1.A-to-B Automation (Supervised Learning): Mapping clear inputs to specific outputs, such as classifying an email as spam or predicting if a user will click an online ad.
2.Generate Text (Generative AI/LLMs): Predicting the next logical word in a sequence based on vast amounts of internet data to act as a copy editor, summarizer, or brainstorming partner.
3.Visual Inspection: Quickly identifying visual anomalies, such as spotting a scratch on a smartphone rolling off an assembly line.
4.Speech Translation and Recognition: Transcribing audio to text or translating English to another language in near real-time.
5.Medical Diagnosis on Standardized Data: Accurately identifying conditions like pneumonia, provided it has been trained on thousands of consistent, high-quality images.
1.Artificial General Intelligence (AGI): AI cannot perform any and all intellectual tasks a human can; we are still decades away from true AGI.
2.Predict Fundamentally Random Systems: It cannot accurately predict future stock market prices based solely on historical data.
3.Learn from Small Data (Few-Shot Learning): AI cannot master a new subject by reading a textbook and looking at 10 pictures the way a human student can.
4.Interpret Nuanced Human Intentions: It struggles with complex, highly variable human behaviors, such as accurately decoding what a pedestrian waving at a self-driving car wants it to do.
5.Adapt to Unseen or Messy Data: AI lacks human robustness; if it is trained on pristine data and then encounters slightly skewed or differently formatted data in the real world, its performance plummets.
1.What Fintech AI Can Do (Fraud Detection): Using supervised learning, an AI can take transaction details (Input A: location, amount, time) and accurately flag suspicious activity (Output B: Is this fraud? 0/1) faster than a human reviewer.
2.What Fintech AI Cannot Do (Historical Market Prediction): As highlighted in the videos, an AI cannot look solely at the historical price chart of a cryptocurrency or stock (Input A) and reliably predict its price exactly one month from now (Output B), because the market is influenced by unpredictable, real-world variables outside of historical price points.
3.UPI/PhonePe transaction data is structured and high volume — ideal conditions for AI. This makes fraud detection, credit scoring, and spend categorisation strong AI PM opportunities in Indian fintech.