Limits of machine learning

Capire il Machine Learning

Sara Billen

Curriculum Manager, DataCamp

Data quality

Garbage in garbage out

  • Garbage in garbage out
  • Output quality depends on input quality
Capire il Machine Learning

How it can go horribly wrong

Amazon's gender-biased recruiting tool

Robot looking at people with magnifying glass

  • Recruiting software to help review resumes
  • Preferred men because it learned from historic data when more men were hired
  • It downgraded resumes that
    • contain the word "women"
    • implied the applicant was female
Capire il Machine Learning

How it can go horribly wrong

Microsoft's AI chatbot

Tay's twitter account

Tay tweet examples

Capire il Machine Learning

Beware

caution sign

  • Don't blindly trust your model
  • Awareness is key
  • Pay attention to your data

$$

A machine learning model is only as good as the data you give it

Capire il Machine Learning

Quality assurance

  • High-quality data requires:
    • Data analysis
    • Review of outliers
    • Domain expertise
    • Documentation

magnifying glass over data

Capire il Machine Learning

Explainability

Capire il Machine Learning

Explainability

  • Transparency to increase trust, clarity, and understanding
  • Use cases: business adoption, regulatory oversight, minimizing bias
Capire il Machine Learning

Explainable AI

Black box

  • Deep learning
  • Better for "What?"
  • Highly accurate predictions

Explainable AI

  • Traditional machine learning
  • Better for "Why?"
  • Understandable by humans
Capire il Machine Learning

Example: Explainable AI

insurance-customer-churn-analytics.jpg

  1. Prediction: Will the patient get diabetes?
  2. Inference: Why will this happen
Capire il Machine Learning

Example: Inexplicable AI

Handwriting.jpg

Prediction only: Which letter is this likely to be?

Capire il Machine Learning

Let's practice!

Capire il Machine Learning

Preparing Video For Download...