What a University Is For大学:让人长出自己的路大學:讓人長出自己的路 · 06
A Reading Is Not Yet a Verdict
一个读数,还不是一个判决
一個讀數,還不是一個判決
Scores, detectors, recommendations, and degrees change people's routes. The question is who explains them, who can review them, and how far their assurances actually reach.
分数、检测、推荐和文凭都能改变人的路。要问的不是有没有评价,而是谁解释它,谁能复核,以及这一次究竟证到了哪里。
分數、檢測、推薦和文憑都能改變人的路。要問的不是有沒有評價,而是誰解釋它,誰能複核,以及這一次究竟證到了哪裡。
Imagine an assignment flagged as highly likely to contain AI-generated writing. The percentage looks like a finished sentence: the student did not write it. But what has actually been observed? Who turns this indication into an allegation, under which requirements, and through what decision about the course result?
Melbourne's staff guidance does not treat a high detector score alone as sufficient grounds for an academic-misconduct allegation. Further evidence is needed, potentially including drafts, notes, or discussion of how the work developed. This specifies an evidential limit. It does not decide any particular student's case.
Even an accurate reading would not complete everything. Using a tool does not automatically establish absence of judgment. A machine-produced number cannot relieve its adopter of choosing applicable requirements, considering evidence, receiving responses, and making the conclusion. Those actions remain actions, not additional decimal places supplied by the detector.
Assessment needs explanation because it changes routes. Grades, degree decisions, references, and decisions using assessment to discipline or fund can open and close opportunities. Calling this power does not accuse assessors of bad intentions. It identifies effects requiring grounds and checks, including when the assessor means well.
An adviser can genuinely support a student while holding consequential assessment power. Sincerity does not remove positional pressure. Checks are not intended to eliminate assessment but to prevent a necessary judgment from acquiring control over every challenge to it. Support and power can be simultaneously present without being identical.
Procedures should follow actual effects. Ungraded feedback and a qualification deciding graduation need not carry equally heavy machinery. But a short remark affecting funding, a degree, or departure cannot be made light merely by tone or a small number of recipients. Limited reach can still be deep.
Formats alone decide little. MIT begins first-year grading with Pass/No Record; Brown does not calculate undergraduate GPAs while retaining course-assessment options. These are bounded arrangements, not proof that evaluation has disappeared or stated all its limits. Computing a GPA does not automatically establish a defective institution either.
Ask whether meeting an indicator is recorded as replacing judgment of the work. A rubric can help. If challenged about unreliable material, an assessor who only cites the rubric omits its application. Explaining which material fails and why, while remaining open to objection, uses a rule without making it an exemption.
Narrative is no automatic escape. Hampshire's account of narrative evaluation also calls for criteria at the beginning of a learning activity. Words do not remove the measure. They may provide another way to state it and its reasons. A paragraph needs accountability as surely as a score does.
An indicator made the sole, unchangeable target can encourage satisfying the display while missing the work it was meant to show. Can is not always. Actual effects need evidence. Rejecting omitted judgment does not establish numbers as inherently corrupt or prose as inherently truthful. Both formats can be handled responsibly.
Some consequential conclusions require separation of people. Degree and qualification decisions should be made from a position separate from the adviser. Advisers may submit opinions that students can see and answer; assessment and review must not depend solely on them. The reason concerns dependency, not an assumption of professional incompetence.
Ordinary courses need not all acquire separate graders. Instructors can teach and assess, while the university explains how checks address the pressure of those combined roles. Transcript grades need reasons and an objection route outside the instructor. Altering an academic conclusion still requires relevant competence and responsibility, not merely a different office.
When the instructor is also the student's adviser, both relationships matter. Changing hats does not remove pressure. Everyday feedback can have lighter arrangements, but the consequences of a formal grade cannot be reduced by calling it only a classroom matter. Checks should follow the actual path of consequences.
Discipline needs another separation: accusing and concluding should not occupy the same position. A zero for alleged misconduct concerns an act, not simply poor writing. A teacher can raise the matter; another position should make or confirm the disciplinary conclusion. One numerical result should not conceal which kind of decision occurred.
References have two ends. Writers answer for grounds and for any use of unrelated obedience as a condition. Adopters answer for use, weight, and alternative sources. A famous writer does not guarantee a sound letter. Questionability does not entitle an applicant to rewrite it into praise. Evaluation can remain unfavorable and still be answerable.
Automating assessment is another adoption. Why this system, how does a reading become a consequence, what review exists, and when can use stop? A machine's output is not completed judgment, and a human confirmation click cannot establish it either. Responsibility does not travel away with the calculation.
Assessment also tells everyone what the university rewards. Suppose verified student data work is credited entirely to an adviser, or extra output obtained by threatening unrelated opportunities is celebrated as leadership. In its own contribution to the outcome, assessment rewards an already identified use of people as instruments.
A university cannot guarantee elimination of all external benefits from such conduct. It can refuse to add its own reward. That principle needs documented acts and contributions, not speculation about inward purity. Rank cannot supply the missing explanation of who performed the work or how it was obtained.
A degree should therefore be a bounded record. What work was verified, what material and criteria supported the assessment, and what remains outside its guarantee? Particular competence can be recognized seriously. It cannot certify sound judgment in every future matter simply because a university has attached its name.
An employer's use introduces another decision. The university answers for its record; the user for its adoption. Facts about work, an assessor's judgment, and later effects should remain distinct. Even an honest record can be strategically satisfied. A well-designed form cannot guarantee removal of that possibility.
The UK's 2022 External Examining Principles describe independent expert input while expressly stating that the principles are not mandatory under UK regulatory systems. Actual powers depend on appointment and institutional rules. Externality can matter without a title certifying authority to reconsider every conclusion or proving how the arrangement works in practice.
Faculty assessment belongs in the same inquiry. DORA opposes using journal measures as proxies for individual research quality. CoARA combines peer-centered qualitative assessment with responsible quantitative use. These are normative documents, not evidence of implementation everywhere. Neither supplies a reason to abolish all indicators or to stop examining their uses.
Interdisciplinary work particularly resists substitution of one field's score for judgment of the whole. Competent reviewers and the meeting of criteria require arrangements. Student and faculty evaluation can share mechanisms without sharing identical duties or consequences. Drawing the comparison should not merge the positions being compared.
Trust and checking need not be alternatives. Princeton's 2026 arrangements introduce proctors for in-class examinations while retaining the Honor Code and student Honor Committee's adjudication. The proctoring document limits observation: suspicion must not interrupt the examination, and records go to the relevant process. Observation is not the conclusion itself.
Those documents do not certify a complete procedure. How a student answers evidence during adjudication requires other records. Removing the unproctored format should not remove standing to respond. Adding checking need not automatically mean treating everybody as untrustworthy. Particular rules, powers, and opportunities to answer still matter.
Checking should not target people simply by identity or turn a previous check into a lasting identity. It may be general or event-based, with questionability of its rules. Readings must not become unanswerable verdicts. Checking has power and needs its own limits; it need not first be renamed assessment to deserve them.
The checking arrangements are not final courts beyond inquiry. Committees, second readers, and external experts can err and must answer for their own decisions. Their role examining others cannot confer permanent immunity from examination. Completion of one case need not end every later question about the arrangement.
No greater accuracy is guaranteed merely by installing review. No grading system wins here, and no student is declared to have cheated. The demand restores work compressed into a number: its evidential meaning, the judgment made from it, the receiving address for objections, and the reach of each conclusion.
The percentage may remain on screen, and a grade may eventually be necessary. The aim is not to erase numbers but to keep people present: adopters still judge, those assessed can respond, and a finished conclusion remembers that it completed this occasion within this scope, not every possible question about a person.
设想一篇学生作业被系统标成高度疑似AI生成。屏幕上的数很醒目,读起来像一句已经说完的话:他没有自己写。可一个读数,到底说明了什么?谁要把这份提示变成指控,又凭什么让它决定这门课的结果?
墨尔本大学给教职员的说明,明确不把高AI检测分数本身当成提出学术不端指控的充分根据,要求另外的证据。可以找草稿、笔记,也可以谈作品怎样形成。这份文本说的是工具的证据限度,不是对任何一个学生的判决。
即使读数准确,事情仍未全部完成。用过工具,不自动等于没有判断;一个数来自机器,也不使采用它的人免去选择:适用什么要求,怎样处理证据,允许怎样回应,最后由谁作结论。
评价之所以需要这些说明,是因为它会改变人的路。一门课的分数、学位资格、推荐信,或采用评价作出的处分与资助决定,都可以打开或关上一些机会。把这种作用叫权力,不是在指控评价者有恶意。
一位导师真心支持学生,同时握着能决定其去留的评价权,这两件事可以都是真的。心好不能消除位置上的压力。制衡也不是要把评价减少到没有,而是让需要的判断,不能顺带取得控制一切追问的权限。
程序应当随实际作用安排。一次不计成绩的练习反馈,与决定能否拿学位的资格考,不必使用同样重的流程。但一句话若实际卡住学位、资金或退出,不能因为它说得轻、只经过几个人,就把作用也说轻。
首先别凭形态判。MIT第一年开始的一段时间采用通过或无记录,布朗不计算学生GPA,同时仍有课程评价选项。这些安排各有范围;不计绩点,不证明没有考核,也不证明评价已说明所有限度。计绩点本身,同样不能直接定罪。
真正要看的,是某个指标有没有被记成对工作本身的判断已经完成。一张写作评分表可以很有帮助;如果学生问材料为什么不可靠,答复只有“按细则扣分”,细则就代替了具体说明。若能说清用法,也能受问,按表评分本身没有错。
文字评价也不天然更好。汉普郡学院介绍叙述式评价时,同样要求在学习开始时确定评价标准。换成一段文字,不等于离开了尺子;它可能只是把尺子与理由换一种方式写出来,仍需要由人负责。
指标成为唯一、不可改动的目标,可能诱使人学会表格想看的一切,而绕开原来想评的工作。只是可能,不是每一处都如此;效应需要按事实查。反对判断被省去,不等于宣布数字必坏、叙述必真。
有些结论需要人与人分开。学位与资格这类决定,应由与导师分开的评定位置作出;导师可以提出意见,学生能看到并回应,评定与复核不由导师一人控制。这是对深依赖的安排,不是宣布导师没有专业能力。
一门普通课程则不必因此全部另聘评分人。任课者可以教,也可以评;大学仍应说明怎样处理同一人给条件又考核的压力。计入成绩单的分数,需要有根据,异议与复核由任课者以外的位置受理。改学术结论,还须具备相应能力与职责。
导师兼任课,两处作用要一起看,不能换一顶帽子就卸掉原关系的压力。日常反馈可以轻一些,正式成绩的后果却不能只靠“这只是一次课堂评价”缩小。制衡跟着实际的路走,不跟着称谓走。
处分还有另一道分开:提出指控的人,不兼作出处分结论的人。因涉嫌不端给作业零分,采用的是对行为的评价,不只是这篇文章写得不好;教师可以提出,但结论应由另一位置作出或确认。不同动作不能靠同一个零分混过去。
推荐信也有两端。写信的人根据什么评价,是否借别处的服从决定推荐,值得问;用信的人为什么采用,给多少分量,有无替代来源,也值得问。信由名人写,不等于其中每一判断可靠。可追问,不使申请人获得改写成好话的权力。
用自动化评价,也是作了一次采用。为什么选它,怎样把读数变成结果,有没有复核,何时停止,都需要有人回答。“机器读出来了”不等于判断完成,人最后按了确认也不够。责任不会随着计算被一起交出去。
评价还会告诉所有人,这所大学奖励什么。若已经查明资料由学生整理,却把功劳全记给导师;若靠卡住成员别的机会逼出额外劳动,却把增加的数字记成带队有方,评价就在自己这一端,再奖励了一次把人当手段的动作。
大学不能保证这样的动作在外面从此没有好处。它能负责的是自己的评价与资源不再添上这一份。原则应落实在可核的贡献与做法上,不靠猜一个人内心是否纯洁,也不把位置高低当成功劳已经归谁的根据。
学位因此应是一份有限的记录。它说明核实过什么工作,用哪些材料和标准作了怎样的评价,证到哪里,哪里没有保证。某项任务确实完成,可以认真确认;它不担保这个人在全部未来事项上都会判断得好。
这份纸到了雇主手里,又有新的采用。大学对自己的记录作答,用它筛人的一方对自己的用法作答。工作的事实、评价者的判断、后来赋予它的效力,应当分开。即使记录如实,也仍可能有人学会专门凑它要求的几样,这个限度不能被一张好表格消除。
英国2022年的外部考评原则,把独立专家意见写在学位评价一侧,同时明确它们不是英国各监管体系下的强制要求。外部意见能做什么,要放回具体聘任与校内规则看;有一个外部头衔,不等于已经有人能独立改判一切。
同一套问题也落到教授身上。DORA反对拿期刊指标替代对单篇研究的质量判断,CoARA把同行的定性评价与负责任的量化使用放在一起。这是已有的规范文本,不证明任何一所大学已照它运行,也不是凡指标都该退出。
跨学科的工作尤其不能只用一边的分数,替整项工作作答。谁有能力评、怎样让几种标准相遇,需要另外安排。老师与学生的评价可以并读其机制,职责与后果却不能因此混成同一种。
信任与核查也不是只能选一边。普林斯顿2026年的监考安排要求课堂考试有人监考,同时保留荣誉制度与学生荣誉委员会处理涉嫌违规的职责。其监考文件限定监考只作观察,发现可疑也不在现场打断考试;记录要交给相应程序。
这里只能据文本说这些变化,不能认证整个程序已经完备。学生如何在裁断阶段回应证据,仍要看另一些规则。撤去无人监考这种信任,不应顺带撤掉学生的申辩地位;增加核查,也不自动就是把所有人当作不可信。
核查要避免按一种身份预先锁定人,或把查过一次变成终身标签。可以普遍查,也可以按具体事件查;规则本身要能问,观察和读数不能直接变成不可回应的判词。核查有自己的权力,需要自己的限制,不必改名成评价才值得制衡。
这些制衡也不是新的终点。委员会、第二评阅、外部专家都可能判断失当,也都应就自己的那一判作答。不能因为他们专门审别人,就取得别人永远不能审他们的地位。
这篇不许诺设了复核就更准确,不替某种评分制度宣布胜利,也不判那个学生有没有作弊。它先让一项本来被压缩的工作重新展开:读数是什么证据,这次评价怎样完成,谁能受理疑问,各自真正说明到哪里。
屏幕上的百分比可能还在那里,成绩也终究需要作出。重要的不是把所有数抹掉,而是不要让人消失:采用工具的人仍要判断,受到评价的人仍能回应,结论作完以后,也仍记得它只完成了这一回、这一范围里的事。
設想一篇學生作業被系統標成高度疑似AI生成。螢幕上的數很醒目,讀起來像一句已經說完的話:他沒有自己寫。可一個讀數,到底說明了什麼?誰要把這份提示變成指控,又憑什麼讓它決定這門課的結果?
墨爾本大學給教職員的說明,明確不把高AI檢測分數本身當成提出學術不端指控的充分根據,要求另外的證據。可以找草稿、筆記,也可以談作品怎樣形成。這份文本說的是工具的證據限度,不是對任何一個學生的判決。
即使讀數準確,事情仍未全部完成。用過工具,不自動等於沒有判斷;一個數來自機器,也不使採用它的人免去選擇:適用什麼要求,怎樣處理證據,允許怎樣回應,最後由誰作結論。
評價之所以需要這些說明,是因為它會改變人的路。一門課的分數、學位資格、推薦信,或採用評價作出的處分與資助決定,都可以打開或關上一些機會。把這種作用叫權力,不是在指控評價者有惡意。
一位導師真心支持學生,同時握著能決定其去留的評價權,這兩件事可以都是真的。心好不能消除位置上的壓力。制衡也不是要把評價減少到沒有,而是讓需要的判斷,不能順帶取得控制一切追問的權限。
程序應當隨實際作用安排。一次不計成績的練習回饋,與決定能否拿學位的資格考,不必使用同樣重的流程。但一句話若實際卡住學位、資金或退出,不能因為它說得輕、只經過幾個人,就把作用也說輕。
首先別憑形態判。MIT第一年開始的一段時間採用通過或無記錄,布朗不計算學生GPA,同時仍有課程評價選項。這些安排各有範圍;不計績點,不證明沒有考核,也不證明評價已說明所有限度。計績點本身,同樣不能直接定罪。
真正要看的,是某個指標有沒有被記成對工作本身的判斷已經完成。一張寫作評分表可以很有幫助;如果學生問材料為什麼不可靠,答覆只有「按細則扣分」,細則就代替了具體說明。若能說清用法,也能受問,按表評分本身沒有錯。
文字評價也不天然更好。漢普郡學院介紹敘述式評價時,同樣要求在學習開始時確定評價標準。換成一段文字,不等於離開了尺子;它可能只是把尺子與理由換一種方式寫出來,仍需要由人負責。
指標成為唯一、不可改動的目標,可能誘使人學會表格想看的一切,而繞開原來想評的工作。只是可能,不是每一處都如此;效應需要按事實查。反對判斷被省去,不等於宣佈數字必壞、敘述必真。
有些結論需要人與人分開。學位與資格這類決定,應由與導師分開的評定位置作出;導師可以提出意見,學生能看到並回應,評定與複核不由導師一人控制。這是對深依賴的安排,不是宣佈導師沒有專業能力。
一門普通課程則不必因此全部另聘評分人。任課者可以教,也可以評;大學仍應說明怎樣處理同一人給條件又考核的壓力。計入成績單的分數,需要有根據,異議與複核由任課者以外的位置受理。改學術結論,還須具備相應能力與職責。
導師兼任課,兩處作用要一起看,不能換一頂帽子就卸掉原關係的壓力。日常回饋可以輕一些,正式成績的後果卻不能只靠「這只是一次課堂評價」縮小。制衡跟著實際的路走,不跟著稱謂走。
處分還有另一道分開:提出指控的人,不兼作出處分結論的人。因涉嫌不端給作業零分,採用的是對行為的評價,不只是這篇文章寫得不好;教師可以提出,但結論應由另一位置作出或確認。不同動作不能靠同一個零分混過去。
推薦信也有兩端。寫信的人根據什麼評價,是否借別處的服從決定推薦,值得問;用信的人為什麼採用,給多少分量,有無替代來源,也值得問。信由名人寫,不等於其中每一判斷可靠。可追問,不使申請人獲得改寫成好話的權力。
用自動化評價,也是作了一次採用。為什麼選它,怎樣把讀數變成結果,有沒有複核,何時停止,都需要有人回答。「機器讀出來了」不等於判斷完成,人最後按了確認也不夠。責任不會隨著計算被一起交出去。
評價還會告訴所有人,這所大學獎勵什麼。若已經查明資料由學生整理,卻把功勞全記給導師;若靠卡住成員別的機會逼出額外勞動,卻把增加的數字記成帶隊有方,評價就在自己這一端,再獎勵了一次把人當手段的動作。
大學不能保證這樣的動作在外面從此沒有好處。它能負責的是自己的評價與資源不再添上這一份。原則應落實在可核的貢獻與做法上,不靠猜一個人內心是否純潔,也不把位置高低當成功勞已經歸誰的根據。
學位因此應是一份有限的記錄。它說明核實過什麼工作,用哪些材料和標準作了怎樣的評價,證到哪裡,哪裡沒有保證。某項任務確實完成,可以認真確認;它不擔保這個人在全部未來事項上都會判斷得好。
這份紙到了雇主手裡,又有新的採用。大學對自己的記錄作答,用它篩人的一方對自己的用法作答。工作的事實、評價者的判斷、後來賦予它的效力,應當分開。即使記錄如實,也仍可能有人學會專門湊它要求的幾樣,這個限度不能被一張好表格消除。
英國2022年的外部考評原則,把獨立專家意見寫在學位評價一側,同時明確它們不是英國各監管體系下的強制要求。外部意見能做什麼,要放回具體聘任與校內規則看;有一個外部頭銜,不等於已經有人能獨立改判一切。
同一套問題也落到教授身上。DORA反對拿期刊指標替代對單篇研究的品質判斷,CoARA把同行的定性評價與負責任的量化使用放在一起。這是已有的規範文本,不證明任何一所大學已照它運行,也不是凡指標都該退出。
跨學科的工作尤其不能只用一邊的分數,替整項工作作答。誰有能力評、怎樣讓幾種標準相遇,需要另外安排。老師與學生的評價可以並讀其機制,職責與後果卻不能因此混成同一種。
信任與核查也不是只能選一邊。普林斯頓2026年的監考安排要求課堂考試有人監考,同時保留榮譽制度與學生榮譽委員會處理涉嫌違規的職責。其監考文件限定監考只作觀察,發現可疑也不在現場打斷考試;記錄要交給相應程序。
這裡只能據文本說這些變化,不能認證整個程序已經完備。學生如何在裁斷階段回應證據,仍要看另一些規則。撤去無人監考這種信任,不應順帶撤掉學生的申辯地位;增加核查,也不自動就是把所有人當作不可信。
核查要避免按一種身份預先鎖定人,或把查過一次變成終身標籤。可以普遍查,也可以按具體事件查;規則本身要能問,觀察和讀數不能直接變成不可回應的判詞。核查有自己的權力,需要自己的限制,不必改名成評價才值得制衡。
這些制衡也不是新的終點。委員會、第二評閱、外部專家都可能判斷失當,也都應就自己的那一判作答。不能因為他們專門審別人,就取得別人永遠不能審他們的地位。
這篇不許諾設了複核就更準確,不替某種評分制度宣佈勝利,也不判那個學生有沒有作弊。它先讓一項本來被壓縮的工作重新展開:讀數是什麼證據,這次評價怎樣完成,誰能受理疑問,各自真正說明到哪裡。
螢幕上的百分比可能還在那裡,成績也終究需要作出。重要的不是把所有數抹掉,而是不要讓人消失:採用工具的人仍要判斷,受到評價的人仍能回應,結論作完以後,也仍記得它只完成了這一回、這一範圍裡的事。
Source & reading note原论文与阅读说明原論文與閱讀說明
Eight essays draw on the University prequel and seven papers. They examine a university direction publicly treating people as ends, not a ranking of real institutions or a belief test for admission and employment. Imagined encounters are distinguished from linked institutional texts, read within their dates and stated scope. Unverified arrangements are not declared absent, and written policies do not certify implementation. Normative requirements, suggestions, role responsibilities and existing rules remain distinct. Basic protections do not disappear when an institution changes direction; questioning does not itself abolish fees, assessment or professional judgment. These essays offer no case-specific legal, admission or disciplinary advice. English is independently written.
八篇据《大学论》前置篇与七篇正文改写,讨论公开把人作为目的的大学方向,不给现实院校作整体排名,也不以信仰审查作为入学或任职条件。设想场景与已列来源的制度文本分开;真实规则只按材料日期、适用范围和页面能说明的程度引用,未核不写成不存在,规则不代证实际执行。作者提出的规范要求与建议、不同行动的职责及现行制度安排分别说明。人的基本保护不因一所大学改变方向而消失;收费、考核与专业判断也不因受问而一概取消。不提供法律、招生或学术处分的个案建议。简繁中文同文,英文独立重写。
八篇據《大學論》前置篇與七篇正文改寫,討論公開把人作為目的的大學方向,不給現實院校作整體排名,也不以信仰審查作為入學或任職條件。設想場景與已列來源的制度文本分開;真實規則只按材料日期、適用範圍和頁面能說明的程度引用,未核不寫成不存在,規則不代證實際執行。作者提出的規範要求與建議、不同行動的職責及現行制度安排分別說明。人的基本保護不因一所大學改變方向而消失;收費、考核與專業判斷也不因受問而一概取消。不提供法律、招生或學術處分的個案建議。簡繁中文同文,英文獨立重寫。
Read the source paper ↗阅读原论文 ↗閱讀原論文 ↗
MIT first-year grading stagesMIT第一年评分的阶段安排MIT第一年評分的階段安排 ↗
Brown: course-assessment options and no calculated GPA布朗大学:课程评价选项与不计算GPA布朗大學:課程評價選項與不計算GPA ↗
Hampshire: narrative assessment also has advance criteria汉普郡学院:叙述式评价也有事先标准漢普郡學院:敘述式評價也有事先標準 ↗
Princeton's 2026 proctoring and continuing Honor Code普林斯顿2026年监考安排与荣誉制度的区分普林斯頓2026年監考安排與榮譽制度的區分 ↗