This Week's News: OpenAI Can Detect AI-Generated Text
•
10 min read
This article has been machine-translated into English. Therefore, there may be nuances or errors in the content. The Danish version is always up-to-date and accurate. Read the original Danish article here.
This week we are focusing on something many educators have been requesting, namely a tool that can detect whether texts are written with artificial intelligence. It turns out that OpenAI has the capability, but they are reluctant to release it. In addition to referring to The Wall Street Journal, which writes about the tool, we also include a perspective from Jeppe Klitgaard Stricker, who has looked a little at the tool’s accuracy. Also remember to sign up to his newsletter – it likewise deals with AI and education, and he writes rather well.
We also include a report showing that many students are concerned that artificial intelligence could reduce the value of their education. At the same time, a survey from Harvard has revealed that almost 90% of undergraduates use AI for their studies. That may not be surprising, but for 25% of these students AI replaces many traditional learning activities. We link to the study and to Ove Christensen’s blog post on LinkedIn, where he reflects on the text.
As generative artificial intelligence is increasingly used by teachers to give feedback or create teaching materials, the question arises whether we should be transparent about this use. An interesting EdSurge blog post discusses the pros and cons.
In our newsletters we cover a broad span from the primary and lower secondary level to university level, and it can be hard to keep up with that breadth. However, we would like to point you to a new tool from the Playful Learning DK programme, which is aimed at the primary and lower secondary level. Here, with the help of generative artificial intelligence, you can get support for innovative teaching units.
Happy reading
OpenAI holds back on anti-cheating tool for education
An article in The Wall Street Journal reveals that OpenAI has secretly developed a technology that can detect with 99.9% accuracy whether a text was written by ChatGPT. The project has been under internal discussion at OpenAI for two years, and the technology has been ready for launch for about a year.
However, there is internal disagreement about whether the tool should be released to the public, due to concerns that it could lead to false accusations of cheating, particularly affecting certain groups, such as people with English as a second language.
The technology works by making a small change to the way words are selected in the text, leaving a “watermark” that is invisible to the human eye but can be detected by OpenAI’s detection system. A challenge for the system is that the watermark can be removed relatively easily with simple methods such as translation between languages or by adding and removing characters in the text. Simply running it through Google Translate will remove the watermark.
In addition, there is a balancing act regarding how many should have access to such a detection system. If too few have access, it will not be effective, but if too many do, there is a risk that the watermarking technique will be exposed and exploited.
Despite these concerns, OpenAI acknowledges the need to tackle the problem of AI-based cheating, particularly in the education sector. They are therefore carefully considering how they can share the technology in a way that is both effective and responsible. The article also mentions that OpenAI has initially prioritised developing watermarking technologies for images and audio, as the risk of misuse is considered greater there, especially in an election year.
Read the article in The Wall Street Journal
⚠️
Comment: Forget watermarking in texts
At Viden.AI we are writing about this detector because it is a topical media story and because it is interesting that OpenAI has secretly worked on watermarking. It may give hope that we could get a system for the education sector that can show whether students have submitted assignments written by artificial intelligence.
The challenge, however, is that even if OpenAI releases its detector, systems that can remove the watermark will quickly appear. In addition, there are many language models other than ChatGPT that do not have such a watermark. We therefore risk a false sense of security if we believe that watermarks in texts alone can solve all the challenges posed by artificial intelligence in the education system.
At Viden.AI we are writing about this detector because it is a topical media story and because it is interesting that OpenAI has secretly worked on watermarking. It may give hope that we could get a system for the education sector that can show whether students have submitted assignments written by artificial intelligence.
The challenge, however, is that even if OpenAI releases its detector, systems that can remove the watermark will quickly appear. In addition, there are many language models other than ChatGPT that do not have such a watermark. We therefore risk a false sense of security if we believe that watermarks in texts alone can solve all the challenges posed by artificial intelligence in the education system.
What does 99.9% accuracy mean?
Jeppe Klitgaard Stricker, a consultant and expert in digital learning, has also reflected on what OpenAI’s secret tool means. In his new newsletter he questions, among other things, what an accuracy of 99.9% actually covers, and points out that even a small margin of error can have major consequences.
He also raises an important discussion about when the use of AI in assignment writing can really be termed cheating. Instead of focusing on detection tools, he urges educational institutions to rethink their assessment and exam formats and place more emphasis on process and reflection rather than solely judging the final product.
Read his newsletter here:
Can cheating with generative AI be avoided?
OpenAI has a new tool that can reveal AI cheating with 99.9% probability. The problem is simply that it does not solve the underlying issue.OpenAI – the company behind ChatGPT – announced earlier this week in a rather eye-catching article in The Wall Street Journal that the company has for several years had a tool that can expose cheating with generative AI with 99.9% certainty.That may at first sound like very good news for the education sector, because cheating in connection with assign
Jeppe Klitgaard Stricker
New newsletter on AI and education
On LinkedIn, Jeppe Klitgaard Stricker has been very active, sharing valuable perspectives, in-depth analyses and thought-provoking reflections through his many posts. He has chosen to publish a weekly newsletter, and we recommend following it.
You can read his previous posts and sign up here:
New newsletter | stricker.ai
Students concerned about AI’s impact on higher education
A new report reveals that many students are worried that increasing use of artificial intelligence in teaching could reduce the value of education. According to a survey of 3,800 students from 16 countries, 55% of participants believe that excessive use of artificial intelligence contributes to devaluing their education, while 52% feel it negatively affects their academic performance. Only 18% find AI-based courses more valuable than traditional courses. Although many students are nervous about becoming too dependent on artificial intelligence, 86% admit that they regularly use tools such as ChatGPT in their studies.
The report also highlights significant uncertainty about AI guidelines at universities, with 86% of surveyed students not feeling fully informed. This has led to confusion, as students often receive conflicting information from different faculties. The report therefore underlines the need for clearer communication and training in the use of artificial intelligence for both students and educators.
Students fear over-reliance on AI ‘devalues’ higher education
Report stresses AI is ‘new standard’ and universities need to better communicate policies
Survey reveals extensive use of generative artificial intelligence among students
A survey from Harvard University shows that almost 90% of the undergraduates surveyed use generative AI in their studies, and for 25% of these students AI increasingly replaces traditional learning activities such as attending classes or reading required material. The survey also highlights that many students are worried that AI could harm their future job prospects, while half want more courses that address AI’s future impact.
Generative AI is primarily used to answer general questions, help with writing tasks and programming. At the same time, there is a clear economic disparity in access to AI, as students from higher socio-economic backgrounds are more likely to pay for premium AI services, giving them more advantages and reducing their need to use university resources. The survey recommends that universities adapt their teaching and resources to address AI’s growing impact.
Harvard Undergraduate Survey on Generative AI
How has generative AI impacted the experiences of college students? We study the influence of AI on the study habits, class choices, and career prospects of Harvard undergraduates (n=326), finding that almost 90% of students use generative AI. For roughly 25% of these students, AI has begun to substitute for attending office hours and completing required readings. Half of students are concerned that AI will negatively impact their job prospects, and over half of students wish that Harvard had more classes on the future impacts of AI. We also investigate students’ outlook on the broader social implications of AI, finding that half of students are worried that AI will increase economic inequality, and 40% believe that extinction risk from AI should be treated as a global priority with the same urgency as pandemics and nuclear war. Around half of students who have taken a class on AI expect AI to exceed human capabilities on almost all tasks within 30 years. We make some recommendations to the Harvard community in light of these results.
Shikoh Hirabayashi
Also read Ove Christensen’s interesting post on LinkedIn, where he reflects on this study.
Generative artificial intelligence is eroding good study habits
Admittedly. The headline is probably an overstatement or an angle intended to get you to read this post.
Ove Christensen
Should teachers disclose when they use AI in teaching materials?
The discussion about whether teachers should inform students when they use AI to create teaching materials has become topical as the use of generative AI increases.
In the article below, Marc Watkins from the University of Mississippi argues that it is important to teach students the proper use of AI, and that there should be transparency — including when we use the technology ourselves. However, many teachers and experts disagree about the need for such disclosures. Some compare the use of AI to other tools such as calculators, while others think it depends on how AI is used. For example, they believe it should be disclosed if AI is used to provide feedback on a student assignment.
The article suggests that it is important for schools and institutions to develop clear policies for the use of AI, so that expectations are clear.
Should Educators Put Disclosures on Teaching Materials When They Use AI? - EdSurge News
As more educators try new AI tools in their teaching, they’re wrestling with a thorny question: Should they disclose their AI use to students?
Jeffrey R. Young
Blingtron: A creative, virtual sparring partner for teaching in the Danish primary and lower secondary school (folkeskole)
Blingtron (LEG-GPT) is a new tool developed to strengthen teaching in the folkeskole by giving teachers access to an artificial intelligence that can serve as their virtual sparring partner.
The tool has been developed by Ulrik Rosenberg and Simon Mathiasen, who are part of the Play@Heart project under the national Playful Learning DK programme.
Blingtron helps teachers generate, refine and develop innovative ideas as well as create cross-curricular teaching units. What sets Blingtron apart is its ability to stimulate teachers’ imagination, opening up endless learning opportunities. The tool is designed to make teaching both playful and rich in learning, and the aim is to integrate technological literacy naturally into children’s development.
Try it here:
BLINGTRON - Play@Heart
BLINGTRON BLINGTRON (LEG-GPT) is used as a tutor for self-developing competences in generating units that are open, inquiry-based and playful. The tool can be used for inspiration that is adapted to teaching units and tailored according to needs and levels and degrees of abstraction. And hey! Remember to share your Supercalifragilisticexpialidocious BLINGTRON finds! The unit builder The subject Area/topic: Location (indoors/outdoors) Material and/or learning resource Imagine, […]
This week’s news
We must not oversimplify the debate on artificial intelligence
I can ask a chatbot for advice on everything from quantum physics to the meaning of life. I can live-translate conversations so two different nationalities can understand each other using artificial intelligence, and I can AI-generate a recipe for a tasty pasta dish from the sorry leftovers in my fridge. The list…
Mogens Kruse Jakobsen mkjjfm.dk
The national Danish essay is a beached blue whale
Tomas Bjørn Pedersen
Probability is not truth: Why you should be careful when using ChatGPT
Language models use probability to write a text. That makes them unreliable and reflects a problematic understanding of truth.
See Stella’s profile
Tracing the chatbot’s linguistic fingerprint: A linguistic authorship analysis of texts written by Danish students and ChatGPT | NyS, Nydanske Sprogstudier (New Danish Language Studies)
Authors Jonas Nygaard Blom Alexandra Holsting Jesper Tinggaard Svendsen
Future-Proofing Education: AI Will Redefine Learning And Teaching
The education industry faces disparate challenges with outdated systems amid AI’s enormous potential. Fraught with educator resistance, data privacy, and ethical issues, this sector must be redefined to create a future-ready system for learning.
Hessie Jones
Back to the future with AI?
Will artificial intelligence transform school?
The boldest visions for AI in classrooms misunderstand much about education
The Economist
Written by
MSc in IT | Co-founder of Viden.AI | Head of IT and Communications at TietgenSkolen
