Machine Learning And Ted Talk

Question 1

What is machine learning?

Answer 1

A type of artificial intelligence. Develops algorithms that can preform tasks without explicit programming

Question 2

Chartwatch

Answer 2

Al-based tool that monitors 100 variables from patients charts to determine risk of needing ICU care every hour

Question 3

Violence risk prediction

Answer 3

ML algorithm was fairly accurate in predicting which patients would become violent
Strongest predictors were homeless and prior assault

Question 4

How does supervised ML work

Answer 4

1. Start with health records for thousands of patients from psychiatric hospital
2. Spilt into two ( train set larger than test set)
3. Train algorithm on features to predict the outcome (in the training set)
4. Test the performance of algorithm (in the test set)

Question 5

What could possibly go wrong with MI?

Answer 5

It can generate biased predictions due to inherent biases in the training DATA

Question 6

What are the types of biases involved in ML

Answer 6

Diagnostic bias and legal bias

Question 7

Diagnostic bias

Answer 7

Patients in data set may have been wrongfully diagnosed leading to a wrongful ML prediction
Ex. Black patients are diagnosed with schizophrenia at a higher rate than other groups

Question 8

Legal bias

Answer 8

Patients in data set may reside closer to police officers
Ex. Black patients have more unfair encounters with police leading to more convictions

Question 9

Outcomes of ML

Answer 9

ML applications are exciting and could improve areas of our lives
Also potential harmful consequences

Question 10

Modern examples of ML

Answer 10

Alexa
Spotify
Netflix