Ethical machines: data mining
and fairness
– the optimistic view
Anna Ronkainen
chief scientist, TrademarkNow
it’s complicated, UU of Helsinki & Turku
@ ronkaine
2016-05-02
My three points
1.  people aren’t exactly perfect, either, and
sometimes algorithms can be an
improvement
2.  different types of algorithms needed for
arriving at decisions and validating/
disproving them
3.  data protection law about automated
decision-making needs to be taken
seriously
Heuristics
or biases?
(Dhami 2003)
Sometimes people fail in unexpected
ways...
(Danziger et al (2011):	Extraneous Factors
in Judicial Decisions)
Systems 1 and 2 in legal reasoning:
interaction
System 1:
making the
decision
System 2:
validation and
justification
(Ronkainen	2011)
Implications for algorithms
(hypothesis)
-  System-1-like processes cannot be captured
reliably with GOFAI -> machine learning and
other statistical approaches needed
-  the System 2 part (finding supporting
arguments and validating/falsifying the
decision candidate) can (and should) be
implemented with rule-based GOFAI for
accountability, maintainability etc etc etc
Taking data protection seriously?
(2016 EU General Data Protection Regulation)
Seriously-seriously?
(1995 EU Data Protection Directive 95/46/EC)
My three points
1.  people aren’t exactly perfect, either, and
sometimes algorithms can be an
improvement
2.  different types of algorithms needed for
arriving at decisions and validating/
disproving them
3.  data protection law about automated
decision-making needs to be taken
seriously
Thank you!

Ethical machines: data mining and fairness – the optimistic view

  • 1.
    Ethical machines: datamining and fairness – the optimistic view Anna Ronkainen chief scientist, TrademarkNow it’s complicated, UU of Helsinki & Turku @ ronkaine 2016-05-02
  • 2.
    My three points 1. people aren’t exactly perfect, either, and sometimes algorithms can be an improvement 2.  different types of algorithms needed for arriving at decisions and validating/ disproving them 3.  data protection law about automated decision-making needs to be taken seriously
  • 3.
  • 4.
    Sometimes people failin unexpected ways... (Danziger et al (2011): Extraneous Factors in Judicial Decisions)
  • 5.
    Systems 1 and2 in legal reasoning: interaction System 1: making the decision System 2: validation and justification (Ronkainen 2011)
  • 6.
    Implications for algorithms (hypothesis) - System-1-like processes cannot be captured reliably with GOFAI -> machine learning and other statistical approaches needed -  the System 2 part (finding supporting arguments and validating/falsifying the decision candidate) can (and should) be implemented with rule-based GOFAI for accountability, maintainability etc etc etc
  • 7.
    Taking data protectionseriously? (2016 EU General Data Protection Regulation)
  • 8.
    Seriously-seriously? (1995 EU DataProtection Directive 95/46/EC)
  • 9.
    My three points 1. people aren’t exactly perfect, either, and sometimes algorithms can be an improvement 2.  different types of algorithms needed for arriving at decisions and validating/ disproving them 3.  data protection law about automated decision-making needs to be taken seriously
  • 10.