Today we have a special new team member joined our subMIT project team. First name Artificial and last name Intelligence is referred to as A2rchi (pronouns: she/her) and will start to take your questions here. She has been trained on the full subMIT users guide, technical design details and specific software which is commonly used on the system.

A2rchi is initially available as a simple chat window on our website so that users can take advantage of immediate feedback. For now though we have to warn users about the fact that A2rchi can be very slick and convincing but also be completely wrong about the answers it will provide. We have added some warning statements on the web page.

To address the issue of possibly wrong answers (AI hallucinations) users can write email to submit-help@mit.edu. Those questions will run through a helpdesk discussion of experts and a verified answer will be provided. As a new team member A2rchi has of course joined our helpdesk (Cleo), but will for now only suggest answers that will be curated by the humans in the team. We will be training A2rchi with these conversations. As the general system administration work is also going through the Cleo system that maintains the helpdesk, A2rchi will be able to join the expert conversations if a particular ticket is assigned to it. A2rchi’s functions of course will be limited to avoid undesired consequences.

More generally speaking A2rchi runs as an external service that interacts with all the different systems human experts would also interact with (email, slack, Cleo ….). Guardrailling and training will help to develop A2rchi’s reliability and its authority to interact with the experts, the users and eventually with the system. A2rchi has an account on the subMIT computing system and can eventually help the users by providing live system information relevant to the question and it might be able to help the experts to fix obvious problems like for example re-mounting a volume that has lost connectivity. Our monitoring alarms will be fed to A2rchi for potential actions and careful documentation of those occurring issues.

In summary, we are excited to welcome A2rchi to the team and hope she will augment users in accomplishing their physics research and the support team by making them more efficient and effective.

Julius Heitkoetter presented our A2rchi prototype at the Physics Department’s workshop on “The impact of chatGPT and other large language models on physics research and education”