The document discusses Hildebrandt's taxonomy of code-driven law, data-driven law, and text-driven law. It notes that code-driven law involves legal norms articulated in computer code, while data-driven law uses automatic decision-making derived from statistical/inductive methods. Text-driven law refers to legal activity performed by humans using sources like statutes and case law. The document also examines various issues that can arise with computational approaches to law, such as whether training data adequately represents the problem domain and whether all relevant facts and rules are considered. It acknowledges limitations in fully formalizing concepts like justice and cautions that computational ideals assume problems may be solved perfectly.

1. code-driven law?
Workshop on AI and border control, Edinburgh Law School (virtual), 15 June 2021
Giovanni Sileno, University of Amsterdam. g.sileno@uva.nl

2. Hildebrandt’s taxonomy
code-driven law
data-driven law
text-driven law
Hildebrandt, M., Code Driven Law. Scaling the Past and Freezing the Future, In: Is Law Computable? Critical Perspectives on Law
and Artificial Intelligence, eds. Markou, Deakin, Hart Publishers (2020)
legal norms or policies that have
been articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law
automatic decision-making
or predictions used in support derived
from statistical/inductive methods

3. Hildebrandt’s taxonomy
code-driven law
data-driven law
text-driven law
Hildebrandt, M., Code Driven Law. Scaling the Past and Freezing the Future, In: Is Law Computable? Critical Perspectives on Law
and Artificial Intelligence, eds. Markou, Deakin, Hart Publishers (2020)
legal activity performed by humans by
means of sources of norms such as
statute and case law
open-textured concepts
multi-interpretability
→ it can be contested
legal norms or policies that have
been articulated in computer code
automatic decision-making
or predictions used in support derived
from statistical/inductive methods

4. Hildebrandt’s taxonomy
code-driven law
data-driven law
text-driven law
Hildebrandt, M., Code Driven Law. Scaling the Past and Freezing the Future, In: Is Law Computable? Critical Perspectives on Law
and Artificial Intelligence, eds. Markou, Deakin, Hart Publishers (2020)
legal activity performed by humans by
means of sources of norms such as
statute and case law
open-textured concepts
multi-interpretability
→ it can be contested statistical closure logical closure
legal norms or policies that have
been articulated in computer code
automatic decision-making
or predictions used in support derived
from statistical/inductive methods

5. Hildebrandt’s taxonomy
code-driven law
data-driven law
text-driven law
Hildebrandt, M., Code Driven Law. Scaling the Past and Freezing the Future, In: Is Law Computable? Critical Perspectives on Law
and Artificial Intelligence, eds. Markou, Deakin, Hart Publishers (2020)
legal activity performed by humans by
means of sources of norms such as
statute and case law
open-textured concepts
multi-interpretability
→ it can be contested statistical closure logical closure
legal certainty +
legal norms or policies that have
been articulated in computer code
automatic decision-making
or predictions used in support derived
from statistical/inductive methods

6. Hildebrandt’s taxonomy
code-driven law
data-driven law
text-driven law
Hildebrandt, M., Code Driven Law. Scaling the Past and Freezing the Future, In: Is Law Computable? Critical Perspectives on Law
and Artificial Intelligence, eds. Markou, Deakin, Hart Publishers (2020)
legal activity performed by humans by
means of sources of norms such as
statute and case law
open-textured concepts
multi-interpretability
→ it can be contested statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
legal norms or policies that have
been articulated in computer code
automatic decision-making
or predictions used in support derived
from statistical/inductive methods

7. Hildebrandt’s taxonomy
code-driven law
data-driven law
text-driven law
Hildebrandt, M., Code Driven Law. Scaling the Past and Freezing the Future, In: Is Law Computable? Critical Perspectives on Law
and Artificial Intelligence, eds. Markou, Deakin, Hart Publishers (2020)
legal activity performed by humans by
means of sources of norms such as
statute and case law
open-textured concepts
multi-interpretability
→ it can be contested statistical closure logical closure
COMPUTATIONAL LEGALISM
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
legal norms or policies that have
been articulated in computer code
automatic decision-making
or predictions used in support derived
from statistical/inductive methods

8. code-driven law
data-driven law
text-driven law
legal activity performed by humans by
means of sources of norms such as
statute and case law
legal norms or policies that have
been articulated in computer code
automatic decision-making
or predictions used in support derived
from statistical/inductive methods
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?

9. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
in so far humans are involved
data-guided law
automatic decision-making
derived from statistical/inductive
methods

10. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
instance + label
instance + label
instance + label
instance + label
…..
TRAINING
DATASET

11. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
instance + label
instance + label
instance + label
instance + label
…..
is the training data set
representative of
the domain in which
it will be applied?
is the rationality behind labeling
still the same?
TRAINING
DATASET

12. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
instance + label
instance + label
instance + label
instance + label
…..
is the training data set
representative of
the domain in which
it will be applied?
is the rationality behind labeling
still the same?
TRAINING
DATASET
is the training method
producing adequate,
reproducible, robust results?
TRAINING
METHOD

13. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
instance + label
instance + label
instance + label
instance + label
…..
is the training data set
representative of
the domain in which
it will be applied?
is the rationality behind labeling
still the same?
TRAINING
DATASET
is the training method
producing adequate,
reproducible, robust results?
TRAINING
METHOD
→ it can be contested

14. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
fact
fact
….
STRUCTURAL
rule
rule
…..
CONTINGENT

15. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
fact
fact
….
STRUCTURAL
rule
rule
…..
CONTINGENT
were all the relevant rules considered?
were all the relevant facts given?
is the formalization correct?
[is the reasoning tool robust?]

16. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty +
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
fact
fact
….
STRUCTURAL
rule
rule
…..
CONTINGENT
→ it can be contested
were all the relevant rules considered?
were all the relevant facts given?
is the formalization correct?
[is the reasoning tool robust?]

17. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL LEGALISM ideal works
only in so far we forget all what may go wrong

18. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
is this all?

19. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) +
instrumentality (contextual policy goals) +
data-guided law
automatic decision-making
derived from statistical/inductive
methods
those principles
could be to some
extent formalized as
meta-rules

20. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) +
instrumentality (contextual policy goals) +
data-guided law
automatic decision-making
derived from statistical/inductive
methods
those principles
could be to some
extent formalized as
meta-rules
these goals
could be made explicit

21. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) +
instrumentality (contextual policy goals) +
data-guided law
automatic decision-making
derived from statistical/inductive
methods
those principles
could be to some
extent formalized as
meta-rules
these goals
could be made explicit
can we be sure about a certain
formalization?

22. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) +
data-guided law
automatic decision-making
derived from statistical/inductive
methods
those principles
could be to some
extent formalized as
meta-rules
these goals
could be made explicit
never.
can we be sure about a certain
formalization?
(cf. discussions on
universal human
rights, ethical
frameworks)

23. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) +
data-guided law
automatic decision-making
derived from statistical/inductive
methods
those principles
could be to some
extent formalized as
meta-rules
these goals
could be made explicit
may we be ever ready for that?
(explicit articulation of
policy-maker purposes)
never.
can we be sure about a certain
formalization?
(cf. discussions on
universal human
rights, ethical
frameworks)

24. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
those principles
could be to some
extent formalized as
meta-rules
these goals
could be made explicit
may we be ever ready for that?
(explicit articulation of
policy-maker purposes)
not to a full extent.
never.
can we be sure about a certain
formalization?
(cf. discussions on
universal human
rights, ethical
frameworks)
(it may require to be transparent on
highly contestable issues)

25. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
statistical closure logical closure
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
those principles
could be to some
extent formalized as
meta-rules
these goals
could be made explicit
may we be ever ready for that?
(explicit articulation of
policy-maker purposes)
not to a full extent.
never.
can we be sure about a certain
formalization?
(cf. discussions on
universal human
rights, ethical
frameworks)
(it may require to be transparent on
highly contestable issues)
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”

26. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”
the COMPUTATIONAL LEGALISM ideal
works only in so far we forget all what
may go wrong

27. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”
the COMPUTATIONAL LEGALISM ideal
works only in so far we forget all what
may go wrong
let’s accept things may go wrong at all level

28. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”
the COMPUTATIONAL LEGALISM ideal
works only in so far we forget all what
may go wrong
let’s accept things may go wrong at all level
people or other actors should
be able to appeal

29. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”
the COMPUTATIONAL LEGALISM ideal
works only in so far we forget all what
may go wrong
let’s accept things may go wrong at all level
let’s accept we need to integrate
systematically feedbacks from
higher-order courts, jurisprudence or
other normative sources
people or other actors should
be able to appeal

30. code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”
the COMPUTATIONAL LEGALISM ideal
works only in so far we forget all what
may go wrong
let’s accept things may go wrong at all level
let’s accept we need to integrate
systematically feedbacks from
higher-order courts, jurisprudence or
other normative sources
people or other actors should
be able to appeal courts or other actors should
be able to cassate

31. including computational actors
possibility of continuous, automated testing/verification
code-driven law
data-driven law
text-guided law
legal norms or policies that have been
articulated in computer code
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”
the COMPUTATIONAL LEGALISM ideal
works only in so far we forget all what
may go wrong
let’s accept things may go wrong at all level
let’s accept we need to integrate
systematically feedbacks from
higher-order courts, jurisprudence or
other normative sources
people or other actors should
be able to appeal courts or other actors should
be able to cassate

32. including computational actors
(possibility of continuous, automated testing/verification)
code-driven law
data-driven law
text-guided law
derived from legal norms or policies that
have been articulated in computer code
(or derived from other methods)
lower authority w.r.t. human authorities!!!
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”
the COMPUTATIONAL LEGALISM ideal
works only in so far we forget all what
may go wrong
let’s accept things may go wrong at all level
let’s accept we need to integrate
systematically feedbacks from
higher-order courts, jurisprudence or
other normative sources
people or other actors should
be able to appeal courts or other actors should
be able to cassate
code-guided law
monolithical systems
ecological system including
interfaces with humans

33. including computational actors
(possibility of continuous, automated testing/verification)
code-driven law
data-driven law
text-guided law
derived from legal norms or policies that
have been articulated in computer code
(or derived from other methods)
lower authority w.r.t. human authorities!!!
legal activity performed by humans by
means of sources of norms such as
statute and case law decision support systems
legal certainty ?
justice (proportionality, distribution) ?
instrumentality (contextual policy goals) ?
data-guided law
automatic decision-making
derived from statistical/inductive
methods
the COMPUTATIONAL JUSTICE ideal
assumes perfect “conceptualization”
the COMPUTATIONAL LEGALISM ideal
works only in so far we forget all what
may go wrong
let’s accept things may go wrong at all level
let’s accept we need to integrate
systematically feedbacks from
higher-order courts, jurisprudence or
other normative sources
people or other actors should
be able to appeal courts or other actors should
be able to cassate
code-guided law
monolithical systems
ecological system including
interfaces with humans
a serious technological gap exists today...

34. Key points
● There is continuity between institutional and computational activities.
● We don’t need institutions to become more mechanical,
institutional constructs need to be brought into the computational realm.
● As we have a plurality of normative sources, we need a plurality of
computational normative sources.
normware

35. Key points
● There is continuity between institutional and computational activities.
● We don’t need institutions to become more mechanical,
institutional constructs need to be brought into the computational realm.
● As we have a plurality of normative sources, we need a plurality of
computational normative sources.
● Yet, this won’t solve all issues. Legalist standpoints may still undermine
justice. State with right does not coincide with State of right.
normware
even in computational terms...

36. Further (related) reading
“Normware” is very much a concept in progress, parts of this story can be found in:
Sileno, G., Of duels, trials, and simplifying systems, European Journal of Risk Regulation (2020).
Sileno, G., Boer, A., Gordon, G., Rieder, B., Like Circles in the Water: Responsibility as a System-Level Function,
Proceedings of 3rd XAILA workshop on Explainable AI and Law, in conjunction with JURIX (2020)
Sileno, G., Boer, A. and van Engers, T., The Role of Normware in Trustworthy and Explainable AI, Proceedings of
1st XAILA workshop on Explainable AI and Law, in conjunction with JURIX (2018)
Boer A., Sileno, G., Institutional Metaphors for Designing Large-Scale Distributed AI versus AI Techniques for
Running Institutions, in: “Anchoring Institutions”, 2021 (forthcoming), Springer

37. code-driven law?
Workshop on AI and border control, Edinburgh Law School (virtual), 15 June 2021
Giovanni Sileno, University of Amsterdam. g.sileno@uva.nl