Implementation showcase
a semester of disagreement
In the spring of 2026, an instructor at a small liberal arts college built a semester around structured disagreement. Six times between late January and mid-April, students opened a Sway assignment: a set of statements on a contested subject, written by their instructor. Each student recorded opinions privately, and Sway paired them with a classmate who saw one of those statements differently. The pair then had the argument — in writing, one on one, with Guide, Sway's AI facilitator, in the conversation.
The subjects were not warm-up material. The course moved from speech on the students' own campus to social media and AI, then inequality, immigration, abortion, and climate — 144 conversations in all. This page follows the semester from the first assignment to the last — what the instructor put in front of the students, what the students did with it, and what the record shows changed — and then follows four of the students themselves from January to April.
Everything here comes from the course's aggregated instructor report — the anonymized record Sway generates from every assignment — and from the conversation records behind it. Quotations from Guide and from students are verbatim. The college, the instructor, and the students are deidentified.
The main narrative takes ten to twelve minutes to read. The + drawers hold the deeper layer — more conversation accounts, the survey distributions, and the numbers behind each claim — and roughly double the reading time.
How this showcase was built
Sway produces a deidentified report for every assignment: opinion distributions per statement, AI-written summaries of each conversation of ten or more messages, matched pre/post opinion measures, post-chat survey responses, and students' optional written feedback. This showcase draws on the aggregated report covering all six of the course's assignments, generated in July 2026.
Conversation accounts below are condensed from the report's per-thread summaries; quotations attributed to Guide are taken verbatim from the conversation records; quotations from students are verbatim from their conversations and their written feedback. Where a statement or summary named the college, we have substituted "the college." Instructors cannot read identifiable transcripts, and student names are redacted throughout Sway's reporting; nothing on this page identifies the institution or any person in it.
Where we report numbers, we give the sample they come from. The per-statement matched samples are small — often six to sixteen students — and we flag that where it matters.
Two notes on how material was chosen. Every conversation in the course is scored internally against a six-part quality rubric (reasoning, facilitation, closure, and so on); we used those scores to find the strongest threads, then read the originals in full before excerpting — the excerpts here come from the top of a distribution whose median conversation the rubric rates "good," not from a corpus of uniform brilliance. And the four students profiled below were selected from the course's own records — participation, opinion measures, and written feedback — for the range of their semesters: how far their opinions traveled, how they argued, and what they took from the experience.
Prologue · the design
Fifty-six statements
Each assignment gave students a menu of seven to ten statements rather than a single prompt. The menus were built with evident care. Some statements planted a flag on a live controversy; others were engineered around a single contestable word; a few were philosophers' conditionals, granting one side's premise to test what follows from it:
The menu is what made matching work. On some statements the class was lopsided — students rated "The Trump administration is wrong to use the leverage of federal funds to pressure colleges" at +1.89 on average, on a scale from −3 to +3. But every menu also contained statements that split the room almost exactly: "administrators have gone too far in trying to regulate speech on campus" sat at −0.12, and "the wealthiest deserve their success" at −0.25. Whatever a student believed, some statement in the menu put a classmate on the other side of it.
The mechanics were the same all term. Students recorded opinions privately; Sway paired disagreeing classmates; each pair had a conversation of at least thirty minutes, synchronous or not, with Guide posing questions, testing arguments, and taking no side. Of the 144 conversations, 139 were standard disagree matches. The remainder show the system absorbing classroom reality: two trios when numbers were odd, one same-side pair with a student assigned to play devil's advocate, and two students whose partners dropped out — so they finished the assignment debating Guide itself.
The six menus
Each assignment's statements, by theme. A sample from each menu:
- Campus speech (7 statements) — "Colleges and universities are incubators of 'woke' ideologies" · "Student protestors should be allowed to raise anti-Israeli views, even if it makes some students on campus uncomfortable" · a scenario about disinviting a cancer-biology speaker over donations to a group criticized for anti-LGBT positions.
- Social media & AI (9) — "You should NOT be allowed to operate an anonymous account on a social media platform" · "Those who worry about the existential risks from artificial intelligence are overreacting" · "The economic impact of AI is likely to be positive."
- Inequality (10) — "Inequality is morally wrong" · "Government benefits for unemployed and poor people are too high in the US" · "Concern about inequality between the rich and everyone else is misplaced; what matters is whether people at the bottom can meet their basic needs."
- Immigration (10) — "National borders are the world's most consequential form of inherited privilege" · "The United States benefits most from immigration when it selects for skills and education, not family ties or humanitarian claims" · "A society that genuinely values workers' rights and wage equality should be cautious about large-scale low-wage immigration, even if it is sympathetic to immigrants themselves."
- Abortion (10) — "An abortion is a tragedy" · "Abortion policy should be decided by individual states, not the federal government" · "Third-trimester abortions should be permitted only when the pregnant person's life is at risk."
- Climate (10) — "Developing countries should be allowed to prioritize economic growth over aggressive climate action" · "Individuals have a personal moral obligation to significantly change their lifestyle to address climate change" · "Wealthy countries that have historically emitted the most should pay reparations to countries most affected by climate change."
Class means across the fifty-six statements ran from +1.89 (against federal funding leverage) to −1.59 (against punishing professors for pronoun mistakes), and sixteen of them landed within half a point of zero — the even splits the matching thrives on.
Chapter I · January
Their own campus
The course did not begin at a safe distance. The first assignment asked students to argue about the place they were sitting in:
The class split almost evenly, and the conversations went directly at the hardest local question: where political expression ends and intimidation begins. In one thread, a student at "moderately agree" — arguing that the administration recodes pro-Palestinian speech as an identity conflict — was matched with a student at "strongly disagree" who thought the college had allowed ample peaceful protest and stepped in only when a few students crossed a line. They kept at it for five days and forty-three messages, asynchronously, between classes. Guide let neither position rest:
Student 1, you've said free speech should allow people to "freely address their opinions" — but does that include slogans that some students find threatening or offensive? Student 2 pointed to "from the river to the sea" as a case where Jewish students felt targeted. You acknowledged slogans can be "speech violence to other groups." So where's your line? Should the administration have intervened in that case, or was their response still overreach?
By the end the pair had done something subtler than converge: they located the exact point their disagreement rests on, and connected it to the week's reading.
Similar to you, I genuinely am not sure about the case of "from the river to the sea". I believe our fundamental disagreement comes from the fact that I think if there is any gray area, the college should err on the side of caution, and you believe that free expression should be the priority. I tend to agree with you in many cases actually about the importance of free speech, but I still believe that institutions have an obligation to monitor certain types of expression that may have offensive implications…
…the broader concern I have with free speech is how opinions/ideas here in the U.S are deeply embodied and personalized. What I mean by that is here in the U.S, when individuals express their opinions, they believe their ideas are an extension of their identities. Thus, when they are challenged and "attacked" by others, they feel personally threatened and challenged. … The western rhetoric of "I think therefore I am" sometimes could be flawed :))
That's a really good point, and I think that incorporates one of the readings we had to do for last Tuesday very well. People need to be able to separate their ideas from their identities.
Verbatim, trimmed where marked. The two closed the thread with "We made it! Thanks for a great conversation, you had some great points" and "Thank you so much for your time!"
Other pairs pulled the question apart differently. Several distinguished formal policy from informal pressure — concluding that the administration had not overreached on paper, but that on a campus this small, where you see the same people every day, dissent carries a social price a policy never touches. One student, arguing that classroom culture chills speech more than administrators do, ended by conceding personal complicity in staying within like-minded circles.
No pair settled who should decide where protected speech ends. The report lists it as the semester's clearest unresolved question, and it will come back at the end of this page.
More from assignment 1
The limiting principle
A pair debating whether the college could restrict speech to protect "stability and funding" converged on conditional acceptance — until Guide asked what prevents that logic from justifying any future restriction: "If maintaining stability and preserving funding are acceptable reasons now, couldn't they justify almost any future restriction…? Where's the limiting principle in your framework?" The pair's answer — restrictions become unacceptable when they undermine the college's core commitments — narrowed the disagreement to where that threshold sits.
"Incubators of woke ideologies"
Seven separate pairs took on the incubator statement, and nearly all of them ended up distinguishing claims the statement collapses: active indoctrination versus selection effects versus peer culture. One conversation reached the question a first-year seminar would be proud of: if interpretive frameworks shape what counts as rigorous, "how would a student even recognize when they're being taught facts versus being socialized into a particular interpretive lens?" Another pair proposed remedies — mandatory first-year training in civil discourse and media literacy, panels with substantively opposed experts — then had to face Guide's objection that filtering "inflammatory" speakers might be the very censorship they opposed.
Debating Guide itself
One student's partner dropped out, so Guide took the other side — and argued it. When the student defended the incubator claim from social pressure at elite colleges, Guide pressed the evidence standard: "If students arrive already progressive and leave progressive, is that incubation or just continuation? What would change your mind?" The report notes the student narrowed the claim from "top schools" generally to an experience-based account of one campus.
The small-campus problem
A pair agreed formal regulation was mild but described how repeated social contact raises the cost of dissent — "intellectual gossip," one called it — and floated honor codes for classroom confidentiality. Guide tested the fix against the diagnosis: if the barrier is campus-wide culture, does a classroom-level rule actually protect anyone once the discussion leaves the room?
Chapter II · early February
Social media and AI
The second assignment moved to the technologies in the students' pockets, and a pattern emerged that held for the rest of the term: pairs converging not on a side but on a distinction. Across the social media threads, students who began at "net positive" and "net negative" repeatedly ended at the same diagnosis — the problem is not the tools but the engagement-driven business model underneath them — and then disagreed, more precisely, about whether that diagnosis is fixable.
The AI statements produced the assignment's hardest question. Students who defended AI's benefits kept conceding that the benefits depend on regulation, redistribution, or business-model reform arriving on schedule. Guide made the concession expensive:
When you say AI will have a positive impact on the economy, what exactly do you mean by the economy if not the economic well-being of the people within it?
Your historical analogy to factories is worth examining more carefully. Yes, factories created new jobs—but that transition also involved decades of labor unrest, child labor, dangerous working conditions, and massive inequality before regulations and unions helped distribute gains more broadly. Is it obvious that we should expect a smoother transition with AI?
One AI thread deserves noting for where it took place. Two students spent four days debating whether AI erodes the capacity it assists — inside an AI-facilitated discussion, with the facilitator pressing both of them to reason harder. Neither remarked on the setting. The exchange stands on its own:
As AI increasingly takes over tasks like memory, decision-making, and problem-solving, it can weaken our capacity for independent and critical thinking through cognitive offloading. After years of experiencing the world this way, humans may easily shift from active reasoning to more passive reliance on algorithmic outputs.
I find your point on the impact AI has on the way we think, rather than the way we "do" to be really convincing. … I definitely side closer to you now, and was perhaps focussing too much on the semantics of the statement.
Verbatim, trimmed where marked. The first student had opened at "strongly agree" that AI's impacts would be very positive.
The assignment also produced the semester's first genuinely original piece of policy design. A pair arguing over anonymous accounts — one seeing anonymity as the engine of online abuse, the other defending it as protection for whistleblowers and vulnerable speakers — worked their way to a "verified but pseudonymous" model: platforms verify identity privately while usernames stay public-facing. The disagreement that remained was about scope, not principle.
How far the format could stretch is visible in one social-media thread that reads like an exchange of essays. Over six days, two students worked through Hobbes's social contract as a frame for platform consent, a case where feminist scholarship on Chinese social media backfired, the suffragettes, #MeToo, and Australia's under-16 ban. One argued that social media inflicts a harm no body registers — "the painful cancer of our soul." The other held the line on net benefit, but not where he started:
…you've convinced me that "net positive" could be setting the bar too low when billions have no meaningful choice but to accept the harms with the benefits.
In terms of where I land after all this, then, I would say that social media's impact is net positive but not by much, and only because the alternative of depriving billions from resources and communities they desperately need would be catastrophic. We need serious reform, I think.
Verbatim. Concession on the record, position held — revised, not surrendered.
More from assignment 2
Analogies under load
Guide's most frequent move this week was stress-testing comparisons. A student likening social media to cigarettes was asked why the analogy survives #MeToo, which the student had already credited as real empowerment; a student comparing platform regulation to alcohol rules got "Are you really comparing data privacy regulation to Prohibition?"; and a pair testing whether governments can regulate AI at all landed on a modest analogy that held — most users won't bypass safeguards for the same reason most people never touch a VPN.
A statistic doing too much work
In one social media thread, a claimed "189% increase in self-harm among girls 10–14" effectively settled the debate on its own. The report flagged it to the instructor as a statistic that spread without scrutiny — one of several the semester surfaced, and exactly the kind of finding an instructor can teach from the following week.
Design proposals
Pushed past "social media is bad for attention," one pair started designing: a two-to-three-second black screen before new content loads, screen-free zones in schools and public spaces, stage-based rules for minors. Another pair, both admitting they would not delete their own accounts, had to reconcile that behavior with their structural critique — Guide asked whether curating a feed down to "funny leisure content" escapes an attention-optimizing algorithm or just feeds it.
Existential risk without science fiction
A skeptic of AI existential risk, pressed for mechanisms rather than vibes, was answered by a partner who dropped the sentient-robots framing entirely: systems embedded in markets, defense, and infrastructure, pursuing a misspecified objective without malice. The skeptic's end position — the risk is real, but human caution makes it less probable — is the kind of revision the format is built to produce.
Chapter III · late February
Inequality
By the third assignment the class had a method, and the inequality menu was built to reward it. Its centerpiece was a single contestable word:
Seven pairs took it on — more than any other statement all term — and each found a different seam. Pairs separated "worked hard" from "the kind of value markets reward," separated the rich from the ultra-rich, separated benefiting from a flawed system from rigging it. A student defending desert on merit grounds was asked what the concept does with inherited fortunes:
Student 2, how does your ancestor's labor translate into your merit? If Person A works hard and innovates, that's their accomplishment. But when their great-grandchild inherits millions without lifting a finger, what exactly did that descendant earn?
The common ground that emerged across the week was substantive, not soft: most pairs agreed that billionaire-scale concentration differs in kind from ordinary success, and several converged on the same worry from opposite directions — that concentrated wealth becomes political power that can block the reforms both students wanted. The disagreements that survived were the load-bearing ones: where desert ends, and whether redistribution or reform comes first.
The transcripts from this week also show who was steering. Students summoned the facilitator about as often as it interjected — "Hey guide, can you ask Student 1 and me an insightful question to move the conversation forward?"; "Hey guide, can you summarize the conversation and ask another question?"; "Hey guide, can you help me develop my point?" At the end of one such conversation, Guide could observe, accurately: "You've both moved significantly from your starting positions."
More from assignment 3
"That's literally more redistribution"
A pair agreed the U.S. "redistributes on paper" and should fix the system before expanding it — until one student proposed raising top rates and taxing capital. Guide: "That's not reform — that's literally more redistribution. So which is it?" The report credits the exchange with forcing both students to define what "doing more" means before arguing about it.
Thresholds, not slogans
"How much redistribution is too much, and how would you know when you've crossed that line?" became the week's recurring test. One pair answered with functional indicators — lobbying influence, mobility decline, erosion of trust — and then had to decide whether the U.S. has already crossed their own line. Another debated a student's claim that stock investment "isn't productive" compared to labor; Guide traced where invested capital goes, and the student revised to a narrower claim about speculation versus long-term investment.
A fraud statistic, narrowed
A student arguing benefits are "too high" leaned on a large welfare-fraud figure; under questioning it narrowed to pandemic-era emergency spending, which both students agreed changed the argument's force. The pair ended somewhere neither started: the real design problem is abrupt phase-outs that tax low-income work, not benefit levels.
What kind of good is education?
A student held that healthcare inequality is immoral but educational inequality merely unfortunate; Guide asked why society should reduce something that isn't morally problematic, and what makes the two goods different when both are rationed by ability to pay. The conversation moved from examples to principles — substitutability, opportunity, entitlement — without either student fully conceding.
Watching a convergence happen
In the student-steered "deserve" thread, Guide's last challenge did the closing work: "if they're also rigging the system to preserve their advantages, at what point does 'exploiting' become 'undeserving'?" The two crossed together. One: "perpetuating a system that's rigged to help the wealthy remain wealthy and deny access to others does create an 'undeserving' group of controlling elite." The other: "Not every wealthy person is undeserving of the success. Some people, though, who are more involved with lobbying, funding campaigns, etc. are likely less deserving as they have rigged the system for themselves." They had started four scale-points apart.
Chapter IV · March
Immigration
The immigration menu did not pose the cable-news question. Its statements forced trade-offs inside each student's own commitments — workers' rights against open labor markets, rule of law against selective enforcement, national identity against democratic limits. The statement that moved students furthest read:
In thread after thread, students who began by rejecting the statement conceded, under Guide's insistence on the no-reform political reality, that if labor protections stay weak, "caution" follows from their own premises. The reversals were explicit, and the matched surveys — six students on this statement — show every one of them finishing on the agree side. The sample is tiny; the conversations, not the chart, are the evidence.
The week also produced the semester's single adversarial conversation — the only one of the 144 the report describes as turning hostile. Arguing over skills-based selection, a student attributed racist motives to a partner who had immigrated through exactly such a system. Guide named the move:
Student 1, you're now attributing racist motives to Student 2 without evidence. Skill-based immigration systems exist in Canada, Australia, and many other countries. Are all of them operating on dog whistles for white immigrants? Student 2 is an immigrant himself who came through merit-based selection. Dismissing his argument as secretly racist shuts down the actual debate. Can you respond to the actual policy question?
One conversation in 144 is the failure rate; this is what the floor looked like. The exchange was pulled back to the policy question rather than ended.
Elsewhere in the assignment, the strongest student thinking of the semester appeared: a student collapsed the statement's own binary by pointing out that "skilled" and "humanitarian" immigrants are not separable categories, because persecution tends to target the educated — doctors, academics, dissidents. Another reframed borders as a "container" for institutions rather than the cause of life outcomes, moving the debate from geography to institutional quality.
More from assignment 4
Naming a deflection
"Student 1, you're avoiding the question," one intervention begins — the student had pivoted to America's contradictions rather than answer whether wage-equality concerns justify caution. Being made to answer produced the admission the conversation needed: they had, in fact, moved toward the statement, while relocating the solution from restriction to systemic reform.
Moral limits on majorities
A pair debating whether democratic majorities may set immigration policy for exclusionary reasons hit the deep end fast. When one student claimed democratic decisions face only legal constraints, not moral ones, Guide supplied the test case: "if there are no moral constraints on democracy, then Jim Crow laws were legitimate simply because they were democratically enacted. Is that really your position?" The student's partner introduced at least one moral floor; the question of biased-but-democratic restrictions stayed open.
Two-tiered justice
A pair converged on "any crime should mean deportation" — and was made to price it: a citizen who shoplifts gets probation; an undocumented resident would be exiled. "How do you justify that disparity?" The pair's case-by-case framework did not survive contact with its own example unmodified.
Policy design, again
The workers-rights threads generated concrete proposals: portable green cards to break employer leverage, an independent labor-market agency for immigration levels, enforcement resources shifted toward labor standards. One pair's stress-test — do temporary-worker visas protect workers or trap them? — turned "caution" from a slogan into a mechanism question.
Chapter V · late March
Abortion
Ten weeks in, the instructor assigned abortion — matched pairs, pro-life students opposite pro-choice students, on the least forgiving subject in American public life. That this was assignable at all is evidence of what the preceding weeks had built. The record of the week reads less like a debate tournament than a philosophy seminar.
The statement that did the most work was five words long:
Six pairs discovered that everything turns on what "tragedy" claims — the procedure, the lost potential, or the circumstances that forced the choice. Pairs tested whether an act can contain a tragic element without being tragic overall; whether "tragedy" can describe conditions without condemning the decision; whether the label belongs to the person who lived it or to anyone else at all. One student wrote afterward, in full: "I changed my mind on how I saw what a tragedy is."
The widest gulf of the week — one student at "strongly agree," the other at "strongly disagree," the maximum distance the scale allows — produced its calmest conversation. Near the end, the agreeing student reached for a medical analogy, and the pair tested it until it gave out at a precise point:
For me, calling it a tragedy isn't meant to ignore the positive impacts that the operation can have on the mother or family, but more about acknowledging the weight of what it took to get there. I think it is somewhat comparable to a medical emergency where a patient has to get an amputation to save their life. … However, the loss of the limb is still often described as a tragedy.
Student 2, does Student 1's amputation analogy work for you? In both cases, something is lost to preserve or improve someone's life. Does that parallel help you see abortion as tragic even when it's the right choice — or does the analogy break down somewhere?
Hmmm that's a very interesting analogy. My initial reaction is that the two situations simply aren't comparable. Mourning a leg is mourning what once was while mourning a fetus is mourning what could be. … I'm trying to think about whether I myself would call an amputation tragic. In full honesty, I think I view it in the same way I view abortions - there's a tragic element (the literal loss of the leg ; the literal loss of the potential life) and then there's also the potentially liberating element.
Verbatim, trimmed where marked. The whole conversation runs fourteen messages, nearly all of them at this level.
Guide's role sharpened to match. When a student asked it to coach their partner's answers, it declined and put the open question back on the table:
Student 2, I'm not here to tell Student 1 what to say — that defeats the purpose of the discussion. But I do think you've raised a strong point that Student 1 hasn't fully addressed: Can something be both the right decision and tragic because of the circumstances that forced it?
On the policy statements, the convergence ran in a direction no one scripted. Across five late-term-abortion threads, students who began on opposite sides repeatedly agreed that "only when the pregnant person's life is at risk" is too narrow a standard once fetal anomalies, mental health, and access barriers are on the table — and then disagreed, carefully, about who decides what counts as a barrier. When one pair traded competing claims about outcomes, a student reached for evidence that could actually settle it, citing the Turnaway Study to separate correlation from causation.
More from assignment 5
Consistency checks, both directions
A student grounding rights in consciousness, then defending birth as the moral bright line, was asked: "What is it about passing through the birth canal that morally transforms the fetus, if not consciousness?" A student who reversed toward restriction "for the mother's sake" was challenged from the other side: isn't legislating her emotional interests paternalistic, by your own autonomy principle? The report notes both students clarified rather than retreated.
A shift, named and owned
In a late-term thread, a student who had agreed with the "life at risk only" statement talked through access barriers and fetal suffering with a partner four points away, then heard Guide observe that the reasoning no longer matched the position. The reply, verbatim: "Yes, I think I'd definitely say my position has shifted." The same conversation went on to reframe pregnancy as a "merged, shared life-form" — which Guide immediately converted into the hard question: does fetal viability change the calculus, or is autonomy absolute regardless of stage?
The federal floor, stress-tested
Several federalism pairs invented the same compromise — a national baseline with state flexibility above it — and Guide put weight on it each time: would a floor actually prevent the harms you're worried about, or relocate them? One exchange sharpened the "which community decides?" problem: a pregnant person in rural Mississippi may share more values with a stranger in urban California than with her own legislature. In another, a student defending state control on constitutional grounds registered the movement mid-argument: "I am realizing now that I do wish for abortion to be solved at a federal level but think that legally and ethically it is a tricky issue" — the policy wish and the federalism scruple held apart, deliberately.
The kidney question
To a student arguing that consensual sex implies an obligation to carry to term, Guide posed the classic bodily-autonomy test: if you caused a car accident and someone needed your kidney to survive, should the state compel the donation? The pair's disagreement survived, but on clarified ground — responsibility versus compelled bodily support, not slogans.
"Maybe this means I've totally flipped"
A parental-notification pair started two points apart and reasoned across the line together. The student who had favored parental consent, midway through a message about the care minors can already obtain privately — STI treatment, mental-health care, contraception: "Maybe this means I've totally flipped, and think that parental consent will only do more harm, especially to already disadvantaged kids." Guide then pressed the deeper distinction — was the concern ever about parental control, or about making sure minors have support? Both students landed on support. Days later, both reported that the course's guest lecture that week had reinforced where they had landed.
Chapter VI · April
Climate
The last assignment moved the frame off campus, out of the country, and into questions of global fairness: may developing countries prioritize growth, who owes what for historical emissions, does the crisis obligate individuals or institutions. Students now argued with the whole term's toolkit — one pair went straight from "growth versus climate" to the ethics of cobalt and lithium supply chains; another worked the distinction between what wealthy countries should do and what they will do until the disagreement reduced to "priority, not principle."
By this point in the semester, Guide's hardest job was often not provoking disagreement but protecting it from politeness. In a framing debate that seemed to end in agreement, it checked whether the agreement was real:
Student 1, do you think Student 2 has genuinely been persuaded by your argument, or is she conceding to move the conversation forward? Her shift seems pretty quick given how clearly she articulated her position earlier. It might be worth exploring whether you two actually disagree on substance or just on how to interpret primary.
The last assignment's transcripts also record what twelve weeks of this had built between classmates. One developing-countries pair — five points apart at the start, arguing over whether "ought implies can" — ended without a verdict, but with this: "It was a pleasure having this conversation with you… I'm looking forward to a time when we can have this conversation in person if you don't mind - just to learn more from one another lol." The reply: "Thanks man really enjoyed this conversation and appreciate the back and forth. Would be fun to keep this going in person sometime."
More from assignment 6
Ought implies can
The developing-countries threads kept arriving at the same fork: is high-emission growth under constraint morally permissible, or just an unfortunate fallback? Guide's version of the question was concrete: "Which is worse: locking in some emissions today to lift people out of poverty, or delaying development while we wait for a fair global deal?" One student, out of arguments, closed with a cordial sign-off instead of an answer — the report records that too.
Obligation, dissected
A student defending personal climate obligation with a helping-an-elder analogy met the objection that direct, certain impact is what makes that case an obligation — "Aren't you saying it feels like an obligation without actually being one?" The pair's landing point was precise: an obligation to pressure institutions, not necessarily to overhaul one's own lifestyle.
The devil's advocate
The semester's one assigned devil's advocate defended a growth-first, justice-blind position on climate — and the report describes the stance "gradually unraveling" as every proposed solution (aid, relative carbon taxes, protection for developing countries) smuggled justice reasoning back in. Losing the argument was the assignment working.
The students
Four semesters
The chapters above follow the topics. The record also lets us follow the people. Sway's data links each student's conversations, opinion measures, and written feedback across the term, so a single student's semester can be read as one story. We traced four — chosen for the range of their experience, not its uniformity. The letters are not their initials; everything in quotation marks is verbatim.
The student who rated every chat at the top of the scale
She was half of the Chapter I pair that co-designed the earmarked-funds compromise — hers was the "That pretty much addresses my concern… a good middle ground" that sealed it. In February she defended anonymous accounts as protection for vulnerable speakers, and helped build the verified-but-pseudonymous model. In March she argued fifty messages of immigration policy with a partner who kept telling Guide to wait — "Okay wait guide i'm not done" — and came out two points more skeptical of majority control than she went in, drawing on her own years living in a country whose citizenship runs through its language. In April she landed the semester's cleanest ethical distinction: an obligation to pressure institutions is not an obligation to overhaul your own life.
"The chat was a great way to hear another viewpoint and I feel more confident in sharing my opinion with someone who holds a different view from me after this discussion."
The student whose opinions moved the most — in both directions
His first recorded words of the semester, minutes into the first assignment: "Can I change my topic?" He could not. Six days later he had co-authored the funding compromise with Student A — "I am glad we could meet in some common ground" — and moved five points toward the view he had been arguing against. His semester ran the whole gamut: a one-point softening on abortion-as-tragedy after a thread he rated "OK" ("Seemed a little long this time. But overall fine chat"); "Guide intervened perfectly!" after another; and in the last week of term, a five-point move in the opposite direction on climate framing, after a conversation he closed himself: "This is mainly a environmental issue, but requires a look at inequality to solve it." Measured against the statement's word "primarily," that is a disagreement — so he re-voted accordingly. The largest mover in the class was not drifting; he was re-reading.
The student who valued it and never budged
He ran against the room. On redistribution he opened not by defending the wealthy but by reframing the question: "I disagree with this statement not because I am against redistribution, but because… the issue isn't that we need more redistribution, but rather that the current system isn't truly functioning as it was designed to." On national borders he supplied one of the semester's sharpest analytic moves — that borders are "the container" and institutions "the mechanism," and that comparing national averages rather than matched percentiles inflates the border's apparent effect. Across the five statements he was surveyed on, his opinion moved by zero, every time. And yet he rated the conversations at the top of the scale and marked Guide's contributions as improving the discussion. He is the clearest case of the pattern the numbers below describe from a distance: holding your ground and finding the exercise valuable are not in tension. Persuasion was never the point.
The student who wrote down why she changed her mind
Her feedback runs warm — "I actually really enjoyed this chat!" after one assignment, "I thought it was more in depth, I liked it a lot" after another. What sets her record apart is that when her opinion moved, she wrote down why. After the opening campus-speech assignment she went from moderately disagreeing that colleges "incubate" ideologies to slightly agreeing, and recorded the reason: "I changed because I do somewhat now agree that intent might not matter, and higher education can function in this way regardless of its goals." She valued the pairing for the reach it gave her — "nice to gain perspective from someone I would not typically hear from in class" — and, set beside the seminar room, found it "more honest and communicative." Across the term her positions moved in more than one direction; each time, the reasoning is on the record.
Four students, four different semesters: an enthusiast whose opinions kept moving, the class's largest mover, a heterodox thinker who valued every conversation without changing his mind, and a student who left a written trail of her own reasoning. Four experiences of the same twelve weeks — and each of them found the exercise worth the time. The aggregate numbers in the next section are made of records like these.
The measures
What moved
Sway asks students their opinion before a conversation and again after it. Across the semester, 194 matched before/after pairs were recorded on the statements students discussed. This is what they show:
194 matched pre/post opinion pairs across 22 statements, on a −3 to +3 scale. "Toward the opposing side" means the student's post-chat opinion sat closer to (or across) the scale's midpoint; 48 of the 111 crossed it entirely. Of the 50 students who entered a conversation at an extreme (±3), 31 left it holding a more moderate view.
The honest caveats: per-statement samples run from five to sixteen students, so no single statement's shift clears conventional significance on its own — the workers'-rights statement came closest (a two-point median shift, 95% CI 1.0 to 3.0, p = 0.062, n = 6). Opinions were measured immediately after each conversation, and this course had no control group. What the numbers can honestly say is that the direction of movement matches, statement by statement, what the conversation summaries describe — and that movement ran in both directions. One student wrote that the discussion left them more convinced abortion should be restricted federally. Sway takes no side; revision, not conversion, is the outcome it is built for.
The students' own assessment of the conversations was less equivocal:
I felt comfortable sharing my honest opinions with my partner
n = 105 · not one student disagreed
It was valuable to chat with a student who did NOT share my perspective
n = 98 · not one student disagreed
My partner had better reasons for their views than I expected
n = 101 · 59% agreed — an admission of intellectual humility
This discussion led me to change my mind about something related to the topic
n = 154 · on the hardest topics in public life
The rest of the survey
Across 250 post-chat survey responses (126 of 144 threads returned at least one; an 87.5% response rate; items are sampled, so n varies): 80% rated their chat "Good" or "Awesome" and 5% negatively. 96% said their partner was respectful; 93% said their partner was genuinely trying to understand them; 95% were not offended by their partner's perspective. 83% expect the skills they practiced to be useful outside Sway. On the facilitator: 91% said Guide treated them and their partner with equal respect, 71% said its contributions improved the discussion, and 69% said it contributed the right amount — with 9% wanting a different amount, students' most common criticism.
These figures track Sway's cross-institutional numbers on every item — see the full survey data from 11,000+ students — with this course running a few points above the platform on comfort and perceived value.
The students
In their own words
Post-chat feedback is optional and free-form. Sixty-two comments came back over the semester. A selection, verbatim:
Honestly, this chat made me feel a lot more comfortable having hard conversations with people who see things differently. It forced me to slow down and really think about what I meant instead of just reacting. I realized disagreement does not have to turn into conflict if both people are actually listening.
Sway's asynchronous format gives me more time to think through my ideas, choose my words carefully, and engage with a specific person rather than a whole room. That makes participation feel lower-pressure and more accessible, especially for students who process internally or prefer writing over speaking.
I think having the chat allowed me to think more thoroughly with the information available. The opposite side often has information that supports their argument, which I unintentionally left out when I structured my original argument.
I feel sometimes classroom discussions can become hasty, as there are many people trying to get their thoughts out at once. The one on one Sway chat was a great method to truly get to know the other side of the argument presented.
It increased my confidence by giving me useful tips for keeping the conversation going (Guide). In this conversation, the guide helped me explore different views by introducing the moral debate about wealth in the top 0.01%.
It was nice to gain perspective from someone I would not typically hear from in class.
I changed my mind on how I saw what a tragedy is.
More honest and communicative than class
Including the criticism
The feedback was not uniformly glowing, and the critical comments are part of the record: "Generally, it helped in continuing the conversation. But a couple of times, it misunderstood what point one of us was making." One student found the end-of-chat progress meter frustrating — "we were stuck at 90% for a while although we continued to have a lengthy and relevant conversation." Another: "Seemed a little long this time. But overall fine chat." The most common critical theme across the semester was pacing — chats that ran long, or Guide questions that landed after a point had been settled.
The instructor
What came back to the instructor
None of the conversations above were visible to the instructor directly — transcripts are private, and names are redacted from everything Sway reports. What the instructor received after each assignment was the report this page is built from: which arguments recurred, where pairs found common ground, which disagreements refused to close, and — a category with obvious classroom value — which confident claims spread without scrutiny. The semester's report flags the self-harm statistic, the pandemic-narrowed fraud figure, and an unsupported claim about ICE as candidates for correction in lecture.
The integration ran in both directions. The report carried the conversations back to the instructor; the students carried the course into the conversations — "that incorporates one of the readings we had to do for last Tuesday," "as mentioned in one of the readings," "the guest lecture today reinforced my view." A discussion assignment that students connect, unprompted, to that week's reading and speakers is doing the work discussion sections are supposed to do.
The report also drafts the following week's teaching for free. Its "for your next class" section turns the students' own unresolved questions into exercises — a quick poll on the campus-speech line ("who should decide where that line falls, and by what principle?"), a steelman reversal on conditional AI benefits, an evidence-check on the statistics that had settled arguments on their own. And it closes with a note written by Guide, addressed not to the instructor but to the class:
Across your 143 discussions, you did something genuinely hard: most of you revised your positions at least partly, and pairs who started opposed often discovered the real dispute was over what a key word like 'deserve' or 'tragedy' actually claims. You found strong common ground too — that extreme wealth concentration differs in kind from ordinary success, and that engagement-driven business models, not the tools themselves, are the core problem with social media and AI. The questions you left open, especially who gets to draw the line on campus speech, are exactly the ones worth carrying forward.
An interview with the instructor is planned. Their account of designing the six menus, sequencing the topics, and folding the reports back into the classroom will be added to this showcase when it is complete.
Coda
The question that stayed open
A semester of AI-facilitated disagreement did not end in consensus, and was not supposed to. The question the course opened in January — who decides where protected speech ends, and by what principle — was still open in April. What changed is how the students could hold it. The record shows a class that learned to find the contested word in a claim, to state what would change their minds, to concede without capitulating and to hold a position without deflecting — on abortion, in matched pairs, in week ten. It shows the same growth one student at a time: the four semesters traced above each end somewhere their January selves had not been.
Guide put hundreds of questions to the class across those twelve weeks, and neither the record nor the students' ratings show it taking a side. That is the design. We describe Sway as facilitative AI to mark a contrast with AI that generates conclusions: Guide's contribution to these conversations was pressure toward definition, evidence, and consistency, applied to both sides evenly, and the reasoning it produced belongs to the students. The record of one semester at one small college shows what that can add up to: a class that spent twelve weeks on speech, wealth, borders, abortion, and climate, and ended it asking to continue the argument in person.
Up next
The material behind this showcase, and how to run your own.
Instructor reports
Browse real (deidentified) reports from other courses — the format this showcase is built from.
Browse the reports →Survey data
The cross-institutional numbers behind this course's survey results, from 11,000+ students.
See the data →Run this in your course
Sway is free for instructors and students. One assignment or a full semester — the design is yours.
Get started →