Engineering TED Talk: Computer Science in Social Justice

Columbia Engineering: TED TALK

Engineering and social justice: Computer Science

1st speaker: Julia Hirschberg

• Identification of hate speech (friend vs. stranger hate speech)
• AI then and now:
  • Original goal: create machines with human intelligence in reasoning, NLP, robotics, vision (machines that can replace humans)
  • Today AI has applications to many areas: healthcare, education, entertainment, sustainability, transportation, and commerce
  • But, rather than replication/replacing human intelligence, the term “collaborative AI” is becoming popular–how can AI help humans, not replace them
• More positive contributions of AI:
  • Virtual reality to study, treat, and simulate autism traits
  • Snapchat recently helped over 400,00 new voters register to vote
  • New predictive model for disaster relief, smart agriculture, medicine delivery, and education in developing countries
  • Ai remove bias from judging for 2020 olympic gymnastics
  • AI can provide electronic strike zone in baseball and many other sports
  • Computer vision techniques helping customers choose makeup colors by matching their picture to one of 40 shades → fashion & computer science
• AI faces many challenges:
  • Self-driving cars are not yet safe
  • AI are taking over people’s jobs
  • AI can invade privacy and create & circulate fake news
  • Deep learning systems perpetuate biases of the data they are trained on (MT, job search, face recognition)
    • Face recognition: dark faces are gorillas or just do not appear
  • Machine translation in pronominally gender-neutral languages’ pronouns: Doctors and programmers are men; nurses and homemakers are women
  • Software that warns people using Nikon cameras when the person they are photographing seems to be blinking tends to interpret asians as always blinking
  • Facebook ads have targeted particular genders or ethnicities for jobs, excluding women and ethnic minorities
• AI software is being used to make serious decisions on:
  • Loan-worthiness
  • Emergency response
  • Medical diagnosis
  • Job candidate selection
  • Parole determination
  • Criminal punishment
  • Educator performance
  • ^often without user awareness of its limitations

2nd speaker: Kathy McKeown

• Objective: develop a system to automatically detect aggression and loss in social media posts by gang-involved  youth
  • Challenges: size of labeled dataset
  • Domain-specific language
  • Context critical

3rd speaker: Dr. Shih-Fu Chang

• Experts believe the increased use of social media among gang-involved youth may be an important factor in the uptick in gang violence in cities across America
• Imagine a world where social media yields clues that identify risk and protective factors for gang violence and prevent the use of firearms
• Some projects he is working on:
  • Image processing/AI/computer vision to understand gang violence
  • Visual search technology for fighting online human trafficking: try to understand illegal information on the dark web, used by 200+ law enforcement agencies or NGOs to locate victors or identify groups engaged in human trafficking