NEW YORK STATE ASSEMBLY MEMORANDUM IN SUPPORT OF LEGISLATION submitted in accordance with Assembly Rule III, Sec 1(f)
BILL NUMBER: A8110
SPONSOR: Vanel TITLE OF BILL:
An act to amend the criminal procedure law and the civil practice law
and rules, in relation to the admissibility of evidence created or proc-
essed by artificial intelligence
PURPOSE OR GENERAL IDEA OF BILL:
Relates to the admissibility of evidence created or processed by artifi-
cial intelligence.
SUMMARY OF SPECIFIC PROVISIONS:
60.80 (1),(2),(3),(4): Evidence 'created' versus 'processed' -- The
primary distinction between evidence that was created or processed is
the result of the system. If "new information" is generated from the AI,
then it was "created." If a "conclusion" was generated from the AI, then
it was "processed." "New information" is information that did not exist
prior to the AI's intervention, while a "conclusion" is an interpreta-
tion of the information that did exist.
60.80(5): "Reasonably deducible" -- There currently exist many forms of
evidence analysis which use AI systems, but only as a means to achieve
its ends more quickly. These would not fall under the standard set forth
in this bill because the accuracy and reliability of the conclusions
drawn, or information created by them, would not be substantially
compromised if AI was not used. Instead, the AI simply makes the work
more efficient.
60.80(6)(a): "The evidence is separate from, and not derived from, the
artificially created evidence." - The evidence must be supported by
something other than the artificially created evidence.
60.80(6)(b): The independent evidence must be admissible under the
existing rules of evidence.
60.80(6)(c): "The independent evidence bears a close and significant
relationship to the artificially created evidence in such a manner that
it reinforces or corroborates the information created from the arti-
ficially created evidence. -- Courts should only accept such evidence
which corroborates the evidence or reinforces information. This inquiry
is highly fact specific and can vary depending on other evidence
presented.
60.80(7)(a): "The specific use case of the AI system must also be vali-
dated through rigorous scientific or technical testing, demonstrating
that it consistently produces accurate and reliable results in varied
environments." -- The accuracy and reliability of the specific use case
of the AI that is used in the case at hand can be determined by looking
at prior testing of the AI system. This testing should be done in varied
environments to ensure the highest probability of an accurate result
regardless of the environment. The level of accuracy should be deter-
mined on a case-by-case basis. Courts should not get caught up on the
question of whether AI generated evidence should be admitted where the
evidence has little weight considering the weight of other evidence.
60.80(7)(b): "The AI has been subjected to testing or application in
environments that are similar or analogous to the specific context in
which it is being used in the proceeding and such testing or application
produced accurate and reliable results." - Courts must ask whether the
AI has been used in the way that it is being used in a similar or analo-
gous situation to the situation at hand. The "context" to which this
part refers is the domain or set of circumstances in which the AI system
is operating.
60.80(7)(c): "The AI has not been subjected to any variables that, based
on scientific or technical testing of the system, have a substantial
probability of causing a materially inaccurate or unreliable result. In
assessing the probability, the court shall consider the weight of the
artificially created evidence relative to other admitted evidence." --
This is similar to the previous element, but with a minor but important
difference. This element states that if we have knowledge of the fact
that the AI system produces inaccurate or unreliable results when
subjected to a certain variable, then the evidence should not be used.
So, even if the AI is generally accurate, and even if it has been tested
to be accurate in the given context, was a variable introduced that
would cause an inaccuracy?
60.80(8): Permits the courts to limit the disclosure of certain informa-
tion.
4551 - The same as the foregoing but as applied in the civil context.
JUSTIFICATION:
Artificial intelligence (AI) systems are sophisticated technologies that
can independently analyze data to generate novel conclusions and new
information. As these technologies continue to advance in accuracy and
capability, their use cases also expand. One crucial aspect that society
must examine is Al's ability to create and process evidence for use in
criminal and civil court proceedings.
A common challenge in many AI systems is known as the "black box" prob-
lem. This complexity arises from the nature of Al's operation, partic-
ularly within neural networks. These networks function as a series of
interconnected processing units guiding the AI's operations and, in
turn, undertake a process to generate an output or prediction based on
given data.
The crux of the black box problem is that no one, not even the AI's
programmers, can precisely understand how the AI reaches its conclusions
from the data. This ambiguity introduces an issue with evidence created
or processed entirely or partially by AI; we cannot discern how the AI
arrived at a specific conclusion. However, through rigorous testing, we
can Verify whether those conclusions are accurate and reliable.
This unique challenge posed by AI necessitates a reevaluation of how we
approach evidence, particularly when we can't be certain of the steps
involved in its creation or processing.
Therefore, this bill outlines the essential steps that must be undertak-
en to ensure that no individual is unjustly penalized in a criminal or
civil proceeding due to evidence created or processed by AI. It strives
to set a legal framework that both recognizes the potential of AI and
safeguards the principles of justice.
PRIOR LEGISLATIVE HISTORY:
New bill
FISCAL IMPLICATIONS:
TBD
EFFECTIVE DATE:
This act shall take effect on the first of January next succeeding the
date on which it shall have become law.
STATE OF NEW YORK
________________________________________________________________________
8110
2023-2024 Regular Sessions
IN ASSEMBLY
October 13, 2023
___________
Introduced by M. of A. VANEL -- read once and referred to the Committee
on Codes
AN ACT to amend the criminal procedure law and the civil practice law
and rules, in relation to the admissibility of evidence created or
processed by artificial intelligence
The People of the State of New York, represented in Senate and Assem-bly, do enact as follows:
1 Section 1. The criminal procedure law is amended by adding a new
2 section 60.80 to read as follows:
3 § 60.80 Rules of evidence; admissibility of evidence created or proc-
4 essed by artificial intelligence.
5 1. Evidence created, in whole or in part, by artificial intelligence
6 shall not be received into evidence in a criminal proceeding unless the
7 evidence is substantially supported by independent and admissible
8 evidence and the proponent of the evidence establishes the reliability
9 and accuracy of the specific use of the artificial intelligence in
10 creating the evidence.
11 2. Evidence processed, in whole or in part, by artificial intelligence
12 shall not be received into evidence in a criminal proceeding unless the
13 proponent of the evidence establishes the reliability and accuracy of
14 the specific use of the artificial intelligence in processing the
15 evidence.
16 3. Evidence is created, in whole or in part, by artificial intelli-
17 gence where the artificial intelligence produces new information from
18 existing information not present in or reasonably deducible from the
19 existing information.
20 4. Evidence is processed, in whole or in part, by artificial intelli-
21 gence where the artificial intelligence produces a conclusion based off
22 of its analysis, interpretation, or transformation of existing informa-
23 tion where such conclusion is not reasonably deducible from the existing
24 information.
EXPLANATION--Matter in italics (underscored) is new; matter in brackets
[] is old law to be omitted.
LBD13089-01-3
A. 8110 2
1 5. Evidence is not reasonably deducible from the existing information
2 where the reliability or accuracy of the information created or conclu-
3 sion drawn from the existing information would be substantially compro-
4 mised without the use of artificial intelligence as a result of the
5 complexity, uncertainty, or subtlety of the information.
6 6. Evidence is substantially supported by independent and admissible
7 evidence where:
8 (a) The independent evidence is separate from, and not derived from,
9 the artificial intelligence that generated the artificially created
10 evidence;
11 (b) The independent evidence is admissible under the existing rules of
12 evidence;
13 (c) The independent evidence bears a close and significant relation-
14 ship to the artificially created evidence in such a manner that it rein-
15 forces or corroborates the information created from the artificially
16 created evidence.
17 7. The reliability and accuracy of the specific use of the artificial
18 intelligence in creating or processing the evidence is sufficient where
19 the proponent of the evidence has a qualified expert testify and such
20 testimony is sufficient to prove that:
21 (a) The specific use of the artificial intelligence has been validated
22 through rigorous scientific or technical testing, demonstrating that it
23 consistently produces accurate and reliable results in varied environ-
24 ments;
25 (b) The artificial intelligence has been subjected to testing or
26 application in environments that are similar or analogous to the specif-
27 ic context in which it is being used in the proceeding and such testing
28 or application produced accurate and reliable results;
29 (c) The artificial intelligence has not been subjected to any vari-
30 ables that, based on scientific or technical testing of the system, have
31 a substantial probability of causing a materially inaccurate or unreli-
32 able result. In assessing the probability, the court shall consider the
33 weight of the artificially created evidence relative to other admitted
34 evidence.
35 8. Where expert testimony would include trade secrets, privileged
36 government information, or information about law enforcement techniques
37 that, if disclosed, would unduly compromise their ability to effectively
38 use their systems for their intended purpose, the court, in its
39 discretion, may impose appropriate measures to protect such information.
40 § 2. The civil practice law and rules is amended by adding a new
41 section 4551 to read as follows:
42 § 4551. Rules of evidence; admissibility of evidence created or proc-
43 essed by artificial intelligence. (a) Evidence created, in whole or in
44 part, by artificial intelligence may not be received into evidence in a
45 civil proceeding unless the evidence is substantially supported by inde-
46 pendent and admissible evidence and the proponent of the evidence estab-
47 lishes the reliability and accuracy of the specific use of the artifi-
48 cial intelligence in creating the evidence.
49 (b) Evidence processed, in whole or in part, by artificial intelli-
50 gence may not be received into evidence in a civil proceeding unless the
51 proponent of the evidence establishes the reliability and accuracy of
52 the specific use of the artificial intelligence in processing the
53 evidence.
54 (c) Evidence is created, in whole or in part, by artificial intelli-
55 gence where the artificial intelligence produces new information from
A. 8110 3
1 existing information not present in or reasonably deducible from the
2 existing information.
3 (d) Evidence is processed, in whole or in part, by artificial intelli-
4 gence where the artificial intelligence produces a conclusion based off
5 of its analysis, interpretation, or transformation of existing informa-
6 tion where such conclusion is not reasonably deducible from the existing
7 information.
8 (e) Evidence is not reasonably deducible from the existing information
9 where the reliability or accuracy of the information created or conclu-
10 sion drawn from the existing information would be substantially compro-
11 mised without the use of artificial intelligence as a result of the
12 complexity, uncertainty, or subtlety of the information.
13 (f) Evidence is substantially supported by independent and admissible
14 evidence where:
15 1. The independent evidence is separate from, and not derived from,
16 the artificial intelligence that generated the artificially created
17 evidence;
18 2. The independent evidence is admissible under the existing rules of
19 evidence;
20 3. The independent evidence bears a close and significant relationship
21 to the artificially created evidence in such a manner that it reinforces
22 or corroborates the information created from the artificially created
23 evidence.
24 (g) The reliability and accuracy of the specific use of the artificial
25 intelligence in creating or processing the evidence is sufficient where
26 the proponent of the evidence has a qualified expert testify and such
27 testimony is sufficient to prove that:
28 1. The specific use of the artificial intelligence has been validated
29 through rigorous scientific or technical testing, demonstrating that it
30 consistently produces accurate and reliable results in varied environ-
31 ments;
32 2. The artificial intelligence has been subjected to testing or appli-
33 cation in environments that are similar or analogous to the specific
34 context in which it is being used in the proceeding and such testing or
35 application produced accurate and reliable results;
36 3. The artificial intelligence has not been subjected to any variables
37 that, based on scientific or technical testing of the system, have a
38 substantial probability of causing a materially inaccurate or unreliable
39 result. In assessing the probability, the court shall consider the
40 weight of the artificially created evidence relative to other admitted
41 evidence.
42 (h) Where expert testimony would include trade secrets, privileged
43 government information, or information about law enforcement techniques
44 that, if disclosed, would unduly compromise their ability to effectively
45 use their systems for their intended purpose, the court, in its
46 discretion, may impose appropriate measures to protect such information.
47 § 3. This act shall take effect on the first of January next succeed-
48 ing the date on which it shall have become a law.
