Topic: Welcoming the Future: AI and Administration
Overview: MIT will soon be launching the Working Group for AI in Administration and Operations to understand the fast-evolving landscape of AI and generative AI tools and current capabilities. The input from table discussions will be shared with the Group. Because the generative AI landscape is evolving so quickly, we encourage you to be thoughtful in your responses to the questions below.
See table questions and discussion notes below.
1. What AI tools are you using now?
- Google AI, when you Google something there is an AI generated response.
- ChatGPT, helped with writing performance reviews. Makes them easier to complete.
- 2 people are using tools minimally in the ways described by Libraries (Stephanie).
- A person from Professional Education shared that AI is the most important topic for their unit. They are exploring getting licenses for Microsoft Copilot Pro for everyone in the unit and working on a Departmental Charter for use of AI as well as training.
- A person didn’t think we’re allowed to use AI tools based on MIT’s guidance.
- One table reported not using generative AI yet. Others reported using ChatGPT, Claude, dictation software, and the Zoom AI companion.
- One table reported using AI to summarize meeting minutes, drafting a cover letter, and understanding a Python program.
- One table was amazed by how well ChatGPT works for taking minutes. Had also used AI to check due diligence about companies, to check a narrative or get a second view on it, and to try writing in a style.
- Fathom is similar to Otter, for facilitating on Zoom with industry leaders. Creates notes and creates next steps. Consent has come up before, along with the issue of turning it off when privacy is needed.
- ChatGPT for Excel formulas, Claude is more literary.
- Gemini for drafting from prompts, and for emails.
- Evernote for notetaking, with a new feature that AI cleans up the formatting of the notes.
2. What AI tools could you be using?
- Microsoft Copilot, ChatGPT, Claude, ABBYY.
- Two tables were uncertain, with one table not sure yet how to use generative AI for financial tasks.
- One table is hesitant to use it because of data security and worry about losing skills. Nonetheless, they might try it for some HR tasks.
- Some attendees said that generative AI had been recommended by a coworker or friend.
- Could be using generative AI to prepare PowerPoint slides.
3. What concerns do you have about generative AI?
- AI technology is scary.
- Resistance to adaptation, fear that employees aren’t needed anymore, that it might take over people's jobs.
- How do we know if the output is factually correct?
- Administrators need to remember that you can’t opt out. This is something that you need to learn. Other tools administrators use i.e. (Brio, Tableau) they don’t have to learn and there are ways around using them.
- Proprietary and confidential information, data security, inherent bias, and where our information is going once it's been fed in.
- Logging in versus not logging in to using an AI tool.
- Opting out of having private information collected.
- Being burdened with more spam from vendors, recruiters.
- Not knowing what is original content from a candidate.
- Being left behind by not using generative AI.
- Not paying as much attention to the output.
- Losing the humanity in communication, and losing the communication skills we currently have.
- Security and data breaches.
- AI might make some skills such as writing and critical thinking obsolete.
- Information is limited depending on the model's training.
- Concern about IP rights and plagiarism.
- Anonymizing the information to make sure it's not taking in personal information is important. We saw risk if too many details are fed into the tools.
- It was interesting to hear Meghan McCollum Fenno mention there is concern on using the tools on work computers.
4. What excites you about generative AI?
- It’s great to use for charts, get instructions on how to do things, writing job descriptions and interview questions.
- It gives a starting point. Good for drafting, but needs human intervention.
- It’s a great tool to help with brainstorming ideas and gathering thoughts.
- There is a Zoom function that uses ChatGPT that can summarize Zoom meetings.
- Brio is no longer supported, so maybe it could be used to access the data warehouse.
- Learning the many ways to use generative AI, but within limits. It was great to hear how Libraries is using it.
- Help with data mining/generating reports instead of using traditional programing language.
- See how MIT will incorporate AI in upcoming systems upgrade.
- It's good for narrative documents, and helpful for tweaking messaging.
- It can reduce workload and make one's job easier.
- It could free up time for higher order work.
5. What support do you need, training, information?
- It would be helpful to know if Research Administration Services has any policies on using AI for grant applications.
- Would like to know general practices from central offices.
- IS&T’s guidance gives the impression that we shouldn’t use AI and we need much more information and support to help us understand how to explore the power of AI and maximize its use.
- Guidance on how to choose which AI tool to use.
- Concerns that the focus on the digitizing project will prevent appropriate attention being given to the use of AI on campus and we will be very behind in 7 years when the digitizing project is scheduled to end.
- Create a “sandbox” for users to test AI solutions.
- Need to learn how to use good prompts.
- Need lots of support: best practices, website updates, proper use, what not to do and how to use it.
- Quality assurance.
- Real world examples.
- Disclaimers.
- This meeting was a great experience for sharing ideas on how to make everyday tasks more efficient. More of these discussions with input from power users would be helpful.
- Would like to get training about how these tools can be used for efficiency.
- More training about confidentiality would be helpful.
Transcription of a Speaker's Q&A:
- Difference between ChatGPT and Claude? Claude is trained on a different language model. If the task is for drafting English language, Claude seems to be better. Also mentioned that it is trained to be more secure and accurate. Chat GPT seems better trained for mathematical information.
- What confidence is there that we will enjoy the extra time, and reap the benefits of the extra time? With more efficiency people do more of the work. Perhaps we should establish that if AI frees up 10% of our work time, we will use 3-4% of this time to do more strategic work.
- As admins at MIT, what is our institute guidance for using GenAI? IS&T has guidance on its website, including a list of approved tools. The list is likely dynamic and will keep changing. The list of tools that have licenses has been scrutinized by members of IS&T. Recourse is limited if something goes wrong. Enterprise licenses have been reviewed by OGC. Advice is to use a critical eye and don't share confidential information. Pre-publish results and sensitive information shouldn't be entered into these systems.
- Is there a plan to exploit the full power of the tools? It was mentioned that users can opt out of data being used for training by changing the settings of the tool. It will not save the entered information. Copilot is a Microsoft version of ChatGPT, and offers protection from leaking information.
- The Libraries used Otter AI to transcribe meetings. They need transcription software for someone who has a medical condition. There aren't guidelines on consent, where to store the transcriptions, or how long they need to be stored for. Libraries is working with OGC and Labor Relations about the transcriptions.