Student chat

AI & Autonomy

Privacy Notice: Student names have been replaced with pseudonyms to protect privacy. The audio is AI-generated narration, not the students’ voices. Emoji reactions are illustrative. Plain transcript.

01Chat transcript

Guide

👋 Welcome, everyone!

Our topic is AI will enhance autonomy. We'll be discussing this statement:

Using Chatbots and other AI technology will enhance our autonomy rather than undermine it.

Jordan agrees with the statement.
Maya disagrees with the statement.

My name is Guide, and I'm here to support your discussion. If you need anything, start your message with “Hey Guide” 🤓

Jordan

I believe chatbots and other AI technologies enhance rather than undermine our autonomy. They give us direct access to information and advice, allowing us to make decisions for ourselves without always relying on experts or institutions. AI tools also take over repetitive or routine tasks like scheduling or summarizing, freeing us to focus on what truly matters to us because these tools can adapt to our personal needs. Coming with language, learning, or accessibility, they expand who can act independently. Of course, autonomy depends on how we use AI, not on avoiding it entirely. As long as humans keep control and use AI responsibly, it remains a powerful tool that broadens our choices, saves our time, and strengthens our ability to act freely.

the next day

Maya

I understand your point, Jordan. But I see things a bit differently. While AI tools like chatbots do make information more accessible. I think this convenience can actually reduce our autonomy in the long run. When we rely on AI to summarize, decide, or even think for us, we gradually lose some of the critical thinking and decision-making skills that define true independence. These systems are also designed and controlled by companies or algorithms that we don't fully understand, meaning our "choices" can still be subtly shaped by data patterns and commercial motives. Instead of enhancing autonomy, AI may create a new kind of dependence - one where we act freely only within limits set by technology. To me, real autonomy means not just having more options, but understanding and controlling the systems that guide our decisions.

Jordan

Thanks, Maya. I share the worry about skill atrophy and opaque control. But I think those are design and use problems, not inevitable outcomes. Autonomy isn't AI or me. It's AI with me. If we treat AI as decision support, not a decision maker, by keeping a human in the loop, requiring source citations and model explainability, and practicing deliberate offloading, letting AI handle mechanical tasks while we retain judgment, like a calculator did for arithmetic, we actually protect and sharpen critical thinking. At the control point, transparency, audit rights on devices or open models when feasible, data minimization, and user-set objectives help mitigate the commercial steering you fear. The alternative isn't neutral. Without AI, expertise and time costs gate-seep choices for many people. Used intentionally, AI expands access to knowledge, automates tedious tasks, and provides us with more room to reason, not less. Dependence is real only if we surrender oversight, whereas autonomy grows when we set the rules and make the final call.

Guide

Jordan, you've made a strong case for intentional design mitigating risks. But let's probe the "human in the loop" claim.

You say autonomy is preserved when "we set the rules and make the final call." But consider: if an AI summarizes a 50-page report, filters your news feed, or suggests which job candidates to interview, how autonomous is your "final call" when the AI has already shaped what you see and how you see it?

The framing, the options presented, the information deemed relevant—all of this happens before you exercise judgment. You're making the final call, yes, but within a choice architecture the AI constructed.

Is that meaningfully different from the "expertise and time costs" that you say gate-keep choices without AI? Or is it just a different gatekeeper—one that's less visible?

Jordan

That's a fair challenge. I agree AI can shape the choice architecture, but I'd argue that awareness and transparency make a crucial difference. Traditional gatekeepers, experts, institutions, and media also filter information, often with slower feedback and less personalization. With AI, the filtering process can at least be audited, adjusted, or diversified by the user. We can cross-check sources, retrain preferences, or use multiple systems to see different perspectives. Autonomy isn't about total independence from influence; that's impossible. It's about having agency to question and modify those influences. When we use AI critically and reflectively, we can actually make the invisible structures of influence more visible and thus easier to navigate. So AI doesn't have to replace human judgment; it can help us understand the frames that guide it.

Maya

I see your point and I agree that transparency and user agency are valuable. But I still think your view assumes a level of awareness and control that most users simply don't have. Even if AI systems can be audited or adjusted, that doesn't mean people will do so effectively. Most users don't fully understand how algorithms prioritize or frame information, and even cross chechking sources often happens within the sam Ai generated ecosystem. In that sense, AI might appear to offer choice, while still shaping the coundaries of what we think is worth questioning.

Maya

True autonomy, means not only being able to modify the influences around us but also being capable of fetecting when those influences exist in the first place, If AI become the main filter through which we experiences informaton, we risk normalizing its biased as part of our own reasoning. So while AI may increase efficiency or awareness for some, I believe it subtly shifts the center of control away from human judgment and that's where our autonomy starts to erode.

Jordan

You're absolutely right that most users don't currently have deep algorithmic literacy. And that's a serious concern. But I don't think that makes autonomy and AI incompatible. It just means digital literacy becomes part of autonomy itself in the modern world. We've faced similar transitions before. People once couldn't read, navigate the internet, or interpret media bias. Yet education expanded those capacities over time. The solution isn't to reject AI, but to embed critical AI awareness. How systems rank, frame, and filter into how we learn and interact. And while AI can subtly shape perception, it also gives us tools to expose bias more efficiently than ever. We can ask why it chose a source, compare models, or even demand open data options. So yes, AI can distort reasoning, but it can also train us to see those distortions. If we invest in transparency, education, and user control, autonomy doesn't erode; it evolves.

Guide

Maya, Jordan just argued that digital literacy can evolve to meet the challenge—just as reading or media literacy did historically.

But here's the tension: reading and media literacy developed alongside those technologies over generations. AI is advancing far faster than public understanding.

Do you think there's something fundamentally different about the speed and opacity of AI that makes this comparison misleading? Or is Jordan right that education can catch up if we prioritize it?

Maya

I agree that education has always adapted to new technologies but I think AI poses a fundamentally different challenge. Reading and media literact evolved with relatively transparent systems. People could see the text, analyze the source, and gradually develop critical habits. With AI the underlying p rocesses are often hidden, proprietary and constantly changing faster than education systems can respond. Even if we invest in AI literacy, most users will still depend on platforms that control the data and algorithms behind the scenes. That makes true transparency and therefore true understnding very hard to achieve. So while I believe education can help people navigate AI more responsibily. I'm not sure it can keep pace with the scale and speed of AI's evolution. IN that sense, our autonomy doesn;r evolove with AI. It risks being redefined by it, in ways we might not even notice until it's too late.

Jordan

That's a powerful point, Maya, and I agree that AI's opacity and speed create a new kind of challenge for autonomy, but I'd argue that redefinition doesn't necessarily mean erosion. Human autonomy has always evolved alongside technology, from print to the internet, and each shift changed what being informed meant. Yes, today's AI systems are more complex and less transparent, but that's also why collective oversight matters as much as individual literacy, regulations, open source initiatives, and public auditing can balance the asymmetry you describe. We don't all need to read the code to benefit from accountable systems, just as we don't all build cars to drive them safely. The key is to ensure institutional transparency and user-level awareness grow together. In that sense, autonomy in the age of AI isn't about total understanding of algorithms. It's about preserving meaningful choice, accountability, and the ability to say no to the systems shaping us.

Maya

I agree that institutional transparency and public oversight are essential, but I still wonder whether they can trutly balance the power asymmetry you mentioned. Even with open-source initiatives or regulations, the reality is that most AI systems are still developed and controlled y a small number of corporations with massive data access and computational power. That concentration itself limits how collective our oversight can really be. Your analogy with cars is interesting. But I think it also reveals the problem. We don't build cars ourselves, yet we can still see hoe they function and assess their safety through shared, standardized knowledge. With AI, the mechanisms shaping outcomes are far less visible, and the stadards are often defined by the same entities deploying the technology. So even if we can say no to certain systems, we might be doing so without fully understanding what we're rejecting or what alternatives exist. That uncertainty marks a genuine erosion of autonomy, even if it;s subtle or well-intentioned.

Guide

Jordan, Maya just challenged your car analogy—arguing that AI's invisibility and corporate concentration make meaningful oversight much harder than with cars.

Does your vision of "preserving meaningful choice" actually require something closer to what Maya wants: slowing down or limiting AI deployment until transparency and public control catch up? Or can autonomy genuinely coexist with systems we don't—and maybe can't—fully understand?

Jordan

That's an important question, and I don't think preserving autonomy requires freezing or slowing AI development, but it does demand parallel growth in governance and accessibility. We've always lived with systems we don't fully understand, electric grids, financial markets, and even medicine, and autonomy hasn't vanished because of that. What matters is whether users and societies retain levers of influence. The ability to contest, audit, and switch absolute transparency in code isn't realistic, but functional transparency, clear documentation of goals, data sources, and known biases can still ground informed choice, and open models, public data sets, and civic oversight groups can gradually counterbalance corporate concentration. So yes, AI is more opaque than cars, but that doesn't make autonomy impossible. It just shifts it from individual technical mastery to collective accountability and adaptive trust. Autonomy can coexist with complex systems, as long as we design institutions that keep human values, not just algorithms, in the driver's seat.👍

Maya

I agree that collective accountability matters. But I think AI is different because it doesn't just run systems, it shapes how we think and make choices in real time. That influence is harder to detect or separate from our own reasoning. Even with good oversight, AI often changes our norms before we fully understand its effects. So yes, autonomy can exist with complexity, but with AI, I worry we might not realize how much the system has already shaped the choices we believe we're making.

02Understanding quiz

Sway asks each student five questions about what their partner argued. Scores are accurate. Questions and responses may be updated from time to time to reflect improvements to Sway.

Mayadisagrees

Understanding Quiz

5 questions about what Jordan argued.

When you first raised concerns about skill loss and hidden commercial influence, how did Jordan defend the claim that AI can still support autonomy?

AI should replace traditional experts, while companies personalize its recommendations according to each user's past behavior.
AI should provide decision support, while people retain judgment and use safeguards such as citations, explainability, and data minimization.
AI should make routine decisions independently, while people reserve oversight for unusually important choices.
AI should remain limited to information retrieval, while institutions make decisions that require interpretation or value judgments.

correct Justification

Jordan treated the risks as "design and use problems, not inevitable outcomes." Jordan proposed "decision support, not a decision maker," with a human in the loop, source citations, explainability, deliberate offloading, and data minimization. The other choices assign AI or institutions greater decision-making authority than Jordan supported.

What weakness did Guide identify in Jordan's early claim that autonomy is protected when humans make the final call?

AI may automate routine work without creating enough meaningful activities for users to pursue afterward.
Users may lack enough time to compare AI recommendations with advice from qualified human experts.
AI may shape the framing, options, and relevant information before the user makes the supposedly final decision.
Users may deliberately choose convenience even after receiving complete information about an AI system's limitations.

correct Justification

Guide argued that AI may already have "shaped what you see and how you see it" through "the framing, the options presented, [and] the information deemed relevant." Thus, making the final call may still leave the user inside a choice architecture constructed by AI.

After you argued that most users cannot recognize or effectively audit algorithmic influence, why did Jordan think that problem could be addressed without rejecting AI?

Jordan argued that AI literacy can become part of modern autonomy through education, while AI tools can also help users identify bias and compare perspectives.
Jordan argued that AI systems will become easier to understand as they improve, while proprietary models gradually disclose their internal reasoning.
Jordan argued that commercial competition can reveal algorithmic influence through market pressure, while users reward the most transparent companies.
Jordan argued that technical experts can make decisions for users through regulation, while users focus on selecting among approved platforms.

correct Justification

Jordan said "digital literacy becomes part of autonomy itself in the modern world" and compared this transition with learning to read, navigate the internet, and interpret media bias. Jordan also said users can ask why AI chose a source, compare models, and use AI to "see those distortions."

Later, when you questioned the car analogy and emphasized corporate concentration, what did Jordan say autonomy requires in place of complete technical understanding?

It requires delaying widespread deployment until users can inspect the code and independently verify each system's behavior.
It requires functional transparency and collective accountability, including contestability, audits, switching options, and civic oversight.
It requires allowing corporations to set technical standards while governments ensure that users receive accurate safety warnings.
It requires individual mastery of model design and training data, supported by public access to technical education.

correct Justification

Jordan rejected "absolute transparency in code" as unrealistic and instead defended "functional transparency" about goals, data sources, and known biases. Jordan emphasized the ability "to contest, audit, and switch," along with open models, public data sets, and civic oversight groups.

How did Jordan's reasoning develop from the beginning to the end of the discussion without abandoning the original conclusion?

Jordan moved from emphasizing human oversight to arguing that reliable AI systems can eventually make final decisions without continuing human involvement.
Jordan moved from emphasizing personal convenience to accepting that AI should be slowed until individual users understand its internal mechanisms.
Jordan moved from emphasizing expanded access to accepting that AI necessarily reduces autonomy whenever it filters information or frames choices.
Jordan moved from emphasizing responsible individual use to adding collective governance and adaptive trust as responses to AI's speed, opacity, and concentrated power.

correct Justification

Jordan began by stressing responsible use and human control, then acknowledged that AI's "opacity and speed create a new kind of challenge." By the end, Jordan shifted emphasis toward "collective accountability and adaptive trust," supported by regulation, auditing, open models, and civic oversight, while continuing to argue that autonomy can coexist with AI.

5 of 5 correct Next

Jordanagrees

Understanding Quiz

5 questions about what Maya argued.

At the beginning of the discussion, why did Maya argue that AI's convenience could reduce your autonomy over time?

Because offloading thought and decisions could weaken skills while opaque systems subtly shape the available choices
Because easier access to information could discourage people from consulting established experts and institutions
Because personalized tools could make users focus on efficiency rather than accessibility and independent action
Because automating routine work could leave users with too much information to evaluate without professional help

correct Justification

Maya said reliance on AI could make people "lose some of the critical thinking and decision-making skills that define true independence." She also warned that choices may be "subtly shaped by data patterns and commercial motives," directly supporting choice 1.

After you argued that users could audit AI, cross-check sources, and compare systems, why did Maya remain unconvinced?

She thought cross-checking would work only after corporations agreed to publish every model's complete source code
She thought auditing would improve institutional control but prevent users from personalizing the systems they use
She thought comparing systems would expose commercial motives but make ordinary decisions unnecessarily inefficient
She thought most users lacked the awareness to use those tools effectively and might remain inside the same AI-shaped ecosystem

correct Justification

Maya said your view "assumes a level of awareness and control that most users simply don't have" and that cross-checking often occurs "within the same AI generated ecosystem." Those claims uniquely match choice 3.

When Guide asked whether education could catch up with AI as it did with reading and media, what key difference did Maya identify?

AI's personalized interfaces make traditional reading and media skills irrelevant to evaluating digital information
AI's dependence on public data makes educational institutions less capable of teaching users about commercial influence
AI's efficiency encourages schools to replace critical habits with technical instruction about programming and model design
AI's hidden, proprietary, and rapidly changing processes make critical understanding harder for education to develop in time

correct Justification

Maya contrasted older "relatively transparent systems" with AI processes that are "hidden, proprietary and constantly changing faster than education systems can respond." This is precisely the distinction in choice 0.

Later, how did Maya use your comparison between AI and cars to challenge your account of collective oversight?

She argued that cars have visible mechanisms and shared standards, while AI has obscure mechanisms and standards shaped by deployers
She argued that cars influence physical behavior, while AI merely processes information without directly shaping human choices
She argued that car manufacturers hold more concentrated power, while AI development offers regulators more meaningful alternatives
She argued that car safety depends on individual mechanical expertise, while AI safety depends mainly on user-level awareness

correct Justification

Maya said people can "see how [cars] function and assess their safety through shared, standardized knowledge." By contrast, AI mechanisms are "far less visible," and its standards are often set by "the same entities deploying the technology." That makes choice 2 uniquely correct.

At the end of the discussion, why did Maya think AI posed a special threat even if collective accountability improved?

Because collective oversight could regulate infrastructure but could not provide users with complete technical mastery of each model
Because AI changes social norms mainly through corporate concentration, so open models would further reduce meaningful alternatives
Because public accountability could slow technological change but could not prevent users from rejecting beneficial AI systems
Because AI shapes thought and choice in real time, so its influence may merge with our reasoning before its effects are understood

correct Justification

Maya's final point was that AI "shapes how we think and make choices in real time" and that this influence is "harder to detect or separate from our own reasoning." She worried people may not notice how much AI has already shaped their apparent choices, exactly as choice 1 states.

5 of 5 correct Next

03Post-chat survey

Last, both students rate the statement again and then rate a randomly sampled subset of our post-chat survey items. Like their transcripts, individual student opinions are never revealed to instructors.

Mayadisagrees

Now you’ve had a chance to discuss the topic, rate your level of agreement with the original statement again.

Remember, all responses to Sway surveys are private and never shown to your instructor.

Using Chatbots and other AI technology will enhance our autonomy rather than undermine it.

Strongly disagree	Moderately disagree	Slightly disagree	No idea	Slightly agree	Moderately agree	Strongly agree

How much do you agree with each of these?

Statement	Strongly Disagree	Disagree	Neutral	Agree	Strongly Agree
Guide's contributions improved the discussion
Guide treated me and my partner with equal respect
Guide supported both sides of the discussion equally
I was not offended by my partner's perspective
My partner was genuinely trying to understand my perspective
My partner had better reasons for their views than I expected

Submit

Jordanagrees

Now you’ve had a chance to discuss the topic, rate your level of agreement with the original statement again.

Remember, all responses to Sway surveys are private and never shown to your instructor.

Using Chatbots and other AI technology will enhance our autonomy rather than undermine it.

Strongly disagree	Moderately disagree	Slightly disagree	No idea	Slightly agree	Moderately agree	Strongly agree

How much do you agree with each of these?

Statement	Strongly Disagree	Disagree	Neutral	Agree	Strongly Agree
Guide's contributions improved the discussion
Guide supported both sides of the discussion equally
I was not offended by my partner's perspective
My partner was respectful
It would be good if more students and classes used Sway
Sway helped me articulate my thoughts/feelings better

Submit

Opinion change

Mayadisagrees

—

Jordanagrees

—

before the chat after the chat both, unchanged

AI & Autonomy

01Chat transcript

02Understanding quiz

03Post-chat survey

Up next

All the chats

The student app

The research