Please enable JavaScript.
Coggle requires JavaScript to display documents.
When Machine Learning goes off the rails - Coggle Diagram
When Machine Learning goes off the rails
What Makes Machine Learning Risky
Concept drift
inputs the systems uses and its outputs isn't stable over time or may be misspecified.
ex:
the color of someone's skin
Covariate shift
machine learning based system using data from large urban hospitals
but using in rural areas
Risk ==> inaccurate decision
Agency Risk
That aren't under the control of a specific business or user.
Because it is a complex system.
The data may come from third-party vendors.
the doctor would most likely be held liable for any harm only if he or she did not the system/
the liability risks from doctors to the developers of the machine-learning-enable medical device and the data providers
Moral Risk
Should Tesla program the car ?
How should we program a car to value the lives of three elderly people against, say , the life of one middle-age person?
How should business balance trade-offs among, say, privacy, fairness, accuracy , and security?
Biases related to demographic groups.
White people v.s Black people
machine-learning systems may be deemed unfair to a certain group on some dimensions.
To Lock Or Not To Lock
company allow it to continuously evolve
FDA (medical device)
they don't want to permit the use of devices whose diagnostic procedures or treatment pathways keep changing in ways .
lock can' move the following dangers:
Inaccurate decisions ( unsure system change)
Environmental changes (Car)
Agency risk ( third vendors)
Moral risks (biases)
only tested and locked versions at intervals
A Tool Kit For Executives
How should executives manage the existing and emerging risks of machine learning?
Treat machine learning as if it's human.
Demand employee learn to use the SYSTEM
Compare Employee and System
Think like a regulator and certify first.
Before go to market, business should certify machine-learning
developing standards for such certification
Monitor continuously
set up ways to check the technology behave within appropriate limits.
Ask the right questions.
accuracy and competitiveness.
if we don't lock, how much improve with the volume of new data in machine-learning system.
biases
what data used to train.
how to predict unlock will produce less-biased results than lock ??
the environment
how will the environment in which the offering is used change over time?
agency
third-party components, including data sources.
should we allow other organizations to use machine-learning which we develop?
Idea In Brief
Problem
machine learning systems don't always make ethical or accurate choices.
Causes
the system often make decisions based on probabilities.
the environment may evolve in an unanticipated way
the complexity makes it difficult to determine whether or why they made a mistake
Solutions
Test the system appropriately
monitor