Last updated March 2024
Please share your work with Generative AI technology using this google form.
See also the examples of Faculty engagement with GenAI when teaching at Amherst College
Click here to read what Faculty are sharing with the Amherst College Community (AC login required)
Using Generative AI To Build Equitable Access to Technology
Can Generative AI help Amherst College IT provide better responses to requests for assistance? Amherst College IT responds to a wide variety of requests from the Amherst College community using a ticketing system. As they respond to each direct request, Amherst IT also looks to understand larger patterns in the requests that suggest common underlying causes that may not be readily apparent. To better achieve this standing mission, a team of Amherst IT staff is exploring whether the natural language processing capabilities of Generative AI Tools can be used to find such patterns in anonymized help request tickets and subsequently use them to recognize and respond to similar issues in later tickets.
The Goal
The emerging task is to incorporate generative artificial intelligence into the analysis of help tickets for Augmented Intelligence, supporting human intelligence and judgment instead of replacing it. The consideration of how to analyze the needs of many in an equitable way, working towards fair and just outcomes with attention to individual needs and circumstances, calls for an approach of Human in the Loop Automation, where human analysis, intelligence, and intervention are supplemented by an additional source of information that would be labor-intensive to obtain without the assistance of AI tools.
Academic Technology Tools Used
- OpenAI API to access GPT-4 for Natural Language Processing
- SolarWinds Service Desk for managing Amherst IT help request tickets
Featured Staff
- Stefan Antonowicz, IT Systems and Networking
- Justin Chen, IT Systems and Networking
- David Yang, IT Support Services
Problem Statement
Providing reliable wi-fi internet access to everyone in the Amherst College community is ongoing work for the Amherst College IT staff. The technology infrastructure requires continuous maintenance and upgrading to adapt to how buildings are arranged, how they are used, and people's needs. One step in this process is creating maps of where the wireless signal is strong and where it is unusable, using tools such as Ekahau Connect Wi-Fi Heatmapping technology.
An example of a heat map showing the variation in wi-fi signal strength in a residential building. Some areas have strong signal strength (green) and other areas have weaker signal strength (red). Image Credit: From Ekahau’s Website Wi-Fi Heatmap Software - Visualize Coverage and Capacity | Ekahau
Stefan Antonowicz and the System and Networking group regularly deal with the technical challenges in maintaining wireless infrastructure. One central issue is how the community experiences an area of poor connectivity — cell phones and computers accommodate spotty coverage in various ways. As a result, “wireless connections do not drop, they slow.” A slow connection may appear to be a different issue entirely, including a problem with the user’s device or trouble somewhere else in the network. When a user reports issues to Amherst IT, the immediate response may provide a solution for the individual requester that works for them, but does not reveal the underlying issue with wi-fi connectivity.
Working with Justin Chen and David Yang, Antonowicz hopes to uncover patterns in the help requests that point to underlying issues in the Amherst IT infrastructure, taking advantage of new opportunities with Generative AI technology. While it would take a prohibitive amount of time for a person to carefully audit all possible issues related to all the help tickets, this is possible using Generative AI tools with natural language processing capabilities that help to identify descriptions related to wi-fi connections even without precise search terms. Antonowicz sees the broad potential for tools built around natural language processing to add additional sources of information, even if they require different and specialized supervision.
Indeed, the team’s first challenge was to consider whether this technology could be appropriate for the task at hand. Yang says his first question is, “What if I’m wrong?” Generative AI can fabricate information that does not reflect reality. Stefan Antonowicz speaks to using it with fault adaptive design, considering how a process can be changed to respond to all the ways it might fail. For the case of using Generative AI to investigate user reports, everyone involved in the work needs to understand that it should be used in addition to, rather than instead of, existing processes, augmenting the team's capabilities.
Ensuring Data Privacy and Security
An additional concern is that Generative AI technology often uses its input as additional data for building its language models, while he Amherst College Community should be confident that their requests for help will be kept private and secure from unauthorized access. For this reason, the OpenAI technology was selected, which is a way to access the power of GPT-4 Large Language Models without consenting that data be used for training future models. Furthermore, the Amherst College IT staff worked together to develop a manual procedure for initially removing sensitive information — anything personal that would not already be publicly available, and also anything identifying individuals, such as their names. Only then were the help request tickets entered into the OpenAI model. Future work may include ways to automate this process in a way that is private, secure, and reliable.
Once the tickets were anonymized, they were entered into OpenAI with a task formulation, essentially: "Is this ticket about a wifi issue? If so, where is the location of the issue? Take care to reply with 'uncertain' if there are cases where locations are mentioned, but there is no issue." The response to this prompt helped identify a list of areas for the team to investigate, where additional analysis with tools like the Ekahau Wi-Fi Heatmapping technology could be warranted.
Emerging Outcomes
- Amherst College IT is exploring the use of Generative AI to find underlying needs from the Amherst College community in the patterns that arise in requests for help.
- Responsible use of many Generative AI tools requires ongoing and case-specific work to prioritize privacy and security and understand the limitations of what it generates.