TEDx Talks - Demystifying AI: From medicine to infrastructure | Anita Raja | TEDxCUNY
The discussion highlights the use of AI in predicting preeclampsia, a condition affecting 5-7% of pregnant women and a leading cause of maternal deaths in the U.S. Traditional methods only detect it at 20 weeks, but AI can identify biomarkers as early as 11-13 weeks, allowing for preventive actions. The AI algorithms developed by the team at Hunter College and Columbia University integrate data from multiple sources to find subtle dependencies between features and outcomes. This approach has been successful, although initial biases in the data were corrected to improve accuracy for underrepresented groups. The methodology is versatile and can be applied to other medical and non-medical problems, such as predicting infrastructure failures like pipeline bursts in New York City. The speaker emphasizes the importance of responsible AI, ensuring data integrity, transparency, and avoiding misinformation, to harness AI's potential for societal benefit.
Key Points:
- AI can predict preeclampsia as early as 11-13 weeks into pregnancy, allowing for early intervention.
- The approach integrates data from multiple sources to identify subtle dependencies and predict outcomes.
- Initial biases in AI models were corrected to improve accuracy for underrepresented groups.
- The methodology can be applied to other fields, such as predicting infrastructure failures.
- Responsible AI involves ensuring data integrity, transparency, and avoiding misinformation.
Details:
1. 🔍 Understanding Preeclampsia and its Challenges
- Preeclampsia affects 5 to 7% of pregnant women, leading to significant health issues and being the primary cause of maternal deaths in the United States.
- Current diagnostic methods predict preeclampsia only at 20 weeks of pregnancy, which is often too late for effective intervention, highlighting the urgent need for earlier detection methods.
- The severe emotional and societal impacts of preeclampsia necessitate the development of preventive measures.
- AI has the potential to revolutionize preeclampsia detection by identifying risk factors much earlier in pregnancy, thereby allowing timely interventions and reducing both maternal and fetal risks.
2. 🧬 Innovative AI Approaches to Predict Preeclampsia
2.1. Economic Impact and Strategic Approach
2.2. Advancements in Biomarker Identification
3. 🤖 Demystifying Artificial Intelligence
- AI algorithms developed at Hunter College are used to tackle complex healthcare issues, such as preeclampsia, demonstrating AI's transformative potential in medicine.
- These algorithms are likened to clever assistants, adept at reasoning about objectives, managing deadlines, and optimizing costs, thereby effectively addressing intricate problems across various fields.
- While AI garners significant interest and enthusiasm, it is often portrayed in the media as a dystopian threat, emphasizing the need for clearer understanding and communication of AI's realistic capabilities and future potential.
- Beyond healthcare, AI applications extend to areas such as finance, where they help optimize trading strategies, and transportation, where they improve route planning and traffic management.
- Understanding the mechanics of AI algorithms enhances their application, allowing for more tailored solutions in diverse sectors and fostering innovation.
4. 🧠 The Evolution and Capabilities of AI
- AI endeavors to develop machines with intelligent behaviors, inspired by the brain's functions like perception, learning, and abstract thinking.
- AI technologies, such as generative AI and neural networks, are influenced by the brain's neuron structure but don't aim to replicate it fully.
- Humans are superior in creativity, common sense reasoning, and adaptability, while AI specializes in identifying patterns in extensive data sets rapidly.
- AI algorithms, despite being powerful, lack the deep understanding and comprehensive cognition that humans possess.
- The interest in AI dates back to the 1950s, with notable advancements from projects like DARPA, illustrating the historical pursuit of AI capabilities.
- The initial wave of AI involved rule-based systems effective in specific domains like medical diagnosis and chess but limited outside their programmed rules.
- Current AI applications include image and speech recognition, autonomous vehicles, and personalized recommendations, showcasing its evolving capabilities.
5. 📈 Machine Learning and AI Applications in Daily Life
- Second-generation AI systems tackle knowledge gaps by learning and adapting without explicit instructions, employing advanced statistical methods to detect data patterns.
- Machine learning enhances daily tasks such as tagging friends in photos by learning from historical data to recognize individuals in new images.
- AI-driven recommendation systems, like those on streaming platforms, boost product sales by analyzing user viewing habits to suggest relevant content.
- User feedback is vital for refining AI outputs as skewed data may result in inaccurate recommendations.
- Generative algorithms, used in chatbots, are trained on extensive text datasets to predict subsequent words, enabling seemingly intelligent interactions.
- AI algorithms sometimes produce incorrect results, known as 'hallucinations,' underscoring the need for accuracy and bias assessments.
- In healthcare, AI systems optimize patient diagnosis and treatment plans by analyzing large datasets, improving outcomes and efficiency.
- Financial sectors leverage AI for fraud detection and personalized customer service, enhancing security and user experience.
6. 🏥 Applying AI to Medical Challenges: Preeclampsia Case Study
6.1. General Applications of AI and Machine Learning in Medical Fields
6.2. Case Study: AI in Preeclampsia Diagnosis
7. 🔬 Addressing Bias and Improving AI Models
7.1. Early Detection of Preeclampsia
7.2. Addressing Bias in AI Models
7.3. Implications for Future AI Development
8. 🌆 AI in Infrastructure: Predicting Pipeline Failures
- On August 29, 2023, a 20-inch pipeline burst under Times Square, releasing 1.8 million gallons of water into the subway, disrupting 300,000 passengers.
- New York City's infrastructure includes 7,400 miles of pipes, many dating back to the 1890s, necessitating repair or replacement.
- Factors determining pipe replacement include age, diameter, material, soil resistivity, pressure zones, and underground root growth.
- Lack of accurate maps complicates New York's pipe maintenance and replacement efforts.
- AI is employed to integrate various data sources to predict likely pipe bursts, aiming for proactive maintenance.
- Mathematical methods, previously used for predicting preeclampsia, are applied to forecast pipeline failures, targeting timely repairs to prevent incidents.
9. 🌐 Responsible AI and its Future Impact
- AI technologies offer unprecedented potential but also pose challenges that must be addressed through responsible AI practices.
- Key challenges include ensuring data comes from trusted sources, avoiding intentional bias, ensuring transparency and interpretability in AI reasoning, and preventing the spread of misinformation.
- Responsible AI involves both technical and governance issues, requiring the creation of algorithms that contain risks while allowing for innovation.
- If implemented responsibly, AI can empower individuals and help foster a resilient society where technology and personal aspirations coexist harmoniously.
- Successful examples of responsible AI include systems that incorporate feedback loops to continuously improve accuracy and fairness.
- Governance frameworks are essential to oversee AI development, ensuring compliance with ethical standards and legal requirements.