Data Ethics in Machine Learning and Data Mining
What is data ethics?
Nowadays we can see lots of devices. People are creating new innovative devices or digital and technological products for various purposes. So, people are more connected with those devices and those devices relate to some other devices or technologies. So, they are collecting data and generating data and also, they are processing data. And the growth of these data output is rapidly increasing. So basically, in the data ethics we work in the direction of the ethical implications of data generation and data analysis.
There are set of moral standards that govern the use of computers called Computer Ethics which explicitly looked at the ethical implications of various types of computer devices. But in data ethics we only concern about data.
Basically, data ethics inspired by three things.
(Big) Data + Algorithms + Data-driven Practice = Data Ethics
Generation and the ability to process greater amounts and greater types of data than has ever been possible before. To do that processing we need algorithmic systems to make sense of data which can be automated, and which can be human run. In other words, we need new ways of looking at the data. Finally, once we have the data and once we have a way to actually look at it, there’ll be lots of different practices that are now becoming increasingly data driven. In other words, the traditional sort of practices, that we have things like medicine are now increasingly being driven by the insights that we can generate from this data producing.
Ethics of Data Science
Ethics of Data
- Privacy
→Re-identification
→Group privacy
- Trust in whom?
- Transparency of what?
Ethics of Algorithms
- Responsibilities and accountability
- Ethical design of algorithms requirements
- Ethical auditing of algorithms
Ethics of Practices
- Deontological code
- Consent
- Privacy of data subjects
- Secondary use
Why Ethics?
Data science provides huge opportunities to improve private and public life. It is coupled to significant ethical challenges, such as:
- Fairness
- Responsibility
- Respect of human rights
Current issues in Machine Learning algorithms
- Hard to explain the final decision to users since ML systems look like black boxes (NN-based algorithms).
- Some of the current ML algorithms behave unfair.
- DM/ML systems need to be used by professionals outside engineering/math communities.
- DM/ML systems should be incorporated into social and legal systems.
