This technology can be a tool designed to enable people to easily and accurately cite the sources of information when quoting or referring to it in written communication. It enables users to cite the source of the information they copy online in an easy to read, consistent format.

This technology would help reduce the amount of misinformation and misattribution of content, as well as reduce plagiarism. It also ensures that credit is given to the original authors of content.

If you’re interested in building this idea, feel free to contact and please cite me ;)

Technical frame

The simplest form of Copy with Citation would be to have it be a web-based tool that creates a citation for anything a user copies from an online source. This can be in the form of a browser extension. Another important feature can be preserving the atomicity of the copied text and its corresponding citation. This would mean that if the user makes any changes to the text — removing certain words and phrases or even paraphrasing the original text, the citation remains in the document unless the copied text is completely deleted.

However, to build it as a robust tool, it could detect the source of the information for any text that the user writes and provide users with an easy way to cite it. This could be done by using natural language processing and data mining techniques, as well as integrating the tool into existing online writing tools. This is technologically feasible currently since it would require similar techniques such as keyword analysis employed by current plagiarism checkers such as TurnItIn¹.

The technical frame for this technology consists of several components: natural language processing, data mining, integration into existing online writing tools, and user privacy considerations.

Natural Language Processing (NLP) involves using computer algorithms to analyze text and extract relevant meaning from it. NLP can be used by Copy with Citation to parse through quotes and references in order to detect which source they come from. This process can be further refined by utilizing data mining techniques such as clustering and classification in order to categorize the various sources correctly.

Integrating Copy with Citation into existing online writing tools is another important component of its technical frame. By doing so, users will not have to install any additional software or go out of their way in order to use the service — they’ll simply be able access it right within their current word processor or other writing tool. Additionally, integrating the tool could potentially provide access to more powerful features than if someone were just using a standalone version of the program since these programs often come equipped with more advanced capabilities related specifically for citation management purposes like auto-generating bibliographies from cited sources². This would also give the users independence on what citation format to follow such as APA, MLA, etc.

Finally, user privacy must also be taken into consideration when designing Copy with Citation’s technical framework. All collected data should be securely stored and never shared without explicit permission from its owners — even if it is anonymized first — since there are still potential risks associated with sharing such sensitive information publicly without consent regardless how “anonymized” it may seem at face value. Furthermore, users should always have control over what information is being collected about them via an easy-to-understand interface that clearly outlines what kinds of data are being tracked (if any).

Ethical Dimension

The ethical dimension of this technology is that it would help to ensure that content creators and authors are properly credited for their work, which is a fundamental right. This technology would also help to reduce the spread of misinformation and plagiarism, which can have serious consequences.

Plagiarism

One big use case for this tool would be for college students like me. Students face pressure to turn in multiple assignments for dierent classes each with long readings under strict deadlines. These readings often have specific information that needs to be mentioned word-for-word or paraphrased to achieve the right level of depth in assignments. This equates to students often copying terms or phrases from multiple texts, each of which need to be credited in order to avoid plagiarizing. However, this process becomes tedious owing to the multiple strict standards for citations such as APA, MLA, etc. Other times, students may simply forget the multiple sources that they copy specific details from. Thus, even with the best intentions, this leads to high amounts of accidental plagiarism. This problem has only become worse since the pandemic, with plagiarism increasing 13%³.

Current solutions only address this problem at a high level and don’t address the root cause. For example, in plagiarism checking softwares like TurnItIn, the general acceptable similarity scores for students is usually around 15%⁴. The Copy with Citation software would ensure that accidental plagiarism is greatly reduced since it takes care of the tedious part of creating the citation and make it really hard for a malicious student to commit traditional (ctrl+c, ctrl+v) patchwork plagiarism.

Attribution of Credit vs User Privacy

The main ethical challenge posed by this technology is how to balance the need for accurate attribution of content with user privacy concerns.

To address these challenges, the tool should be designed in such a way as to not collect any unnecessary data from users and only store data related specifically to citing sources of information. Additionally, best practices such as encryption should be used wherever applicable so that the user’s personal data remains secure at all times.

Furthermore, mechanisms should be put in place so that users can easily delete all collected data on request if they wish to do so. The tool also needs to make sure that any citations it produces are accurate and up-to-date so there can be no confusion about where the source material came from or who provided it originally. To achieve this, natural language processing and data mining techniques will need to be used whenever possible along with some manually curated databases of relevant sources kept up-to-date by knowledgeable staff members.

Value Driven Design Choice

The specific values that Copy with Citation upholds are that of authorship and ownership, the fundamental right of content creators to be credited for their work. This is important because authors and content creators should have their work attributed to them, as it serves both as recognition for the creator and a way of protecting intellectual property.

The tool should also be designed with user privacy in mind, since all data collected by the tool (such as citations) will need to be securely stored and not shared without permission.

Through its implementation , individuals are able to properly attribute credit where appropriate while protecting themselves from potential legal issues associated with copyright infringement . Furthermore , this technology provides an invaluable resource in promoting reliable knowledge exchange among peers while reducing overall levels of misinformation spread across digital channels.

Another important factor is that the process should be easy and straightforward for users, which can be accomplished by making the tool available on web-based platforms. More features such as auto-fill/auto-complete options could also help simplify the process even further while helping reduce errors in citing sources correctly. Additionally, integration into existing writing tools (such as Microsoft Word or Google Docs) would make it easier for users who are already familiar with those platforms instead of having them learn how to use yet another software package from scratch.

Finally, another value-driven design choice would be integrating this technology with existing online writing tools such as word processors or content creation sites like Medium or Wordpress. This integration will make it even easier for people who need to quickly reference sources when writing articles or essays etc., thus reducing any potential barriers preventing them from properly attributing credit where due. This integration will make it even easier for people who need to quickly reference sources when writing articles or essays etc., thus reducing any potential barriers preventing them from properly attributing credit where due.

Overall, Copy with Citation upholds an important value: authorship and ownership — giving credit where credit is due while protecting intellectual property rights — through several key design choices; ease of use via web-based platforms; accurate citations generated through natural language processing; robust privacy measures; and finally integration with existing online writing tools like word processors etc. These features taken together will enable users of this technology to easily acknowledge creators while preserving their right over their work in a secure manner.

Citations

It would be very weird to not cite sources for this idea, so here goes:

1. Turnitin.com Editors. (2022, December 2). The Detection is in the Details. Turnitin. https://www.turnitin.com/blog/the-detection-is-in-the-details

2. EasyBib®: Free Bibliography Generator — MLA, APA, Chicago citation styles. (n.d.). https://www.easybib.com/

3. Report: Students Plagiarized More When Instruction Moved Online -. (2021, April 9). THE Journal. https://thejournal.com/articles/2021/04/09/report-students-plagiarized-mo re-when-instruction-moved-online.aspx

4. Kasen,J. (2021, September 6). Acceptable Turnitin similarity: How much is too much percentage. Grade Bees. https://gradebees.com/acceptable-percentage-for-turnitin/

Sorry you couldn’t make it. Not to worry, we can catch you up with the minutes of the hearing. I will read them out to you —

FADE IN:

INT. COURT ROOM — DAY

The room is filled with people as the trial is underway. The PROSECUTOR stands up and addresses the STUDENT.

PROSECUTOR

So, Student, can you walk us through the day you made the paper submission?

STUDENT

To be honest, it was quite a blur. I woke up at 12 on that fateful day. My brain was foggy and not in the best spirit, and I checked the course website to see that I had this assignment due today… I mean today then, that day!

PROSECUTOR

This was a required submission?

STUDENT

Yes, I needed to submit this paper to pass the course.

PROSECUTOR

And you didn’t know this assignment that you needed to pass the course was due till the last day?

STUDENT

… I knew about the assignment but had the dates messed up and other assignments for which I had to take late days.

(Points to the jury)

You know how rigorous our university is with its expectations of students!

PROSECUTOR

I don’t see how that’s relevant here, but let’s say this wasn’t premeditated. Is the board supposed to believe that you completed an assignment that other students took an average of a week to complete in the span of a day without any external help or support?

STUDENT

Yes sir, that is correct

PROSECUTOR

Oh, and is this the norm for you? Completing work in significantly less time than your peers?

DEFENSE COUNSEL

Objection! Irrelevant

BOARD

Overruled, continue.

STUDENT

No sir, this is perhaps a one-off occurrence given the circumstances. You know, like an athlete high on adrenaline?

PROSECUTOR

OK, given your athlete comparison, did you take the help of illegal performance enhancement drugs for your career best?

STUDENT

As far as I know sir, Caffeine is a legal drug

PROSECUTOR

Alright, now is it true Student that you take an interest in Machine Learning? Would you say it is a hobby of sorts?

STUDENT

Sure

PROSECUTOR

And you have an active fascination with certain kinds of ML models, text-generation, especially?

STUDENT

I am aware of them, yes.

PROSECUTOR

And is it not true that you have a workstation capable of training several high-end ML models very quickly at your whims and fancy?

DEFENSE COUNSEL

Objection! Irrelevant

PROSECUTOR

The question is relevant if the student submitted a paper that contained text generated by an ML algorithm he trained.

BOARD

Objection overruled, please continue

PROSECUTOR

Thank you Board. Now Student, is it not true that you used an ML algorithm to generate the text in your paper?

STUDENT

I don’t know what you’re talking about.

PROSECUTOR

Were you not talking to your friend Student 2 about training a GPT-2 model on a large corpus of texts the night before the paper was due?

STUDENT

I honestly don’t remember. We must have been joking. It was a blur like I said.

Prosecutor: Your Honor, I would like to enter as evidence the chat logs between Student and Student 2 from the night before the paper was due.

BOARD

Very well, proceed.

(The chat logs are entered as evidence)

PROSECUTOR

Now Student, let me ask you again, did you use an ML algorithm to generate the text in your paper?

STUDENT

No, I did not.

PROSECUTOR

Admit it! You thought you could fool everyone. Is this not how you were able to quickly train a model, generate text that passed off as a well-written paper and submit it 29 minutes before the course deadline?

DEFENSE COUNSEL

Objection! Leading, the Prosecutor is badgering the witness.

BOARD

Sustained.

PROSECUTOR

That’ll be all. I have no further questions.
(…)

The court is adjourned for a break.

EXT. COURTROOM — DAY

The jury is seen deliberating amongst themselves.

INT. COURT ROOM — LATER

The DEFENSE COUNSEL brings the PROFESSOR to the stand.

DEFENSE COUNSEL

I would now like to bring the plaintiff Professor to the stand

DEFENSE COUNSEL

Now, Professor, did the Student submit the paper on time?

PROFESSOR

Yes, the paper was submitted 29 minutes before the course deadline.

DEFENSE COUNSEL

And is this an odd occurrence for the student?

PROFESSOR

I don’t understand your question… no?

DEFENSE COUNSEL

So, Student is known to submit their papers on time. And has any one of their previous papers been plagiarized?

PROFESSOR

No, not to my knowledge. We run our submissions through Turnitin and they’ve always been below the threshold of plagiarism.

DEFENSE COUNSEL

Funny you mention TurnItIn. Your honor, I would like to enter as evidence the University’s Policy on Academic Integrity. Professor, is it not true that as per your course policy, using TurnItIn is the standard used to judge plagiarism for all students in all scenarios?

PROFESSOR

…Yes, but this case is different!

DEFENSE COUNSEL

Professor, did this paper pass the plagiarism threshold of TurnItIn?

PROFESSOR

Well… no. But- (Interrupted)

DEFENSE COUNSEL

So, Professor, you’re admitting to accusing a Student of breaking the University’s Policy on Academic Integrity without it passing the traditional plagiarism testing procedure prescribed by the School board?

PROFESSOR

Well, when I ran it through TurnItIn, there were no matches. But, I ran it through another software that explicitly checks for artificially generated content

DEFENSE COUNSEL

What caused you to use it especially this time?

PROFESSOR

I overheard Student and Student 2 discussion about the topic.

DEFENSE COUNSEL

That’s a hearsay argument, but let’s assume you’re right. How does it work?

PROFESSOR

It is by the company ShutAI that uses ML to check whether a paper has been generated by an ML algorithm by analyzing the style, grammar, and other linguistic features.

DEFENSE COUNSEL

And what did it tell you about Student’s paper?

PROFESSOR

It passed the threshold at 68% generated content.

DEFENSE COUNSEL

Did it give any other information detailing the evidence for generation?

PROFESSOR

Um… no, not as far as I can recall.

DEFENSE COUNSEL

So, Professor, if I’m getting you right, you can’t explain how this ML software detects generated content?

PROFESSOR

No, but ShutAI’s a reputable company and their research paper was all over HackerNews.

DEFENSE COUNSEL

Define reputable, Professor

PROFESSOR

Uh ok… It means something with a good reputation?

DEFENSE COUNSEL

That wasn’t my question — Nevermind, Professor, is there anything in the software documentation that indicates how they determined this paper had generated content?

PROFESSOR

No… their code is proprietary

DEFENSE COUNSEL

So, is the Board just supposed to take their word for it?

(Awkward silence ensues)

DEFENSE COUNSEL

So, you’re saying that this software that is not commonly used and is not part of the University’s standard procedure for plagiarism testing is what you based your accusation on?

PROFESSOR

Well, when I saw the results, I also looked at the paper manually and there were whole sections that didn’t make sense!

DEFENSE COUNSEL

So, you’re saying that you found evidence of, in your own words, “not Student’s own original academic work” not just from the software but also from reading the paper?

PROFESSOR

Yes!

DEFENSE COUNSEL

Professor, can you please point to us which section of the paper was artificially generated?

PROFESSOR

Um… well, the whole section on [redacted] had weird sentence structure and it just didn’t sound like something a student would write.

DEFENSE COUNSEL

So, your entire basis of the writing section on [redacted] being generated by an ML algorithm is essentially personal opinion?!

PROFESSOR

What? No, of course not!

DEFENSE COUNSEL

Are you sure? Because it seems to me like you’re going out of your way to make an example of Student.

PROSECUTOR

Objection! The defense is speculating!

PROFESSOR

I… I… don’t know what you’re talking about. I’m just trying to do my job.

BOARD

Order! Order in the court! Defense please refrain from such questioning.

DEFENSE COUNSEL

I have no further questions.

(…)

FADE OUT

Now that you have all the information, the Board requests you to deliberate come up with a decision.

Writer’s Note

This story is based on the theme of utilizing AI / Machine Learning in real life. It also explores the questions of openness, accountability and explainability of ML. With advancement of Machine Learning, especially in the domain of Natural Language processing, there have been many tools that have entered the public’s knowledge.

Recently, one of the famous ones is GPT-3¹. It’s an auto-regressive language generation model that takes in some text and generates more based on context. Unlike ML models of the past, it is not domain specific. GPT-3 has been used in a lot of domains such as sentiment analysis, code generation, classification, summarization and translation. In fact, while this idea was conceived fully independently, for the purpose of this script, I experimented with generating a few dialogues using GPT-3. This turned out to be highly successful, with multiple readers not able to recognize which of the sentences were generated.

With its popularity and recent open beta availability, this has led to its extensive use in a variety of applications. One of these use cases unfortunately happens to be students using GPT-3 or GPT-3 like models to generate text to use in their homework assignments that passes plagiarism checker.²

This is a very slippery slope. On one hand, it is morally less worse than textbook plagiarism, but it is reduced effort compared to a student actually spending the time to work on an assignment completely on their own. Moreover, there still remains the question of the source of the generated text since the model is trained on large and vast corpuses of data. Thus, there may need to be more regulation around the use of these technologies and the penalties associated.

Another aspect covered is differentiating between human written and AI generated texts, which inherently seems to be a problem that ironically may be solved with AI itself.³ There’s a cyclic nature to this question since if there’s a good discriminator between real and generated text, then it can be used to train better generators, until it’s tough to differentiate between. This is a problem in the Computer Vision DeepFake sphere.⁴

Finally, an aspect of this which is fascinating is the lack of transparency around these models. For example, OpenAI’s GPT 3 is still not open source and only available in beta in limited usage.⁵ While the company claims this is due to security considerations, it makes it harder to regulate its use and explain how it works without the public properly knowing about the internal implementations.

References

1. Tom B. Brown, et al. “Language Models are Few-Shot Learners”. CoRR abs/2005.14165. (2020).
2. “Students Are Using AI to Write Their Papers, Because of Course They Are.” VICE, 14 Oct. 2022, https://www.vice.com/en/article/m7g5yq/students-are-using-ai-to-write-their-papers-because-of-course-they-are .
3. Lakshmanan, Ravie. “This AI Tool Is Smart Enough to Spot AI-Generated Articles and Tweets.” TNW | Artificial-Intelligence, 29 July 2019, https://thenextweb.com/news/this-ai-tool-is-smart-enough-to-spot-ai-generated-articles-and-tweets .
4. Anderson, Martin. “Deepfake Detectors Pursue New Ground: Latent Diffusion Models and Gans.” Unite.AI, 14 Oct. 2022, https://www.unite.ai/deepfake-detectors-pursue-new-ground-latent-diffusion-models-and-gans/ .
5. Gershgorn, Dave. “GPT-3 Is an Amazing Research Tool. but OpenAI Isn’t Sharing the Code.” Medium, OneZero, 20 Aug. 2020, https://onezero.medium.com/gpt-3-is-an-amazing-research-tool-openai-isnt-sharing-the-code-d048ba39bbfd .