Beyond a Moment of Silence: How STEM Can Prevent Gun Violence | Divya Nori | TEDxAlpharettaWomen
The speaker argues that while political activism is important for addressing gun violence, technology—specifically AI analysis of social media—offers an untapped, quantifiable method for early detection. This method assigns a numerical coefficient score to phrases to flag potential high-risk language, demonstrated via a mobile app prototype alerting parents to concerning text messages. ## Speakers & Context - Unnamed speaker; giving a presentation on gun violence prevention. - Context involves responses to specific, highly publicized mass shooting events (Sandy Hook, Parkland, Santa Fe). - The speaker aims to show how technological analysis can complement traditional activism in tackling this issue. ## Theses & Positions - Gun violence is an *ongoing concern* that cannot be let slip into the background simply because national headlines shift. - Relying solely on qualitative observations (like T-shirts or specific social media posts) can be misleading regarding risk. - Technology's greatest asset is analyzing *large amounts of data* to find trends in a short amount of time, quantifying risk factors. - STEM offers an untapped perspective that can complement, rather than replace, political activism in addressing social issues like gun violence. - Society has an obligation to create a safer world for future generations. ## Concepts & Definitions - **Word's Coefficient:** A number assigned to a word indicating how strongly it predicts one of two concerning patterns: glorification of violence or suicidal ideation. - **Positive Coefficient:** Indicates a word is predictive or indicative of violence/suicide. - **Negative Coefficient:** Indicates a word suggests the opposite of the concerning patterns (e.g., "happy"). - **Score Threshold:** A specific value above which a message's final calculated score is deemed concerning enough to trigger an alert. ## Mechanisms & Processes - **AI Message Scoring:** A process where a message is broken into words, each word is assigned a coefficient, and these are combined (using an unspecified "not a simple addition expression") to create a final score between zero and one. - **Coefficient Calculation:** Requires vast amounts of data as input for computational techniques. - **Phraze Scoring:** Coefficients can be assigned to *groups of words* (a phrase) rather than just individual words, allowing context to override single-word ambiguity (e.g., "guns make me happy"). - **Sarcasm Handling:** AI models are trained on curated tweets from Millennial and Gen Z audiences to learn and apply conclusions to sarcastic patterns. - **App Monitoring:** The mobile application monitors a teen's outgoing text messages, runs the message through the model, and alerts a parent if the final score surpasses the established threshold. ## Timeline & Sequence - **2012:** Sandy Hook Elementary School shooting (9 children, 6 staff members died). - **2018:** Parkland shooting at Marjory Stoneman Douglas High School; inspired a national walkout. - **Three months later:** Santa Fe High School shooting. - **2019:** Dayton shooter linked to previous shooting in El Paso, Texas. ## Named Entities - **Sandy Hook Elementary School** — location of 2012 shooting; 20 children and 6 staff members died. - **Parkland, Florida** — location of 2018 shooting at Marjory Stoneman Douglas High School. - **Marjory Stoneman Douglas High School** — location of the 2018 shooting. - **Santa Fe High School** — location of subsequent shooting experience. - **Milton, Georgia** — location where the speaker lived, making gun violence seem distant. - **El Paso, Texas** — location of a previous shooting linked to the 2019 Dayton shooter. - **Dayton shooter** — perpetrator linked to previous shooting. - **Sandy Hook shooters** — perpetrators associated with the 2012 incident. - **Google/Social Media** — general platforms where data for analysis is sourced. ## Numbers & Data - Number of deaths at Sandy Hook: **20 children and 6 staff members**. - Age speaker remembered the first shooting: **Nine**. - Number of students involved in 2018 walkouts: **Over two million students**. - Number of minutes of silence held: **17 minutes**. - Data analyzed for patterns: **Past 55 years** of mass shooting data. - Percentage of shooters showing unusual interest in past shootings: **75%**. - Percentage of shooters showing signs of suicidal ideation: **80%**. - Amount of data collected for model training: **Over 1.6 million tweets**. - Final score range: **Between zero and one**. ## Tools, Tech & Products - **AI Models:** Three models built in **Python**. - **Mobile App:** Prototype monitoring outgoing text messages from a teen to a parent. - **Coefficient Scoring System:** The mechanism for quantifying linguistic risk. - **SNS monitoring:** Implicitly relates to social media platforms like Facebook and Instagram. ## Examples & Cases - **Sandy Hook:** Shooting at Sandy Hook Elementary School in 2012. - **Parkland:** Shooting at Marjory Stoneman Douglas High School in 2018. - **Santa Fe:** Shooting at Santa Fe High School, triggering a "wake-up call." - **Perpetrator evidence:** A perpetrator posting a Facebook photo wearing a T-shirt that said *born to kill* days before a shooting. - **Coefficient Example 1 (Positive):** In the phrase *"i was born to kill,"* the word *kill* has a coefficient of positive **0.8**. - **Coefficient Example 2 (Negative):** Words like *happy* or *life* would have a negative coefficient. - **Phrase Example:** The phrase *"guns make me happy"* (which might confuse a single-word model) has a final, highly positive score of **0.95**. - **Sarcasm Example:** The phrase *"math is so hard i want to die lol"* may not indicate genuine concern unless backed up by other messages. ## Trade-offs & Alternatives - **Traditional Activism:** Highly important, but sometimes insufficient when headlines change focus. - **Sole reliance on qualitative observation:** Misleading; drawing serious conclusions from limited signs is problematic. - **Technology Monitoring (App):** Addresses immediacy; but introduces ethical challenges regarding monitoring personal data. - **Data Scope:** Models must account for context (phrases/sarcasm) rather than just isolated words. ## Counterarguments & Caveats - The relationship between statistical patterns (75% interest in past shootings) and actual gun violence is *not causal*; something more complex is at play. - A single warning sign (e.g., T-shirt) or a positive score does not guarantee danger; context is needed. - **Ethical Concern:** Monitoring personal messages raises serious questions about user safety and privacy. - **Current Limitation:** The prototype is not yet a widely available solution; significant work remains. ## Methodology - Analysis of **55 years** of mass shooting data to identify patterns in shooters' past behaviors. - Using computational techniques to collect and analyze **over 1.6 million tweets**. - Building **three AI models in Python** to calculate coefficients for language patterns. - The final output is a score, indicating the degree of concern within a message. ## Conclusions & Recommendations - Society needs to recognize and utilize STEM solutions alongside political activism to tackle social issues like gun violence. - The final goal is ensuring that *never again* is not just a *trending hashtag* but a permanent reality. ## Implications & Consequences - Technology allows for a shift from purely reactive, post-tragedy activism to proactive, early-warning quantitative assessment of risk language. - Successful implementation could lead to systems that preemptively flag dangerous patterns in communication before violence occurs. ## Verbatim Moments - *"i remember hearing about a school shooting for the first time when i was nine."* - *"an infinite number of silent moments can't save lives."* - *"technology's greatest asset is analyzing large amounts of data and finding trends in a short amount of time."* - *"the word kill has a coefficient of positive 0.8."* - *"that whole phrase actually has a highly positive final score and it's therefore deemed concerning."* - *"the scoring process takes place directly on the phone. the content of the message will never leave the phone only a numerical score."* - *"we can all contribute to making sure that mass shootings truly occur never again."*