22b9d24a81
|
||
|---|---|---|
| .. | ||
| README.md | ||
README.md
Ethical and Responsible AI
You don almost finish dis course, and I dey hope say by now you don sabi well well say AI na somtin wey base on plenty formal mathematical methods wey fit help us find relationship for data and train models to dey do some kain tins wey human beings dey do. For dis time wey we dey so, we dey see AI as one strong tool wey fit help us see pattern for data, and use dat pattern solve new problems.
Pre-lecture quiz
But for science fiction, we dey always see stories wey talk say AI fit be danger to human beings. Most times, dem dey talk say AI go rebel, like say AI wan fight human beings. Dis kain story dey make am look like say AI get emotion or say e fit make decision wey e developers no plan for.
Di kain AI wey we don learn for dis course no be anytin pass big big matrix arithmetic. E be strong tool wey fit help us solve our problems, and like any other strong tool - e fit dey used for good or bad purpose. But di main tin be say, e fit dey misused.
Principles of Responsible AI
To make sure say AI no go dey misused by mistake or on purpose, Microsoft don talk about di important Principles of Responsible AI. Di following ideas dey under dis principles:
-
Fairness dey talk about di big problem of model biases, wey fit happen if we use data wey don already get bias to train AI. For example, if we wan predict di chance wey person get to get software developer job, di model fit prefer men pass women - just because di training data wey we use fit don dey biased towards men. We need balance di training data well and check di model to make sure say e no dey biased, and make sure say di model dey use di correct features.
-
Reliability and Safety. AI models fit make mistake because of di way dem be. Neural network dey return probabilities, and we need remember dis one when we dey make decisions. Every model get di own precision and recall, and we need understand am to stop di harm wey wrong advice fit cause.
-
Privacy and Security get some AI-specific wahala. For example, di data wey we use train model go somehow dey "inside" di model. On one side, e dey increase security and privacy, but on di other side - we need remember di data wey di model take train.
-
Inclusiveness mean say we no dey build AI to replace people, but to help people and make our work dey more creative. E still join fairness, because when we dey deal with underrepresented communities, di data wey we collect fit dey biased, and we need make sure say dem dey included and di AI dey handle dem well.
-
Transparency. Dis one mean say we go always make sure say people sabi say AI dey used. Plus, anywhere we fit, we go wan use AI systems wey people fit understand.
-
Accountability. When AI models dey make decisions, e no dey always clear who go take responsibility for di decision. We need make sure say we sabi who go take responsibility for di AI decisions. Most times, we go wan put human beings for di process of making important decisions, so dat na real people go dey accountable.
Tools for Responsible AI
Microsoft don create di Responsible AI Toolbox wey get plenty tools:
-
Interpretability Dashboard (InterpretML)
-
Fairness Dashboard (FairLearn)
-
Error Analysis Dashboard
-
Responsible AI Dashboard wey include:
- EconML - tool for Causal Analysis, wey dey focus on "what-if" questions
- DiCE - tool for Counterfactual Analysis wey go show you di features wey you need change to affect di model decision
If you wan sabi more about AI Ethics, check dis lesson for di Machine Learning Curriculum wey get assignments.
Review & Self Study
Follow dis Learn Path to learn more about responsible AI.
Post-lecture quiz
Disclaimer:
Dis dokyument don translate wit AI translation service Co-op Translator. Even though we dey try make am accurate, abeg sabi say automated translation fit get mistake or no correct well. Di original dokyument for im native language na di main correct source. For important information, e beta make professional human translator check am. We no go fit take blame for any misunderstanding or wrong interpretation wey fit happen because you use dis translation.