Consider critical thinking.
- 3:08We surveyed knowledge workers about their use of AI.
- 3:12They reported that they put less effort into critical thinking
- 3:15when working with AI
- 3:18than when working manually.
- 3:21And this effect was greater
- 3:22when they had greater confidence in AI
- 3:24and less confidence in themselves.
- 3:27Consider memory.
- 3:30When people rely on AI to write for them,
- 3:33they remember less of what they wrote.
- 3:35And when they read AI-generated summaries,
- 3:38it's hardly surprising that they remember less
- 3:40than if they'd read the document.
- 3:42And finally, consider metacognition,
- 3:44which is the ability to think about your own thinking process.
- 3:48Working with AI requires significant metacognitive reasoning
- 3:51about your task goals,
- 3:53decomposing the task,
- 3:55the applicability of gen-AI,
- 3:57your ability to evaluate the output.
- 3:59These are things which are built into the process
- 4:02of working directly with the material,
- 4:04and which become problematic
- 4:06when that material engagement becomes intermediated.
- 4:09Basically, we've become middle managers for our own thoughts.
- 4:13So what's the score?
- 4:15We have fewer ideas.
- 4:17We think about them less critically.
- 4:19We remember them less well,
- 4:21and we have a harder time doing it.
- 4:24Taken together, we can see that AI-assisted workflows
- 4:27can have profound effects on human thinking.
- 4:30And this extends even to seemingly trivial mundane tasks,
- 4:34because these everyday opportunities for exercising our creativity,
- 4:37our critical thinking and our memory
- 4:39are essential for protecting our cognitive musculature
- 4:42and allow us to rise to the occasion
- 4:44when an exceptionally complex task comes our way.
- 4:48Studies show that when we don't use our brains,
- 4:52they get worse at brain things.
- 4:54Nobel Prize committee, please hold your applause.
- 4:59Is this the cost of progress?
- 5:04We've solved the problem of having to think.
- 5:07Unfortunately, thinking wasn't actually a problem.
- 5:09(Laughter)
- 5:11It's like we invented a cure for exercise
- 5:14and then wondered why we're out of breath all the time,
- 5:17you know?
- 5:19It doesn't have to be this way.
- 5:24Beyond AI as an assistant,
- 5:26I believe that AI should be a tool for thought.
- 5:29AI should challenge, not obey.
- 5:32And I believe that right at this moment, we are at a critical juncture
- 5:35where the world of work is poised to be transformed by generative AI,
- 5:39and we must act now
- 5:40to shape and drive that transformation towards humanistic values.
- 5:44Of these two diverging roads,
- 5:47we must take the one less traveled.
- 5:50Beyond getting the job done,
- 5:52a tool for thought helps us better understand the job.
- 5:56Beyond getting it done faster,
- 5:58it helps us get it done better.
- 6:01Beyond getting us to the right answers,
- 6:03a tool for thought helps us ask the right questions.
- 6:07Beyond automating known processes,
- 6:09it helps us explore the unknown.
- 6:12What does this look like?
- 6:14What I'm about to show you is a prototype,
- 6:16developed by my colleagues and me
- 6:18at the Tools for Thought team at Microsoft Research in Cambridge.
- 6:22Now, please bear in mind that this is a live research prototype.
- 6:26It's not a product.
- 6:27And it's just one of a series of explorations
- 6:30that our team is conducting
- 6:31to study how different modes of working with AI
- 6:33can enhance human thought.
- 6:35So let's look at a fictitious example.
- 6:39Clara and her colleagues run a company
- 6:43that sells bottled beverages.
- 6:46They've just had a meeting to discuss a new industry report
- 6:49that seems to have some pretty important findings
- 6:52about consumer preferences for sustainable packaging.
- 6:55Clara's colleagues have asked her to write a proposal
- 6:58arguing for how the company ought to respond.
- 7:01So she really needs to get to grips with this proposal --
- 7:04She really needs to get to grips with this report,
- 7:07understand its findings and its data
- 7:09and how it fits into her business context.
- 7:13She starts by loading some documents into her workspace.
- 7:19There's the meeting transcript to remind her what was discussed.
- 7:22There's a recent internal report from her own business.
- 7:26And of course,
- 7:27there’s the industry report, which she opens.
- 7:29She sees an overview of the document
- 7:31along with section-by-section summaries.
- 7:34Except these aren't really just summaries.
- 7:36We think of them more as lenses.
- 7:38They're customizable micro representations of the text
- 7:41that can emphasize what is most relevant to the task at hand.
- 7:45So in this case, Clara selects the consumer’s lens.
- 7:51She can select a section for deeper reading,
- 7:53in this case the first one.
- 7:55As she reads, she makes notes about her thoughts
- 8:00and highlights excerpts from the document.
- 8:05As she reads,
- 8:06she also sees AI-generated commentary and critiques.
- 8:10We call these provocations.
- 8:13Here's a provocation that raises a potential opportunity,
- 8:17which she highlights and annotates.
- 8:19Note how this process is a hybrid of completely manual reading
- 8:24and completely relying on AI to read for you.
- 8:26Clara still reads, but intentionally and strategically.
- 8:32Now, as Clara is working,
- 8:33she's building up an outline of her argument manually
- 8:37in this pane on the right.
- 8:39This outline is lightly structured
- 8:41and allows her to sketch out the flow of her argument at a high level,
- 8:46while still retaining deep connections
- 8:47and being grounded in the source documents.
- 8:50As a result of which we can already generate a draft of the proposal,
- 8:54and Clara can do things here like add a heading to the outline
- 8:57to generate a paragraph.
- 8:59But what I want to draw your attention to here
- 9:01is that while this text is AI-generated,
- 9:05Clara has a completely different relationship to this text
- 9:07than if she just dropped in some documents and said, write me a report.
- 9:11Because this text is deeply rooted in a cognitively effortful
- 9:16but interactionaly effortless thought process.
- 9:20It reflects Clara’s decisions, Clara’s judgments,
- 9:23Clara’s unique personal, professional expertise.
- 9:30She sees another provocation, this time in the outline.
- 9:35In this case, she decides that while the provocation is useful,
- 9:39she does not need to address it.
- 9:41Unlike typical AI suggestions,
- 9:43provocations are not meant to be applicable all the time.
- 9:46They're instead meant to stimulate your thinking about your work.
- 9:49Because if you understand your work well enough,
- 9:52deeply enough to make the confident decision
- 9:55not to accept a piece of feedback,
- 9:57then the feedback process is still working as intended.
- 10:02But we're not done yet.
- 10:03Clara has entirely new ways of interacting with this text
- 10:06because of generative AI.
- 10:09A really simple example is that she can just resize a paragraph
- 10:12to change its length.
- 10:14She can also rapidly test different versions of this text.
- 10:17For instance, in this paragraph,
- 10:19she's wondering whether it would be more effective
- 10:21if it took a more inspirational or more practical tone.
- 10:24So she selects one of these customizable dimensions.
- 10:30And previews a few alternatives
- 10:32and selects one.
- 10:35And at select strategic points,
- 10:37indeed, she writes.
- 10:39As she writes, she sees provocations that,
- 10:44rather than autocompleting her ideas,
- 10:47they raise alternatives,
- 10:48they identify fallacies,
- 10:50they offer counterarguments
- 10:51to help her strengthen and develop her own argument.
- 10:56There's something you won't find anywhere in this interface.
- 10:59And that's a chat box.
- 11:01Clara’s not having to chat with anything to do her work,
- 11:04yet she is silently and appropriately assisted by her computer
- 11:08as a computer
- 11:09and not as an ersatz human.
- 11:12To put it simply, we have gone from this ...
- 11:19To this.
- 11:20Throughout this process,
- 11:22Clara has been assisted and yes,
- 11:24probably worked faster because of AI.
- 11:27But she's also maintained direct material engagement at strategic points.
- 11:31She read the relevant portions of the document herself.
- 11:34She constructed her decisions on her argument herself.
- 11:37And ultimately it can be said she has written this document herself.
- 11:41Moreover, she worked better because of AI.
- 11:44AI provocations at every stage of the process
- 11:47kept her metacognitively engaged,
- 11:49always looking for critiques, alternatives and lateral moves.
- 11:55We have been studying the effects of tools like this.
- 11:58And the results are promising.
- 12:01You can demonstrably reintroduce critical thinking
- 12:04into AI-assisted work flows.
- 12:06You can reverse the loss of creativity and enhance it instead.
- 12:11You can build powerful tools for memory
- 12:13that enable knowledge workers to read and write at speed
- 12:17with greater intentionality, and remember it, too.
- 12:20It turns out, with the right principles of design,
- 12:24you can build tools that are the best of both worlds.
- 12:27Applying the awesome speed and flexibility of this technology
- 12:31to protect and enhance human thought.
- 12:33These are simple, general principles,
- 12:36like ensuring that the tool preserves material engagement,
- 12:41offers productive resistance,
- 12:43and scaffolds metacognition.
- 12:46And while we've been primarily studying professional knowledge workers,
- 12:49we believe that these principles can extend to all aspects of AI use,
- 12:53including when we use it in our daily lives, our hobbies,
- 12:57and even in education.
- 12:59I repeat, efficiency is not the aim of Tools for Thought.
- 13:03Better thinking is.
- 13:05But sometimes you can have both.
- 13:08I used to think there was no such thing as a free lunch in human thinking.
- 13:12This is so much better than a free lunch.
- 13:14This is a lunch that pays you to eat it.
- 13:17(Laughter)
- 13:20I want to close with some thoughts on the values that we have
- 13:23in developing AI software.
- 13:25What if AI gets to the point where it can do a better job
- 13:29of thinking than humans?
- 13:30Why should we care so much
- 13:31about protecting and augmenting human thought?
- 13:35There's two reasons.
- 13:36First,
- 13:38there may always be ways of thinking that remain unique human strengths
- 13:41of which we may not even be aware.
- 13:44Second, perhaps more importantly,
- 13:46we take the position that the ability to think well
- 13:49is essential for human agency and empowerment and flourishing.
- 13:54This echoes an ancient question.
- 13:57People once asked
- 13:59if writing, if books, if the internet
- 14:02can remember for us,
- 14:03does it matter that we cannot?
- 14:05People once asked if maps can navigate for us,
- 14:07does it matter that we cannot?
- 14:10Now we ask if machines can think for us,
- 14:13does it matter that we cannot?
- 14:16If machines can speak for us,
- 14:18grieve for us, pray for us, love for us,
- 14:23does it matter that we cannot?
- 14:26To me, the answer is pretty obvious.
- 14:29When I began studying human-AI interaction 13 years ago,
- 14:32it was inconceivable to me
- 14:34that we would be asking these questions in my lifetime.
- 14:37But we are.
- 14:38And we must.
- 14:40I'll leave you with this thought.
- 14:45What would you rather have?
- 14:47A tool that thinks for you,
- 14:49or a tool that makes you think?
- 14:51(Applause)
About this transcript
This page contains the full transcript of How to Stop AI from Killing Your Critical Thinking | Advait Sarkar | TED by TED, generated from the public captions YouTube serves with the video. The transcript has 2,173 words across 307 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.