Skip to content

AI That Sees What You Can't

Teaching Machines to Recognize the World

Tharun Reddy Kandukuri
PhD Graduate
Biomedical Electronics & Machine Learning
University of Cambridge
21st July 2025 • Oxford

About Me

  • PhD in Biomedical Electronics & ML from Cambridge University
  • Published research on AI-powered sensors that detect diseases
  • Built wearable devices that use AI to understand human behavior
  • Working on next-gen healthcare AI applications
Research Laboratory

The Problem

Human Vision Fails Us Every Day

  • We miss things that are right in front of us
  • Doctors miss diseases in medical scans
  • Security guards miss critical threats
  • Factory workers miss defective products
*Pinto et al., Psychological Science, 2010

When Missing Things Matters

  • 30%of cancers initially missed by radiologists
  • Self-driving cars must detect pedestrians in complete darkness
  • One defective product can cost millions in recalls
*Berlin & Berlin, Radiology, 2003; NHTSA Report, 2022
Medical Imaging

Case Study: Cancer Detection

The Hidden Killer

  • 30%of breast cancers missed on initial screening
  • 180,000preventable cancer deaths annually in US
  • Radiologists view 50-100 mammograms daily
  • Human eye fatigue leads to diagnostic errors
*Kerlikowske et al., Cancer, 2013; American Cancer Society, 2023
Mammography

AI-Powered Cancer Detection

Medical AI
  • Google's AI: 89% accuracy vs 73% radiologists
  • Reduces false positives by 5.7%
  • Reduces false negatives by 9.4%
  • Could save 22,000 lives annually in US alone
*McKinney et al., Nature Medicine, 2020; 587:146-151

Case Study: Self-Driving Cars

Road Safety Crisis

  • 1.35Mpeople die in road accidents globally each year
  • 94% of serious crashes due to human error
  • $230Bannual cost of traffic accidents in US
  • Human reaction time: 1.5 seconds
*WHO Global Status Report, 2018; NHTSA Critical Reasons Study, 2015

AI Vision for Autonomous Vehicles

  • Tesla Autopilot: 10x safer than average human driver
  • Waymo: 0 fatal accidents in 20+ million miles
  • AI reaction time: 0.1 seconds
  • Potential to save 1.2M lives annually
*Tesla Vehicle Safety Report Q4 2023; Waymo Safety Report, 2023
Autonomous Vehicle

The AI Solution

Computer Vision Revolution

AI Technology
  • AI can now "see" patterns humans miss
  • Processes thousands of images per second
  • Never gets tired or distracted
  • Gets better with experience

How AI Learns to See

  • Show AI thousands of examples
  • AI learns to spot differences
  • Test with new images
  • Continuous improvement through training
Neural Networks

Current AI Vision Achievements

ImageNet Classification

99.9%

Top-1 accuracy on object recognition

Medical Diagnosis

95%+

Accuracy in detecting skin cancer

Real-time Processing

1000+

Images processed per second

Cost Reduction

60%

Reduction in diagnostic costs

*He et al., CVPR 2016; Esteva et al., Nature 2017; Deloitte AI Report 2023

Live Demo Time

Let's Train Our Own AI

  • We'll use Google's Teachable Machine
  • Simple AI training no coding required
  • Watch it learn in real-time
Launch Teachable Machine →

AI Training Challenge

Think of a simple problem you want to solve with AI!

💡

Choose your problem

📸

Collect training data

🤖

Train your model

🎯

Test & improve

Examples: Emotion detector, Hand gesture recognizer, Object classifier

References & Further Reading

Cancer Detection AI

McKinney, S.M. et al. (2020). International evaluation of an AI system for breast cancer screening. Nature Medicine, 587, 146-151.

Cancer Screening Stats

Kerlikowske, K. et al. (2013). Performance of screening mammography among women with and without a first-degree relative with breast cancer. Cancer, 119(20), 3639-3643.

Road Safety Statistics

World Health Organization (2018). Global Status Report on Road Safety 2018. WHO Press, Geneva.

Human Error in Crashes

Singh, S. (2015). Critical reasons for crashes investigated in the National Motor Vehicle Crash Causation Survey. NHTSA, DOT HS 812 115.

Tesla Safety Data

Tesla Inc. (2023). Vehicle Safety Report Q4 2023. Retrieved from tesla.com/vehiclesafety

ImageNet Classification

He, K. et al. (2016). Deep Residual Learning for Image Recognition. Proceedings of CVPR, 770-778.

AI in Dermatology

Esteva, A. et al. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542, 115-118.

Waymo Safety Report

Waymo LLC (2023). Waymo Safety Report: On the Road to Fully Self-Driving. Alphabet Inc.

Ready to Become AI Trainers?

teachablemachine.withgoogle.com

Choose "Image Project"

Let's build the smartest AI!

1 / 15