From violence and thoughts of death in Hamlet to the dramatic double suicide at the end of Romeo and Juliet, the works of Shakespeare are filled with troubling themes. Student assignments about these works often contain phrases that, taken out of context, may look concerning to Artificial Intelligence (AI) systems designed to flag signals of students in crisis.
Understanding Context to Reduce False Positives
When classwork like Romeo and Juliet is flagged by an AI system looking for signs of self-harm or violence, it is considered to be a false positive. These false positives can be especially frustrating when 20 or more students in a class are all working on the same triggering assignment.
Lightspeed Alert proactively looks for signs of students in need of help by scanning online searches, Google and Microsoft productivity suites, and other sources. New machine learning capabilities added to Lightspeed Alert significantly reduce false positives like those found in Shakespeare assignments by consuming and understanding a full body of text, including surrounding context, to understand if content is concerning. In addition to context, these models can understand humor and hyperbolic language.
Types of content that will no longer cause alerts include (but are not limited to—the full list is extensive):
- Romeo and Juliet, To Kill a Mockingbird, Macbeth, and Hamlet homework.
- Chemistry homework (i.e., non-concerning mentions of sodium nitrate).
- Homework referencing battles, wars, or things that are inherently violent but not concerning within context.
- School reports/research of general suicide awareness and the national suicide hotline.
- Alerts containing non-concerning words that have the same letters as a concerning word (i.e., “edit” contains the same letters as “die,” but is not concerning).
- Homework related to reproduction/sexually transmitted diseases and sexual health.
- References to the poem “If we must die” by Claude Mckay.
- The same exact alert is being duplicated an excessive amount of times.
Building a Foundation for Future Learning
As the fields of machine learning and artificial intelligence continue to advance rapidly, Lightspeed is committed to making ongoing investments in our Lightspeed Alert technology to improve the product even further. The new machine learning capabilities help build the foundation for ongoing refinement and updates to the model, including regular updates to incorporate the latest information (i.e., new slang) and improve accuracy even further.