 |
| Screenshot of video from NY Times* |
To: Faculty, Instructors and Teaching Assistants
What you need to know:
• Today Turnitin released a “preview” feature designed to detect the possible presence of AI-generated text in student-submitted work.
• The feature is “trained” on OpenAI’s Large Language Models GPT-3 and 3.5, but not on more recent versions.
• Pending further investigation, UCLA has temporarily opted out of this feature.
• Academic and Administrative leadership are committed to providing the community opportunities to engage and research this new feature to determine how it fits into UCLA's educational landscape.
===
Dear Colleagues:
Today Turnitin released a “preview” feature designed to detect the possible presence of AI-generated text in student work, adding to the product’s longstanding functionalities designed to detect plagiarism.
Turnitin informed the public of this "preview" feature two weeks prior to its availability, which did not provide sufficient time for the university to engage with its governance bodies to review and discuss the major changes introduced by this feature. We believe it is important to address many questions about how it is best used, what limitations it has, and how its conclusions are interpreted to ensure fairness and transparency. Our priority is to understand how this feature would align with academic expectations, accessibility, and the proper protection of the community’s privacy and security.
Due to concerns and unanswered questions around this new feature, we have temporarily opted out of having it enabled (as have many of our peer institutions within the UC system and across higher education). Turnitin is currently unable to allow individual instructors to enable this feature. However, gaining experience with this new feature is a high priority for us so that we can build a solid understanding of how AI detection can impact teaching and learning. We will announce updates as we gain access to review the feature with campus partners and governance, including faculty bodies.
A list of FAQs is listed below.If you have additional questions please contact the Bruin Learn Center of Excellence at BruinLearn-Support@it.ucla.edu or visit the Bruin Learn website.
Sincerely,
Adriana Galván, Dean of Undergraduate Education Interim Vice Provost for Teaching
Jasmine Rush, Interim Dean of Students
Lucy Avetisyan, Associate Vice Chancellor, Chief Information Officer
===
FAQ
What limitations does Turnitin’s AI detection have?
• It has only been trained to detect work generated from OpenAI’s Large Language Models GPT-3 and 3.5 and is not yet trained on newer versions (GPT-4) or other tools (Google Bard). As a result, it will not reliably detect all possible sources of AI- generated text.
• While Turnitin’s internal research estimates a 98% confidence rate in detecting text written by GPT-3 and 3.5, these numbers still contain a margin of error and have not been externally verified. Conversely, a low AI detection score provided by Turnitin does not necessarily mean the text was not AI-generated; it could have been generated by a tool that Turnitin’s detection model has not been trained to identify, or by a person mimicking AI writing styles.
• The AI score and report is only available to instructors, not students.
• Turnitin “AI detection” only supports content submitted in the English language.
When will instructors be able to experiment with this new feature?
This “preview” feature will be made available based on the following criteria:
• Engagement with campus governance, including faculty bodies, to consider the data and research on AI detection confidence rate.
• Campus's due diligence around accessibility, security, and privacy.
○ Specifically, reviewing the privacy policy changes Turnitin has updated and modifications to the user interface possibly impacting accessibility.
• When Turnitin allows UCLA to enable experimentation at the instructor and/or course level.
How should Instructors discuss AI with students?
• Consider sharing community agreements or providing expectations at the beginning of each course on the appropriate use of AI tools and how it will be used in that course (or not). See the recent Teaching Guidance for ChatGPT and Related AI Developments with examples.
Will the Office of Student Conduct accept cases where instructors suspect AI was used?
• Instructors can submit information regarding suspected academic misconduct from any source, including their own professional judgment.
• Reports will be reviewed and instructors may be asked to provide supporting rationale where appropriate.
• As per the student conduct code:
○ Cheating includes, but is not limited to, the use of unauthorized materials from any source.
○ Unless otherwise specified by the faculty member, all submissions, whether in draft or final form, (including a paper, project, exam, computer program, oral presentation, or other work) must either be the Student’s own work, or must clearly acknowledge the source.
===
*The video “a cow at a birthday party”:
No comments:
Post a Comment