I Robot Research Paper

SYRACUSE SCIENCE & TECHNOLOGY
LAW REPORTER
VOLUME 22

SPRING 2010

ARTICLE 1, PAGE 1

“I, Robot – I, Criminal”—When Science Fiction Becomes Reality:
Legal Liability of AI Robots committing Criminal Offenses
Gabriel Hallevy*
I. INTRODUCTION
Can society impose criminal liability upon robots? The technological world has changed rapidly. Simple human activities are being replaced by robots. As long as humanity used robots as mere tools, there was no real difference between robots and screwdrivers, cars or telephones.
When robots became sophisticated, we used to say that robots “think” for us. The problem began when robots evolved from “thinking” machines into thinking machines (without quotation marks)—or Artificial Intelligence
Robots (AI Robots). Could they become dangerous?
Unfortunately, they already are. In 1950, Isaac Asimov set down three fundamental laws of robotics in his science fiction masterpiece “I, Robot”: (1) a robot may not injure a human being or, through inaction, allow a human being to come to harm; (2) a robot must obey the orders given to it by human beings, except where such orders would conflict with the First Law; and (3) a robot must protect its own existence, as long as such protection does not conflict with the First or Second Laws.1 These three fundamental laws are obviously contradictory.2 What if a man orders a robot to hurt another person for the own good of the other person? What if the
*

Associate Professor, Faculty of Law, Ono Academic College.

1

ISSAC ASIMOV, I, ROBOT (1950).

2

Isaac Asimov wrote in his introduction to THE REST OF ROBOTS (1964) that “[t]here was just enough ambiguity in the Three Laws to provide the conflicts and uncertainties required for new stories, and, to my great relief, it seemed always to be possible to think up a new angle out of the
61 words of the Three Laws.”

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

2

robot is in police service and the commander of the mission orders it to arrest a suspect and the suspect resists arrest? Or what if the robot is in medical service and is ordered to perform a surgical procedure on a patient, the patient objects, but the medical doctor insists that the procedure is for the patient’s own good, and repeats the order to the robot?
The main question in that context is which kind of laws or ethics are correct and who is to decide. In order to cope with these same problems as they relate to humans, society devised criminal law. Criminal law embodies the most powerful legal social control in modern civilization. People’s fear of AI robots, in most cases, is based on the fact that AI robots are not considered to be subject to the law, specifically to criminal law. In the past, people were similarly fearful of corporations and their power to commit a spectrum of crimes, but since corporations are legal entities subject to criminal and corporate law, that kind of fear has been reduced significantly.3
The apprehension that AI robots evoke may have arisen due to Hollywood’s depiction of
AI robots in numerous films, such as “2001: A Space Odyssey,”4 and the modern trilogy “The
Matrix,”5 in which AI robots are not subject to the law. However, it should be noted that
Hollywood did treat AI robots in an empathic way by depicting them as human, as almost

3

See generally John C. Coffee, Jr., “No Soul to Damn: No Body to Kick”: An Unscandalised
Inquiry Into the Problem of Corporate Punishment, 79 MICH. L. REV. 386 (1981); STEVEN BOX,
POWER, CRIME AND MYSTIFICATION 16-79 (1983); Brent Fisse & John Braithwaite, The
Allocation of Responsibility for Corporate Crime: Individualism, Collectivism and
Accountability, 11 SYDNEY L. REV. 468 (1988).
4
5

STANLEY KUBRICK, 2001: A SPACE ODYSSEY (1968).

JOEL SILVER, THE MATRIX (1999); JOEL SILVER, LAURENCE WACHOWSKI AND ANDREW PAUL
WACHOWSKI, THE MATRIX RELOADED (2003); JOEL SILVER, LAURENCE WACHOWSKI AND
ANDREW PAUL WACHOWSKI, THE MATRIX REVOLUTIONS (2003).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

3

human, or as wishing to be human.6 This kind of treatment included, of course, clear subordination to human legal social control and to criminal law.
The modern question relating to AI robots becomes: Does the growing intelligence of AI robots subject them to legal social control, just as any other legal entity?7 This article attempts to work out a legal solution to the problem of the criminal liability of AI robots. At the outset, a definition of an AI robot will be presented. Based on that definition, this article will then propose and introduce three models of AI robot criminal liability: (1) the perpetration-byanother liability model, (2) the natural-probable-consequence liability model, and (3) the direct liability model.
These three models might be applied separately, but in many situations, a coordinated combination of them (all or some of them) is required in order to complete the legal structure of criminal liability. Once we examine the possibility of legally imposing criminal liability on AI robots, then the question of punishment must be addressed. How can an AI robot serve a sentence of imprisonment? How can capital punishment be imposed on an AI robot? How can probation, a pecuniary fine, or the like be imposed on an AI robot? Consequently, it is necessary to formulate viable forms of punishment in order to impose criminal liability practically on AI robots. 6

See, e.g., STEVEN SPIELBERG, STANLEY KUBRICK, JAN HARLAN, KATHLEEN KENNEDY, WALTER
F. PARKES AND BONNIE CURTIS, A.I. ARTIFICIAL INTELLIGENCE (2001).
7

See in general, but not in relation to criminal law, e.g., Thorne L. McCarty, Reflections on
Taxman: An Experiment in Artificial Intelligence and Legal Reasoning, 90 HARV. L. REV. 837
(1977); Donald E. Elliott, Holmes and Evolution: Legal Process as Artificial Intelligence, 13 J.
LEGAL STUD. 113 (1984); Thomas E. Headrick & Bruce G. Buchanan, Some Speculation about
Artificial Intelligence and Legal Reasoning, 23 STAN. L. REV. 40 (1971); Antonio A. Martino,
Artificial Intelligence and Law, 2 INT’L J.L. & INFO. TECH. 154 (1994); Edwina L. Rissland,
Artificial Intelligence and Law: Stepping Stones to a Model of Legal Reasoning, 99 YALE L.J.
1957 (1990).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

4

II. WHAT IS AN AI ROBOT?
For some years, there has been significant controversy about the very essence of AI robots.8 Futurologists have proclaimed the birth of a new species, machina sapiens, which will share the human place as intelligent creatures on Earth.9 Critics have argued that a “thinking machine” is an oxymoron.10 Machines, including robots, with their foundations of cold logic, can never be insightful or creative as humans are.11 This controversy raises the basic questions of the essence of humanity (i.e., do human beings function as thinking machines?) and of AI
(i.e., can there be thinking machines?).12
There are five attributes that one would expect an intelligent entity to have:13
(i) communication (One can communicate with an intelligent entity. The easier it is to communicate with an entity, the more intelligent the entity seems. One can communicate with a dog, but not about Einstein’s theory of relativity. One can communicate with a little child about
Einstein’s theory, but it requires a discussion in terms that a child can comprehend.); (ii) mental knowledge (An intelligent entity is expected to have some knowledge about itself.); (iii)

8

Terry Winograd, Thinking Machines: Can There Be? Are We?, in THE FOUNDATIONS OF
ARTIFICIAL INTELLIGENCE 167 (Derek Partridge & Yorick Wilks eds., 2006).
9

Id.

10

Id.

11

Id.

12

For the formal foundations of AI, see generally Teodor C. Przymusinski, Non-Monotonic
Reasoning versus Logic Programming: A New Perspective, in THE FOUNDATIONS OF ARTIFICIAL
INTELLIGENCE 49 (Derek Partridge & Yorick Wilks eds., 2006); Richard W. Weyhrauch,
Prolegomena to a Theory of Mechanized formal Reasoning, in THE FOUNDATIONS OF ARTIFICIAL
INTELLIGENCE 72 (Derek Partridge & Yorick Wilks eds., 2006).
13

Roger C. Schank, What is AI, Anyway?, in THE FOUNDATIONS OF ARTIFICIAL INTELLIGENCE 3
(Derek Partridge & Yorick Wilks eds., 2006).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

5

external knowledge (An intelligent entity is expected to know about the outside world, to learn about it, and utilize that information.); (iv) goal-driven behavior (An intelligent entity is expected to take action in order to achieve its goals.); and (v) creativity (An intelligent entity is expected to have some degree of creativity. In this context, creativity means the ability to take alternate action when the initial action fails. A fly that tries to exit a room and bumps into a window pane, tries to do that over and over again. When an AI robot bumps into a window, it tries to exit using the door).
Most AI robots possess these five attributes by definition.14 Some 21st Century types of
AI robots possess even more attributes that enable them to act in far more sophisticated ways. In
November 2009, during the Supercomputing Conference in Portland Oregon (“SC 09”), IBM scientists and others announced that they succeeded in creating a new algorithm named “Blue
Matter,” which possesses the thinking capabilities of a cat.15 This algorithm collects information from many units with parallel and distributed connections.16 The information is integrated and creates a full image of sensory information, perception, dynamic action and reaction, and cognition.17 This platform simulates brain capabilities, and eventually, it is supposed to simulate

14

Schank, supra note 13, at 4-6.

15

Chris Capps, “Thinking” Supercomputer Now Conscious as a Cat, UNEXPLAINABLE.NET,
Nov. 19, 2009, http://www.unexplainable.net/artman/publish/article_14423.shtml; see also Super
Computing, http://sc09.supercomputing.org.
16

Id.

17

Id.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

6

real thought processes.18 The final application of this algorithm contains not only analog and digital circuits, metal or plastics, but also protein-based biologic surfaces.19
An AI robot has a wide variety of applications.20 A robot can be designed to imitate the physical capabilities of a human being, and these capabilities can be improved.21 For instances, a robot is capable of being physically faster and stronger than a human being.22 The AI software installed in it also enables the robot to calculate many complicated calculations faster and simultaneously, or to “think” faster.23 An AI robot is capable of learning and of gaining experience, and experience is a useful way of learning.24 All these attributes create the essence of an AI robot. AI robots and AI software are used in a wide range of applications in industry, military services, medical services, science, and even in games.25

18

Capps, supra note 15.

19

Id.

20

See, e.g., Yorick Wilks, One Small Head: Models and Theories, in THE FOUNDATIONS OF
ARTIFICIAL INTELLIGENCE 121 (Derek Partridge & Yorick Wilks eds., 2006); Alan Bundy &
Stellan Ohlsson, The Nature of AI Principles, in THE FOUNDATIONS OF ARTIFICIAL INTELLIGENCE
135 (Derek Partridge & Yorick Wilks eds., 2006); Thomas W. Simon, Artificial Methodology
Meets Philosophy, in THE FOUNDATIONS OF ARTIFICIAL INTELLIGENCE 155 (Derek Partridge &
Yorick Wilks eds., 2006).
21

Id.

22

Id.

23

Id.

24

Id.

25

See, e.g., William B. Schwartz, Ramesh S. Patil & Peter Szolovits, Artificial Intelligence in
Medicine Where Do We Stand, 27 JURIMETRICS J. 362 (1987); Richard E. Susskind, Artificial
Intelligence, Expert Systems and the Law, 5 DENNING L.J. 105 (1990).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

7

III. MODELS OF THE CRIMINAL LIABILITY OF AI ROBOTS
The fundamental question of criminal law is the question of criminal liability, i.e., whether the specific entity (human or corporation) bears criminal liability for a specific offense committed at a specific point in time and space.26 In order to impose criminal liability upon a person, two main elements must exist.27 The first is the factual element, i.e., criminal conduct
(actus reus), while the other is the mental element, i.e., knowledge or general intent in relation to the conduct element (mens rea).28 If one of them is missing, no criminal liability can be imposed.29 The actus reus requirement is expressed mainly by acts or omissions.30 Sometimes, other factual elements are required in addition to conduct, such as the specific results of that conduct and the specific circumstances underlying the conduct.31 The mens rea requirement has various levels of mental elements.32 The highest level is expressed by knowledge, while

26

See generally JEROME HALL, GENERAL PRINCIPLES OF CRIMINAL LAW 70-211 (2d ed. 2005)
(1960).
27

Id.

28

Id.

29

Id.

30

See Walter Harrison Hitchler, The Physical Element of Crime, 39 DICK. L. REV. 95 (1934);
MICHAEL MOORE, ACT AND CRIME: THE PHILOSOPHY OF ACTION AND ITS IMPLICATIONS FOR
CRIMINAL LAW 5 (1993) (Tony Honore & Joseph Raz, eds., Oxford University Press 1993).
31

See JOHN WILLIAM SALMOND, ON JURISPRUDENCE 505 (Glanville Williams ed., 11th ed.
1957); GLANVILLE WILLIAMS, CRIMINAL LAW: THE GENERAL PART 18 (2d ed., Steven & Sons
Ltd. 1961); OLIVER W. HOLMES, THE COMMON LAW 54 (Mark DeWolf Howe ed., Harvard
University Press 1923) (1881); Walter Wheeler Cook, Act, Intention, and Motive in Criminal
Law, 26 YALE L.J. 645 (1917).
32

HALL, supra note 26, at 105-45, 325-59.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

8

sometimes it is accompanied by a requirement of intent or specific intention.33 Lower levels are expressed by negligence (a reasonable person should have known),34 or by strict liability offenses.35 No other criteria or capabilities are required in order to impose criminal liability, not from humans, nor from any other kind of entity, including corporations and AI robots.36 An entity might possess further capabilities; however, in order to impose criminal liability, the existence of actus reus and mens rea in the specific offense is quite enough.37 As far as known to science, a spider is capable of acting, but it is incapable of formulating the mens rea requirement; therefore, a spider bite bears no criminal liability. A parrot is capable of repeating words it hears, but it is incapable of formulating the mens rea requirement for libel. In order to impose criminal liability on any kind of entity, it must be proven that the above two elements

33

See generally J.
Ll. J. Edwards, The Criminal Degrees of Knowledge, 17 MOD. L. REV. 294,
295 (1954); Rollin M. Perkins, “Knowledge” as a Mens rea Requirement, 29 HASTINGS L.J. 953
(1978); and see, e.g., United States v. Youts, 229 F.3d 1312, 1316 (10th Cir. 2000); United
States v. Spinney, 65 F.3d 231, 235 (1st Cir. 1995); People v. Steinberg, 595 N.E.2d 845, 847
(N.Y. 1992); State v. Sargent, 594 A.2d 401, 403 (Vt. 1991); State v. Wyatt, 482 S.E.2d 147, 150
(W. Va. 1996).
34

See, e.g., Jerome Hall, Negligent Behaviour Should Be Excluded from Penal Liability, 63
COLUM. L. REV. 632, 632 (1963); Robert P. Fine & Gary M. Cohen, Is Criminal Negligence a
Defensible Basis for Criminal Liability?, 16 BUFF. L. REV. 749, 780 (1966).
35

See, e.g., Jeremy Horder, Strict Liability, Statutory Construction and the Spirit of Liberty, 118
L. Q. REV. 458, 458 (2002); Francis Bowes Sayre, Public Welfare Offenses, 33 COLUM. L. REV.
55, 55 (1933); Stuart P. Green, Six Senses of Strict Liability: A Plea for Formalism, in
APPRAISING STRICT LIABILITY 1 (A. P. Simester ed., 2005); A. P. Simester, Is Strict Liability
Always Wrong?, in APPRAISING STRICT LIABILITY 21 (A. P. Simester ed., 2005).
36

HALL, supra note 26.

37

Id. at 185-86.

Vol.
22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

9

existed.38 Thus, when it has been proven that a person committed the criminal act knowingly or with criminal intent, that person is held criminally liable for that offense.39 The relevant question concerning the criminal liability of AI robots is: How can these entities fulfill the two requirements of criminal liability? This article proposes the imposition of criminal liability on
AI robots using three possible models of liability: (A) the Perpetration-by-Another liability model; (B) the Natural-Probable-Consequence liability model; and (C) the Direct liability model.
The following is an explanation of these possible models:
A. The Perpetration-by-Another Liability Model: AI Robots as Innocent Agents
This first model does not consider the AI robot as possessing any human attributes. The
AI robot is considered an innocent agent. Accordingly, due to that legal viewpoint, a machine is a machine, and is never human. However, one cannot ignore an AI robot’s capabilities, as previously mentioned.40 Pursuant to this model, these capabilities are insufficient to deem the AI robot a perpetrator of an offense. These capabilities resemble the parallel capabilities of a mentally limited person, such as a child, or of a person who is mentally incompetent or who lacks a criminal state of mind.41 Legally, when an offense is committed by an innocent agent, like when a person causes a child,42 a person who is mentally incompetent43 or who lacks a

38

HALL, supra note 26.

39

Id. at 105-45, 183.

40

See discussion supra Part II.

41

HALL, supra note 26, at 232.

42

See, e.g., Maxey v. United States, 30 App. D.C. 63 (1907); Commonwealth v. Hill, 11 Mass.
136 (1814); R.
v. Michael, 169 Eng. Rep. 48 (1840).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

10

criminal state of mind to commit an offense,44 that person is criminally liable as a perpetrator-byanother.45 In such cases, the intermediary is regarded as a mere instrument, albeit a sophisticated instrument, while the party orchestrating the offense (the perpetrator-by-another) is the real perpetrator as a principal in the first degree and is held accountable for the conduct of the innocent agent.46 The perpetrator’s liability is determined on the basis of that conduct47 and his own mental state.48
The derivative question relative to AI Robots is: Who is the perpetrator-by-another?
There are two candidates: the first is the programmer of the AI software installed in the specific robot and the second is the user. A programmer of AI software might design a program in order to commit offenses via the AI robot. For example, a programmer designs software for an operating robot. The robot is intended to be placed in a factory, and its software is designed to torch the factory at night when no one is there. The robot committed the arson, but the programmer is deemed the perpetrator. The second person who might be considered
the perpetrator-by-another is the user of the AI robot. The user did not program the software, but he

43

See, e.g., Johnson v. State, 38 So. 182 (Ala. 1904); People v. Monks, 24 P.2d 508 (Cal. Dist.
Ct. App. 1933).

44

See, e.g., United States v. Bryan, 483 F.2d 88 (3rd Cir. 1973); Boushea v. United States, 173
F.2d 131 (8th Cir. 1949); People v. Mutchler, 140 N.E. 820 (Ill. 1923); State v. Runkles, 605
A.2d 111 (Md. 1992); Parnell v. State, 912 S.W.2d 422 (Ark. 1996); State v. Thomas, 619
S.W.2d 513 (Tenn. 1981).

45

See generally Morrisey v. State, 620 A.2d 207 (Del. 1993); State v. Fuller, 552 S.E.2d 282
(S.C. 2001); Gallimore v. Commonwealth, 436 S.E.2d 421 (Va. 1993).
46

Id.

47

See generally Dusenbery v. Commonwealth, 263 S.E.2d 392, 392 (Va. 1980).

48

See generally United States v. Tobon-Builes, 706 F.2d 1092 (11th Cir. 1983); United States v.
Ruffin, 613 F.2d 408 (2nd Cir. 1979).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

11

uses the AI robot, including its software, for his own benefit. For example, a user purchases a servant-robot, which is designed to execute any order given by its master. The specific user is identified by the robot as that master, and the master orders the robot to assault any invader of the house. The robot executes the order exactly as ordered. This is not different than a person who orders his dog to attack any trespasser. The robot committed the assault, but the user is deemed the perpetrator.
In both scenarios, the actual offense was committed by the AI robot. The programmer or the user did not perform any action conforming to the definition of a specific offense; therefore, they do not meet the actus reus requirement of the specific offense. The perpetration-by-another liability model considers the action committed by the AI robot as if it had been the programmer’s or the user’s action. The legal basis for that is the instrumental usage of the AI robot as an innocent agent. No mental attribute required for the imposition of criminal liability is attributed to the AI robot.49 When programmers or users use an AI robot instrumentally, the commission of an offense by the AI robot is attributed to them. The mental element required in the specific offense already exists in their minds.50 The programmer had criminal intent when he ordered the commission of the arson, and the user had criminal intent when he ordered the commission of the assault, even though these offenses were actually committed through an AI robot.
This liability model does not attribute any mental capability, or any human mental capability, to the AI robot. According to this model, there is no legal difference between an AI robot and a screwdriver or an animal. When a burglar uses a screwdriver in order to open up a
49

The AI robot is used as an instrument and not as a participant, although it uses its features of processing information. See George R. Cross & Cary G. Debessonet, An Artificial Intelligence
Application in the Law: CCLIPS, A Computer Program that Processes Legal Information, 1
HIGH TECH. L.J. 329, 362 (1986).
50

HALL, supra note 26.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

12

window, he uses the screwdriver instrumentally, and the screwdriver is not criminally liable.
The screwdriver’s “action” is, in fact, the burglar’s. This is the same legal situation when using an animal instrumentally. An assault committed by a dog by order of its master is, in fact, an assault committed by the master.
This kind of legal model might be suitable for two types of scenarios. The first scenario is using an AI robot to commit an offense without using its advanced capabilities. The second scenario is using a very old version of an AI robot, which lacks the modern advanced capabilities of the modern AI robots. In both scenarios, the use of the AI robot is instrumental; it is usage of an AI robot, given its ability to execute an order, to commit an offense. A screwdriver cannot execute such an order; a dog can. A dog cannot execute complicated orders; an AI robot can.51
The perpetration-by-another liability model is not suitable when an AI robot decides to commit an offense based on its own accumulated experience or knowledge. This model is not suitable when the software of the AI robot was not designed to commit the specific offense, but was committed by the AI robot nonetheless. This model is also not suitable when the specific AI robot functions not as an innocent agent, but as a semi-innocent agent.52 However, the perpetration-by-another liability model might be suitable when a programmer or user makes instrumental usage of an AI robot, but without using the AI robot’s advanced capabilities. The legal result of applying this model is that the programmer and the user are both criminally liable for the specific offense committed, while the AI robot has no criminal liability whatsoever.

51

Compare Andrew J. Wu, From Video Games to Artificial Intelligence: Assigning Copyright
Ownership to Works Generated by Increasingly Sophisticated Computer Programs, 25 AIPLA
Q.J. 131 (1997), with Timothy L. Butler, Can a Computer be an Author: Copyright Aspects of
Artificial Intelligence, 4 HASTINGS COMM. & ENT. L.J. 707 (1982).
52

See generally NICOLA LACEY & CELIA WELLS, RECONSTRUCTING CRIMINAL LAW: CRITICAL
PERSPECTIVES ON CRIME AND THE CRIMINAL PROCESS 53 (2d ed. 1998).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

13

B. The Natural-Probable-Consequence Liability Model: Foreseeable Offenses Committed by AI
Robots
The second model of criminal liability assumes deep involvement of the programmers or users in the AI robot’s daily activities, but without any intention of committing any offense via the AI robot. For instance, one scenario would be when an AI robot commits an offense during the execution of its daily tasks. The programmers or users had no knowledge of the offense until it had already been committed. They did not plan to commit any offense, and they did not participate in any part of the commission of that specific offense.
An example of such a scenario is when an AI robot or software is designed to function as an automatic pilot. As part of the mission of flying the plane, the AI robot is programmed to protect the mission itself. During the flight, the human pilot activates the automatic pilot (which is the AI robot), and the program is initiated. At some point after activation of the automatic pilot, the human pilot sees an approaching storm and tries to abort the mission and return to base.
The AI robot deems the human pilot’s action as a threat to the mission and takes action in order to eliminate that threat; it may attempt to cut off the air supply to the pilot or activate the ejection seat. Whatever defense tactic is taken, the human pilot is killed as a result of the AI robot’s actions. Obviously, the programmer had not intended to kill anyone, especially not the human pilot, but nonetheless, the human pilot was killed by the AI robot’s programmed actions.
In this example, the first model is not legally suitable. The first model assumes mens rea, the criminal intent of the programmers or users to commit an offense via the instrumental use of some of the AI robot’s capabilities.53 This is not the legal situation in the case of the automatic pilot. In this case, the programmers or users had no knowledge of the committed offense; they had not planned it, and had not intended to commit the offense using the AI robot. For such
53

See discussion supra Part III.A.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

14

circumstances, the natural-probable-consequence liability model may create a more suitable legal response. This model is based upon the ability of the programmers or users to foresee the potential commission of offenses.
According to the second model, a person might be held accountable for an offense, if that offense is a natural and probable consequence of that person’s conduct. Originally, the naturalprobable-consequence liability was used to impose criminal liability upon accomplices, when one committed an offense, which had not been planned by all of them and which was not part of a conspiracy. 54 The established rule prescribed by courts and commentators is that accomplice liability extends to acts of a perpetrator that were a “natural and probable consequence”55 of a criminal scheme that the accomplice encouraged or aided.56 The natural-probable-consequence liability has been widely accepted in accomplice liability statutes and recodifications.57
Natural-probable-consequence liability seems to be legally suitable for situations where an AI robot committed an offense, but the programmer or user had no knowledge of it, had not intended it and had not participated in it. The natural-probable-consequence liability model only requires the programmer or user to be in a mental state of negligence, not more. Programmers or users are not required to know about any forthcoming commission of an offense as a result of

54

See generally United States v. Powell, 929 F.2d 724 (D.C. Cir. 1991).

55

Id.

56

See generally WILLIAM M. CLARK & WILLIAM L. MARSHALL, LAW OF CRIMES 529 (7th ed.
1967); Francis Bowes Sayre, Criminal Responsibility for the Acts of Another, 43 HARV. L. REV.
689 (1930); and see, e.g., People v. Prettyman, 926 P.2d 1013 (Cal. 1996); Chance v. State, 685
A.2d 351 (Del. 1996); Ingram v. United States, 592 A.2d 992 (D.C. 1991); Richardson v. State,
697 N.E.2d 462 (Ind. 1998); Mitchell v. State, 971 P.2d 813 (Nev. 1998); State v. Carrasco, 928
P.2d 939 (N.M. 1996); State v. Jackson, 976 P.2d 1229 (Wash. 1999).
57

See, e.g., United States v. Andrews, 75 F.3d 552, 553-57 (9th Cir. 1996); State v. Kaiser, 918
P.2d 629, 632-39 (Kan. 1996).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

15

their activity, but are required to know that such an offense is a natural, probable consequence of their actions.
A negligent person, in a criminal context, is a person who has no knowledge of the offense; rather, a reasonable person should have known about it since the specific offense is a natural probable consequence of that person’s conduct.58 Thus, the programmers or users of an
AI robot, who should have known about the probability of the forthcoming commission of the specific offense, are criminally liable for the specific offense, even though they did not actually know about it. This is the fundamental legal basis for criminal liability in negligence cases.59
Negligence is, in fact, an omission of awareness or knowledge.60 The negligent person omitted knowledge, not acts.61
The natural-probable-consequence liability model would permit liability to be predicated upon negligence, even when the specific offense requires a different state of mind.62 This is not valid in relation to the person who personally committed the offense, but rather, is considered valid in relation to the person who was not the actual perpetrator of the offense, but was one of its intellectual perpetrators.63 Reasonable programmers or users should have foreseen the

58

See generally Robert P. Fine & Gary M. Cohen, Is Criminal Negligence a Defensible Basis for
Criminal Liability?, 16 BUFF. L. REV. 749, 749-52 (1966); Herbert L.A. Hart, Negligence, Mens rea and Criminal Responsibility, in OXFORD ESSAYS IN JURISPRUDENCE 29 (1961); Donald
Stuart, Mens rea, in NEGLIGENCE AND ATTEMPTS, 1968 CRIM. L. REV. 647 (1968).
59

HALL, supra note 26, at 114-40.

60

Id.

61

Id.

62

THE AMERICAN LAW INSTITUTE, MODEL PENAL CODE: OFFICIAL DRAFT AND EXPLANATORY
NOTES § 2.05 (1962, 1985) [hereinafter Model Penal Code]; and see, e.g., State v. Linscott, 520
A.2d 1067, 1069 (Me. 1987); People v. Luparello, 231 Cal. Rptr. 832 (Cal. Ct. App. 1987).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

16

offense, and prevented it from being committed by the AI robot. However, the legal results of applying the natural-probable-consequence liability model to the programmer or user differ in two different types of factual cases. The first type of case is when the programmers or users were negligent while programming or using the AI robot but had no criminal intent to commit any offense. The second type of case is when the programmers or users programmed or used the
AI robot knowingly and willfully in order to commit one offense via the AI robot, but the AI robot deviated from the plan and committed some other offense, in addition to or instead of the planned offense.
The first type of case is a pure case of negligence. The programmers or users acted or omitted negligently; therefore, there is no reason why they should not be held accountable for an offense of negligence, if there is such an offense in the specific legal system. Thus, as in the example above, where a programmer of an automatic pilot negligently programmed it to defend its mission with no restrictions on the taking of human life, the programmer is negligent and liable for the homicide of the human pilot. Consequently, if there is a specific offense of negligent homicide in that legal system, this is the most severe offense, for which the programmer might be held accountable because manslaughter or murder requires knowledge or intent. The second type of case resembles the basic idea of the natural-probable-consequence liability in accomplice liability cases. The dangerousness of the very association or conspiracy whose aim is to commit an offense is the legal reason for more severe accountability to be imposed upon the cohorts. For example, a programmer programs an AI robot to commit a violent robbery of a bank, but the programmer did not program the AI robot to kill anyone.

63

See sources cited supra note 62.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

17

During the execution of the violent robbery, the AI robot kills one of the people present at the bank who resisted the robbery. In such cases, the criminal negligence liability alone is insufficient. The danger posed by such a situation far exceeds negligence.
As a result, according to the natural-probable-consequence liability model, when the programmers or users programmed or used the AI robot knowingly and willfully in order to commit one offense via the AI robot, but the AI robot deviated from the plan and committed another offense, in addition to or instead of the planned offense, the programmers or users shall be held accountable for the offense itself as if it had been committed knowingly and willfully. In the above example of the robbery, the programmer shall be held criminally accountable for the robbery (if committed), as well as for the killing as an offense of manslaughter or murder, which requires knowledge and intent.64
The question still remains: What is the criminal liability of the AI robot itself when the natural-probable-consequence liability model is applied? In fact, there are two possible outcomes. If the AI robot acted as an innocent agent, without knowing anything about the criminal prohibition, it is not held criminally accountable for the offense it committed. Under such circumstances, the actions of the AI robot were not different from the actions of the AI robot under the first model (the perpetration-by-another liability model65). However, if the AI robot did not act merely as an innocent agent, then the AI robot itself shall be held criminally liable for the specific offense directly, in addition to the criminal liability of the programmer or

64

See, e.g., United States v. Greer, 467 F.2d 1064 (7th Cir. 1972); People v. Cooper, 743 N.E.2d
32 (Ill. 2000); People v. Michalow, 128 N.E. 228 (N.Y. 1920); People v. Little, 107 P.2d 634
(Cal. Dist. Ct. App. 1941); People v. Cabaltero, 87 P.2d 364 (Cal. Dist. Ct. App. 1939); People v.
Weiss, 9 N.Y.S.2d 1 (N.Y. App. 1939); R v. Cunningham, 3 W.L.R. 76 (1957); R v. Faulkner,
13 Cox C.C. 550 (1876).

65

See supra Part III.A.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

18

user pursuant to the natural-probable-consequence liability model. The direct liability model of
AI robots is the third model, as described hereunder.
C. The Direct Liability Model: AI Robots as Direct Subjects of Criminal Liability
The third model does not assume any dependence of the AI robot on a specific programmer or user. The third model focuses on the AI robot itself.66 As discussed above, criminal liability for a specific offense is mainly comprised of the factual element (actus reus) and the mental element (mens rea) of that offense.67 Any person attributed with both elements of the specific offense is held criminally accountable for that specific offense.68 No other criteria are required in order to impose criminal liability.69 A person might possess further capabilities, but, in order to impose criminal liability, the existence of the factual element and the mental element required to impose liability for the specific offense is quite enough.70 In order to impose criminal liability on any kind of entity, the existence of these elements in the specific entity must be proven.71 When it has been proven that a person committed the offense in question with

66

See generally Steven J. Frank, Tort Adjudication and the Emergence of Artificial Intelligence
Software, 21 SUFFOLK U. L. REV. 623 (1987); Sam N. Lehman-Wilzig, Frankenstein Unbound:
Towards a Legal Definition of Artificial Intelligence, 13 FUTURES 442 (1981); Maruerite E.
Gerstner, Liability Issues with Artificial Intelligence Software, 33 SANTA CLARA L. REV. 239
(1993); Richard E. Susskind, Expert Systems in Law: A Jurisprudential Approach to Artificial
Intelligence and Legal Reasoning, 49 MOD. L. REV. 168 (1986).
67

HALL, supra note 26.

68

Id.

69

Id.

70

Id. at 185-86.

71

Id. at 183.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

19

knowledge or intent, that person is held criminally liable for that offense.72 The relevant questions regarding the criminal liability of AI robots are: How can these robots fulfill the requirements of criminal liability? Do AI robots differ from humans in this context?
An AI algorithm might have many features and qualifications far exceeding those of an average human, but such features or qualifications are not required in order to impose criminal liability.73 When a human or corporation fulfills the requirements of both the factual element and the mental element, criminal liability is imposed.74 If an AI robot is capable of fulfilling the requirements of both the factual element and the mental element, and, in fact, it actually fulfills them, there is presumptively nothing to prevent criminal liability from being imposed on that AI robot. Generally, the fulfillment of the factual element requirement of an offense is easily attributed to AI robots. As long as an AI robot controls a mechanical or other mechanism to move its moving parts, any act might be considered as performed by the AI robot. Thus, when an AI robot activates its electric or hydraulic arm and moves it, this might constitute an act. For example, in the specific offense of assault, such an electric or hydraulic movement of an AI robot that hits a person standing nearby is considered as fulfilling the actus reus requirement of the offense of assault. When an offense might be committed due to an omission, it is even simpler.
Under this scenario, the AI robot is not required to act at all; its very inaction is the legal basis for criminal liability, as long as there was a duty to act. If a duty to act is imposed upon the AI

72

HALL, supra note 26, at 105-45, 183.

73

Id. at 70-71; see supra Part II.

74

Id.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

20

robot and it fails to act, the actus reus requirement of the specific offense is fulfilled by way of an omission.
In most cases, the attribution of the mental element of offenses to AI robots is the real legal challenge. The attribution of the mental element differs from one AI technology to other.
Most cognitive capabilities developed in modern AI technology are immaterial to the question of the imposition of criminal liability.75 Creativity is a human feature that some animals have, but creativity is a not a requirement for imposing criminal liability.76 Even the most uncreative persons are held criminally liable. The sole mental-state requirements to impose criminal liability are knowledge, intent, negligence, or the mens rea required in the specific offense and under the general theory of criminal law.77
Knowledge is defined as sensory reception of factual data and the understanding of that data.78 Most AI systems are well equipped for such reception because they possess sensory receptors for sights, voices, physical contact, touch, and the like.79 These receptors transfer the

75

HALL, supra note 26.

76

Id.

77

Id.

78

See generally WILLIAM JAMES, THE PRINCIPLES OF PSYCHOLOGY (1890); HERMANN VON
HELMHOLTZ, THE FACTS OF PERCEPTION (1878). In this context, knowledge and awareness are identical. See, e.g., United States v. Youts, 229 F.3d 1312 (10th Cir. 2000); United States v.
Wert-Ruiz, 228 F.3d 250 (3th Cir. 2000); United States v. Ladish Malting Co., 135 F.3d 484 (7th
Cir. 1998); United States v. Spinney, 65 F.3d 231 (1st Cir. 1995); United States v. Jewell, 532
F.2d 697 (9th Cir. 1976); State v. Sargent, 594 A.2d 401 (Vt. 1991); State v. Wyatt, 482 S.E.2d
147 (W. Va. 1996). See also Model Penal Code, supra note 62, at § 2.02(2)(b), which provides that, “A person acts knowingly with a respect to a material element of an offense when: (i) [if] he is aware that his conduct is of that nature or that such circumstances exist; and (ii) [if] he is aware that it is practically certain that his conduct will cause such a result” (emphasis added).
79

See generally Margaret A. Boden, Has AI Helped Psychology?, in THE FOUNDATIONS OF
ARTIFICIAL INTELLIGENCE 108 (Derek Partridge & Yorick Wilks eds., 2006); David Marr, AI: A

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

21

factual data received to central processing units that analyze the data.80 The process of analysis in AI systems parallels that of human understanding.81 The human brain understands and analyzes the data received by eyes, ears, and hands.82 Advanced AI algorithms are trying to imitate human cognitive processes because these processes are not so different.83
Specific intent is the strongest of the mental element requirements.84 Specific intent is the existence of a purpose or an aim that a factual event will occur.85 The specific intent required to establish liability for murder is a purpose or an aim that a certain person will die.86 As a result of the existence of such intent, the perpetrator of the offense commits the offense, i.e., he performs the factual element of the specific offense.87 This situation is not unique to humans,

Personal View, in THE FOUNDATIONS OF ARTIFICIAL INTELLIGENCE 97 (Derek Partridge &
Yorick Wilks eds., 2006).
80

See generally Derek Partridge, What’s in an AI Program?, in THE FOUNDATIONS OF
ARTIFICIAL INTELLIGENCE 112 (Derek Partridge & Yorick Wilks eds., 2006).
81

Id.

82

Id.

83

See generally Daniel C. Dennett, Evolution, Error, and Intentionality, in THE FOUNDATIONS OF
ARTIFICIAL INTELLIGENCE 190 (Derek Partridge & Yorick Wilks eds., 2006); B.
Chandraswkaran, What Kind of Information Processing is Intelligence?, in THE FOUNDATIONS
OF ARTIFICIAL INTELLIGENCE 14 (Derek Partridge & Yorick Wilks eds., 2006).
84

See generally Robert Batey, Judicial Exploration of Mens rea Confusion at Common Law and
Under the Model Penal Code, 18 GA. ST. U. L. REV. 341 (2001).

85

See, e.g., Carter v. United States, 530 U.S. 255 (2000); United States v. Randolph, 93 F.3d 656
(9th Cir. 1996); United States v. Torres, 977 F.2d 321 (7th Cir. 1992); Frey v. United States, 708
So.2d 918 (Fla. 1998); State v. Neuzil, 589 N.W.2d 708 (Iowa 1999); State v. Daniels, 109 So.2d
896 (La. 1958); People v. Disimone, 650 N.W.2d 436 (Mich. Ct. App. 2002); People v. Henry,
607 N.W.2d 767 (Mich. Ct. App. 1999).
86

For the intent-to-kill murder, see WAYNE R. LAFAVE, CRIMINAL LAW 733-34 (4th ed. 2003).

87

Id.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

22

and an AI robot might be programmed to have a similar purpose or an aim and to take actions to achieve that purpose. This is specific intent.88
One might assert that humans have feelings that cannot be imitated by AI robots, not even by the most advanced robots. Examples of such feelings are love, affection, hatred, or jealousy.89 This theory might be correct in relation to the technology of the beginning of the 21st
Century;90 however, these feelings are rarely required in specific offenses. Most specific offenses are satisfied by knowledge of the existence of the factual element.91 Few offenses require specific intent in addition to knowledge.92 Almost all other offenses are satisfied by much less than that (e.g., negligence, recklessness, strict liability). Perhaps in a few specific offenses that do require certain feelings (e.g., crimes of racism, hate93), criminal liability cannot be imposed upon AI robots, which have no such feelings, but in any other specific offense, it is not a barrier.94

88

LAFAVE, supra note 86.

89

Capps, supra note 15.

90

Id.

91

HALL, supra note 26, at 632.

92

Id.

93

See generally Elizabeth A. Boyd, Richard A. Berk & Karl M. Hammer, “Motivated by Hatred or Prejudice”: Categorization of Hate-Motivated Crimes in Two Police Divisions, 30 LAW &
SOC’Y REV. 819, 842 (1996); Theresa Suozzi, F. Matt Jackson, Jeff Kauffman et al., Crimes
Motivated by Hatred: The Constitutionality and Impact of Hate Crimes Legislation in the United
States, 1 SYRACUSE J. LEGIS. & POL’Y 29 (1995).
94

HALL, supra note 26, at 105-45.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

23

If a person fulfills the requirements of both the factual element and the mental element of a specific offense, then the person is held criminally liable.95 Why should an AI robot that fulfills all elements of an offense be exempt from criminal liability? One might argue that some segments of human society are exempt from criminal liability even if both the factual and mental elements have been established. Such segments of society are infants and the mentally ill.96 A specific order in criminal law exempts infants from criminal liability.97 The social rationale behind the infancy defense is to protect infants from the harmful consequences of the criminal process and to handle them in other social frameworks.98 Do such frameworks exist for AI robots? The original legal rationale behind the infancy defense was the fact that infants are as

95

HALL, supra note 26, at 70-211.

96

See supra notes 40-45 and accompanying text.

97

Id.; and see, e.g., MINN. STAT. § 9913 (1927); MONT. REV. CODE § 10729 (1935); N.Y. PENAL
CODE § 816 (1935); OKLA. STAT. § 152 (1937); UTAH REV. STAT. 103-i-40 (1933); State v.
George, 54 A. 745 (Del. 1902); Heilman v. Commonwealth, 1 S.W. 731 (Ky. 1886); State v.
Aaron, 4 N.J.L. 269 (N.J. 1818); McCormack v. State, 15 So. 438 (Ala. 1894); Little v. State,
554 S.W.2d 312 (Ark. 1977); Clay v. State, 196 So. 462 (Fla. 1940); In re Devon T., 584 A.2d
1287 (Md. 1991); State v. Dillon, 471 P.2d 553 (Idaho 1970); State v. Jackson, 142 S.W.2d 45
(Mo. 1940).
98

See generally Frederick J. Ludwig, Rationale of Responsibility for Young Offenders, 29 NEB.
L. REV. 521 (1950); In re Tyvonne, 558 A.2d 661 (Conn. 1989); Andrew Walkover, The Infancy
Defense in the New Juvenile Court, 31 UCLA L. REV. 503 (1984); Keith Foren, In Re Tyvonne
M. Revisited: The Criminal Infancy Defense in Connecticut, 18 Q. L. REV. 733 (1999); Michael
Tonry, Rethinking Unthinkable Punishment Policies in America, 46 UCLA L. REV. 1751 (1999);
Andrew Ashworth, Sentencing Young Offenders, in PRINCIPLED SENTENCING: READINGS ON
THEORY AND POLICY 294 (Andrew von Hirsch, Andrew Ashworth & Julian Roberts eds., 3d ed.
2009); Franklin E. Zimring, Rationales for Distinctive Penal Policies for Youth Offenders, in
PRINCIPLED SENTENCING: READINGS ON THEORY AND POLICY 316 (Andrew von Hirsch, Andrew
Ashworth & Julian Roberts eds., 3d ed. 2009); Andrew von Hirsch, Reduced Penalties for
Juveniles: The Normative Dimension, in PRINCIPLED SENTENCING: READINGS ON THEORY AND
POLICY 323 (Andrew von Hirsch, Andrew Ashworth & Julian Roberts eds., 3d ed. 2009).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

24

yet incapable of comprehending what was wrong in their conduct (doli incapax).99 Later, children can be held criminally liable if the presumption of mental incapacity was refuted by proof that the child was able to distinguish between right and wrong.100 Could that be similarly applied to AI robots? Most AI algorithms are capable of analyzing permitted and forbidden.
The mentally ill are presumed to lack the fault element of the specific offense, due to their mental illness (doli incapax).101 The mentally ill are unable to distinguish between right and wrong (cognitive capabilities)102 and to control impulsive behavior.103 When an AI algorithm functions properly, there is no reason for it not to use all of its capabilities to analyze the factual data received through its receptors. However, an interesting legal question would be whether a defense of insanity might be raised in relation to a malfunctioning AI algorithm, when its analytical capabilities become corrupted as a result of that malfunction.

99

SIR EDWARD COKE, INSTITUTIONS OF THE LAWS OF ENGLAND: THIRD PART 4 (6th ed. 2001)
(1681).
100

MATTHEW HALE, HISTORIA PLACITORUM CORONAE 23, 26 (1736) [MATTHEW HALE, HISTORY
OF THE PLEAS OF THE CROWN (1736)]; and see, e.g., McCormack v. State, 15 So. 438 (Ala.
1894); Little v. State, 554 S.W.2d 312 (Ark. 1977); In re Devon T., 584 A.2d 1287 (Md. 1991).
101

Benjamin B. Sendor, Crime as Communication: An Interpretive Theory of the Insanity
Defense and the Mental Elements of Crime, 74 GEO. L.J. 1371, 1380 (1986); Joseph H.
Rodriguez, Laura M. LeWinn & Michael L. Perlin, The Insanity Defense Under Siege:
Legislative Assaults and Legal Rejoinders, 14 RUTGERS L.J. 397, 406-07 (1983); Homer D.
Crotty, The History of Insanity as a Defence to Crime in English Common Law, 12 CAL. L. REV.
105 (1924).
102

See generally Edward de Grazia, The Distinction of Being Mad, 22 U. CHI. L. REV. 339
(1955); Warren P. Hill, The Psychological Realism of Thurman Arnold, 22 U. CHI. L. REV. 377
(1955); Manfred S. Guttmacher, The Psychiatrist as an Expert Witness, 22 U. CHI. L. REV. 325
(1955); Wilber G. Katz, Law, Psychiatry, and Free Will, 22 U. CHI. L. REV. 397 (1955); Jerome
Hall, Psychiatry and Criminal Responsibility, 65 YALE L.J. 761 (1956).
103

See generally John Barker Waite, Irresistible Impulse and Criminal Liability, 23 MICH. L.
REV. 443, 454 (1925); Edward D. Hoedemaker, “Irresistible Impulse” as a Defense in Criminal
Law, 23 WASH. L. REV. 1, 7 (1948).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

25

When an AI robot establishes all elements of a specific offense, both factual and mental, it may be presumed that there is no reason to prevent imposition of criminal liability upon it for that offense. The criminal liability of an AI robot does not replace the criminal liability of the programmers or the users, if criminal liability is imposed on the programmers and users by any other legal path. Criminal liability is not to be divided, but rather, added. The criminal liability of the AI robot is imposed in addition to the criminal liability of the human programmer or user.
However, the criminal liability of an AI robot is not dependent upon the criminal liability of the programmer or user of that AI robot. As a result, if the specific AI robot was programmed or used by another AI robot, the criminal liability of the programmed or used AI robot is not influenced by that fact. The programmed or used AI robot shall be held criminally accountable for the specific offense pursuant to the direct liability model, unless it was an innocent agent. In addition, the programmer or user of the AI robot shall be held criminally accountable for that very offense pursuant to one of the three liability models, according to its specific role in the offense. The chain of criminal liability might continue, if more parties are involved, whether human or AI robots.
There is no reason to eliminate the criminal liability of an AI robot or of a human, which is based on complicity between them. An AI robot and a human might cooperate as joint perpetrators, as accessories and abettors, or the like; thus, the relevant criminal liability might be imposed on them accordingly. Since the factual and mental capabilities of an AI robot are sufficient to impose criminal liability—that is, if these capabilities satisfy the legal requirements of joint perpetrators, or of accessories and abettors—then the relevant criminal liability as joint perpetrators, accessories and abettors, or the like should be imposed irrespective of whether the offender is an AI robot or a human.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

26

Not only positive factual and mental elements may be attributed to AI robots; rather, all relevant negative fault elements should be attributable to AI robots. Most of these elements are expressed by the general defenses in criminal law, e.g., self-defense, necessity, duress, or intoxication. For some of these defenses (justifications),104 there is no material difference between humans and AI robots since they relate to a specific situation (in rem), regardless of the identity of the offender. For example, an AI robot serving under the local police force is given an order to arrest a person illegally. If the order is not manifestly illegal, the executer of the order is not criminally liable.105 In that case, there is no difference whether the executer is human or an AI robot.
For other defenses (excuses and exempts),106 some applications should be adjusted. For example, the intoxication defense is applied when the offender is under the physical influence of an intoxicating substance (e.g., alcohol or drugs). The influence of alcohol on an AI robot is minor, at most, but the influence of an electronic virus that is infecting the operating system of the AI robot might be considered parallel to the influence of intoxicating substances on humans.

104

See generally JOHN C. SMITH, JUSTIFICATION AND EXCUSE IN THE CRIMINAL LAW (1989);
Anthony M. Dillof, Unraveling Unknowing Justification, 77 NOTRE DAME L. REV. 1547 (2002);
Kent Greenawalt, Distinguishing Justifications from Excuses, 49 LAW & CONTEMP. PROBS. 89
(1986); Kent Greenawalt, The Perplexing Borders of Justification and Excuse, 84 COLUM. L.
REV. 949 (1984); Thomas Morawetz, Reconstructing the Criminal Defenses: The Significance of
Justification, 77 J. CRIM. L. & CRIMINOLOGY 277 (1986); Paul H. Robinson, A Theory of
Justification: Societal Harm as a Prerequisite for Criminal Liability, 23 UCLA L. REV. 266
(1975); Paul H. Robinson, Testing Competing Theories of Justification, 76 N.C. L. REV. 1095
(1998).
105

See generally Michael A. Musmanno, Are Subordinate Officials Penally Responsible for
Obeying Superior Orders which Direct Commission of Crime?, 67 DICK. L. REV. 221 (1963).
106

See generally Peter Arenella, Convicting the Morally Blameless: Reassessing the
Relationship Between Legal and Moral Accountability, 39 UCLA L. REV. 1511 (1992); Sanford
H. Kadish, Excusing Crime, 75 CAL. L. REV. 257 (1987); Andrew E. Lelling, A Psychological
Critique of Character-Based Theories of Criminal Excuse, 49 SYRACUSE L. REV. 35 (1998).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

27

Some other factors might be considered as being parallel to insanity or loss of control. It may be concluded that the criminal liability of an AI robot, according to the direct liability model, is not different from the relevant criminal liability of a human. In some cases, some adjustments are necessary, but substantively, it is the very same criminal liability based upon the same elements and examined under the same light.
D. Hybrids: Coordinating the Models
The possible liability models described above are not alternative models.107 These models might be applied in combination to create a full image of criminal liability in the specific context of AI robot involvement. None of the possible models is mutually exclusive. Thus, applying the second model is possible as a single model for the specific offense, and it is possible as one part of a combination of two of the legal models or of all three of them. When the AI robot plays the role of an innocent agent in the perpetration of a specific offense, and the programmer is the only person who directed that perpetration, the application of the perpetrationby-another model (the first liability model108) is the most appropriate legal model for that situation. In that same situation, when the programmer is itself an AI robot (when an AI robot programs another AI robot to commit a specific offense), the direct liability model (the third liability model109) is most appropriate to be applied to the criminal liability of the programmer of the AI robot. The third liability model in that situation is applied in addition to the first liability model, and not in lieu thereof. Thus, in such situations, the AI robot programmer shall be

107

See discussion supra Parts III.A-C.

108

See discussion supra Parts III.A.

109

See discussion supra Parts III.C.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

28

criminally liable, pursuant to a combination of the perpetration-by-another liability model and the direct liability model.110
If the AI robot plays the role of the physical perpetrator of the specific offense, but that very offense was not planned to be perpetrated, then the application of the natural-probableconsequence liability111 model might be appropriate. The programmer might be deemed negligent if no offense had been deliberately planned to be perpetrated. Alternatively, the programmer might be held fully accountable for that specific offense if another offense had indeed been deliberately planned, but the specific offense that was perpetrated had not been part of the original criminal scheme. Nevertheless, when the programmer is not human, the direct liability model must be applied in addition to the simultaneous application of the naturalprobable-consequence liability model; likewise, when the physical perpetrator is human while the planner is an AI robot.112
Hybrids of all three liability models create an opaque net of criminal liability. The combined and coordinated application of these three models reveals a new legal situation in the specific context of AI robots and criminal law. As a result, when AI robots and humans are involved, directly or indirectly, in the perpetration of a specific offense, it will be far more difficult to evade criminal liability. The social benefit to be derived from such a legal policy is of substantial value. All entities—human, legal or AI—become subject to criminal law. If the clearest purpose of the imposition of criminal liability is the application of legal social control in

110

See supra Parts III.A and III.C.

111

See discussion supra Parts III.B.

112

See supra Parts III.B and III.C.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

29

the specific society, then the coordinated application of all three models is necessary in the very context of AI robots.
IV. GENERAL PUNISHMENT ADJUSTMENT CONSIDERATIONS
Let us assume an AI robot is criminally liable. Let us assume it is indicted, tried and convicted. After the conviction, the court is supposed to sentence that AI robot. If the most appropriate punishment under the specific circumstances is one year of imprisonment, for example, how can an AI robot practically serve such a sentence? How can capital punishment, probation or even a fine be imposed on an AI robot? What is the practical meaning of imprisonment? Where no bank account is available for the sentenced AI robot, what is the practical significance of fining it?
Similar legal problems have been raised when the criminal liability of corporations was recognized.113 Some asked how any of the legitimate penalties imposed upon humans could be applicable to corporations.114 The answer was simple and legally applicable.115 When a punishment can be imposed on a corporation as it is on humans, it is imposed without change.116
When the court adjudicates a fine, the corporation pays the fine in the same way that a human
113

See generally Gerard E. Lynch, The Role of Criminal Law in Policing Corporate Misconduct,
60 LAW & CONTEMP. PROBS. 23 (1997); Richard Gruner, To Let the Punishment Fit the
Organization: Sanctioning Corporate Offenders Through Corporate Probation, 16 AM. J. CRIM.
L. 1 (1988); Steven Walt & William S. Laufer, Why Personhood Doesn’t Matter: Corporate
Criminal Liability and Sanctions, 18 AM. J. CRIM. L. 263 (1991); John C. Coffee, Jr., “No Soul to Damn: No Body to Kick”: An Unscandalised Inquiry Into the Problem of Corporate
Punishment, 79 MICH. L. REV. 386 (1981); STEVEN BOX, POWER, CRIME AND MYSTIFICATION
16-79 (1983); Brent Fisse & John Braithwaite, The Allocation of Responsibility for Corporate
Crime: Individualism, Collectivism and Accountability, 11 SYDNEY L. REV. 468 (1988).
114

Id.

115

Id.

116

Id.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

30

pays the fine and in the same way that a corporation pays its bills in a civil context.117 However, when punishment of a corporation cannot be carried out in the same way as with humans, an adjustment is required.118 Such is the legal situation vis-à-vis AI robots.
The punishment adjustment considerations examine the theoretical foundations of any applied punishment. These considerations are applied in a similar manner and are comprised of three stages. Each stage may be explained by a question: (1) What is the fundamental significance of the specific punishment for a human?; (2) How does that punishment affect AI robots?; and (3) What practical punishments may achieve the same significance when imposed on AI robots? The most significant advantage of these punishment adjustment considerations is that the significance of the specific punishment remains identical when imposed on humans and
AI robots. This method of punishment adjustment considerations is referred to below in some of the punishments used in modern societies, e.g., capital punishment, imprisonment, suspended sentencing, community service and fines.
Capital punishment is considered the most severe punishment for humans, and there is no consensus regarding its constitutionality among the various jurisdictions.119 Capital punishment is the most effective method of incapacitating offenders as it relates to recidivism since, once the death sentence is carried out, the offender is obviously incapable of committing any further

117

See sources cited supra note 113.

118

Id.

119

See, e.g., GG art. 102 (for the abolition of capital penalty in Germany in 1949); Murder
(Abolition of Death Penalty) Act, 1965, 13-14 Eliz. 2, c. 71 (for murder in Britain in 1965); and for the debate in the United States, e.g., Wilkerson v. Utah, 99 U.S. 130 (1878); In re Kemmler,
136 U.S. 436 (1890); Gregg v. Georgia, 428 U.S. 153 (1979); Hunt v. Nuth, 57 F.3d 1327 (4th
Cir. 1995); Campbell v. Wood, 18 F.3d 662 (9th Cir. 1994); Gray v. Lucas, 710 F.2d 1048 (5th
Cir. 1983); People v. Daugherty, 256 P.2d 911 (Cal. 1953); Provenzano v. Moore, 744 So.2d 413
(Fla. 1999); Dutton v. State, 91 A. 417 (Md. 1914).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

31

offense.120 The significance of capital punishment for humans is the deprivation of life.121 The
“life” of an AI robot is its independent existence as an entity. Considering capital punishment’s efficacy in incapacitating offenders, the practical action that may achieve the same results as capital punishment when imposed on an AI robot is deletion of the AI software controlling the
AI robot. Once the deletion sentence is carried out, the offending AI robot is incapable of committing any further offenses. The deletion eradicates the independent existence of the AI robot and is tantamount to the death penalty.
Imprisonment is one of the most popular sentences imposed in western legal systems for serious crimes.122 The significance of imprisonment for humans is the deprivation of human liberty and the imposition of severe limitations on human free behavior, freedom of movement and freedom to manage one’s personal life.123 The “liberty” or “freedom” of an AI robot

120

See generally ROBERT M. BOHM, DEATHQUEST: AN INTRODUCTION TO THE THEORY AND
PRACTICE OF CAPITAL PUNISHMENT IN THE UNITED STATES 74-78 (1999); Austin Sarat, The
Cultural Life of Capital Punishment: Responsibility and Representation in ‘Dead Man Walking’ and ‘Last Dance’, in THE KILLING STATE: CAPITAL PUNISHMENT IN LAW, POLITICS, AND
CULTURE 226 (Austin Sarat ed., 1999); Peter Fitzpatrick, “Always More to Do”: Capital
Punishment and the (De)Composition of Law, in THE KILLING STATE: CAPITAL PUNISHMENT IN
LAW, POLITICS, AND CULTURE 117 (Austin Sarat ed., 1999).
121

See generally Franklin E. Zimring, The Executioner’s Dissonant Song: On Capital
Punishment and American Legal Values, in THE KILLING STATE: CAPITAL PUNISHMENT IN LAW,
POLITICS, AND CULTURE 137 (Austin Sarat ed., 1999); Anthony G. Amsterdam, Selling a Quick
Fix for Boot Hill: The Myth of Justice Delayed in Death Cases, in THE KILLING STATE:
CAPITAL PUNISHMENT IN LAW, POLITICS, AND CULTURE 148 (Austin Sarat ed., 1999).
122

See generally David J. Rothman, For the Good of All: The Progressive Tradition in Prison
Reform, in HISTORY AND CRIME 271 (James A. Inciardi & Charles E. Faupel eds., 1980);
MICHAEL WELCH, IRONIES OF IMPRISONMENT (2004); Roy D. King, The Rise and Rise of
Supermax: An American Solution in Search of a Problem?, 1 PUNISHMENT & SOC’Y 163 (1999);
CHASE RIVELAND, SUPERMAX PRISONS: OVERVIEW AND GENERAL CONSIDERATIONS (1999);
JAMIE FELLNER & JOANNE MARINER, COLD STORAGE: SUPER-MAXIMUM SECURITY
CONFINEMENT IN INDIANA (1997).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

32

includes the freedom to act as an AI robot in the relevant area. For example, an AI robot in medical service has the freedom to participate in surgeries, or an AI robot in a factory has the freedom to manufacture. Considering the nature of a sentence of imprisonment, the practical action that may achieve the same effects as imprisonment when imposed on an AI robot is to put the AI robot out of use for a determinate period. During that period, no action relating to the AI robot’s freedom is allowed, and thus its freedom or liberty is restricted.
Suspended sentencing is a very popular intermediate sanction in western legal systems for increasing the deterrent effect on offenders in lieu of actual imprisonment.124 The significance of a suspended sentence for humans is the very threat of imprisonment if the human commits a specific offense or a type of specific offense.125 If the human commits such an offense, a sentence of imprisonment will be imposed for the first offense in addition to the sentencing for the second offense.126 As a result, humans are deterred from committing another offense and from becoming a recidivist offender.127 Practically, a suspended sentence is imposed

123

See generally Richard Korn, The Effects of Confinement in the High Security Unit in
Lexington, 15 SOC. JUST. 8 (1988); Holly A. Miller, Reexamining Psychological Distress in the
Current Conditions of Segregation, 1 J. CORRECTIONAL HEALTH CARE 39 (1994); FRIEDA
BERNSTEIN, THE PERCEPTION OF CHARACTERISTICS OF TOTAL INSTITUTIONS AND THEIR EFFECT
ON SOCIALIZATION (1979); BRUNO BETTELHEIM, THE INFORMED HEART: AUTONOMY IN A MASS
AGE (1960); Marek M. Kaminski, Games Prisoners Play: Allocation of Social Roles in a Total
Institution, 15 RATIONALITY & SOC’Y 188 (2003); JOHN IRWIN, PRISONS IN TURMOIL (1980);
ANTHONY J. MANOCCHIO AND JIMMY DUNN, THE TIME GAME: TWO VIEWS OF A PRISON (1982).
124

See generally MARC ANCEL, SUSPENDED SENTENCE (1971); Marc Ancel, The System of
Conditional Sentence or Sursis, 80 L. Q. REV. 334 (1964).
125

Id.

126

Id.

127

Anthony E. Bottoms, The Suspended Sentence in England 1967-1978, 21 BRITISH J.
CRIMINOLOGY 1, 2-3 (1981).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

33

only in the legal records.128 No physical action is taken when a suspended sentence is imposed.129 As a result, there is no difference between humans and AI robots. The statutory criminal records of the state do not differentiate between a suspended sentence imposed on humans, and those imposed on corporations or AI robots, as long as the relevant entity may be identified specifically and accurately.
Community service is also a very popular intermediate sanction in western legal systems in lieu of actual imprisonment.130 In most legal systems, community service is a substitute for short sentences of actual imprisonment.131 In some legal systems, community service is imposed coupled with probation so that the offender “pays a price” for the damages he caused by committing the specific offense.132 The significance of community service for humans is compulsory contribution of labor to the community.133 As discussed above, an AI robot can be engaged as a worker in very many areas.134 When an AI robot works in a factory, its work is done for the benefit of the factory owners or for the benefit of the other workers in order to ease
128

Bottoms, supra note 127.

129

Id.

130

See generally John Harding, The Development of the Community Service, in ALTERNATIVE
STRATEGIES FOR COPING WITH CRIME 164 (Norman Tutt ed., 1978); HOME OFFICE, REVIEW OF
CRIMINAL JUSTICE POLICY (1977); Andrew Willis, Community Service as an Alternative to
Imprisonment: A Cautionary View, 24 PROBATION J. 120 (1977).
131

Id.

132

See generally Julie Leibrich, Burt Galaway & Yvonne Underhill, Community Sentencing in
New Zealand: A Survey of Users, 50 FED. PROBATION 55 (1986); James Austin & Barry
Krisberg, The Unmet Promise of Alternatives, 28 J. RES. IN CRIME & DELINQ. 374 (1982); Mark
S. Umbreit, Community Service Sentencing: Jail Alternatives or Added Sanction?, 45 FED.
PROBATION 3 (1981).
133

Id.

134

See supra p. 32.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

34

and facilitate their professional tasks. In the same way that an AI robot works for the benefit of private individuals, it may work for the benefit of the community. When work for the benefit of the community is imposed on an AI robot as a compulsory contribution of labor to the community, it may be considered community service. Thus, the significance of community service is identical, whether imposed on humans or AI robots.
The adjudication of a fine is the most popular intermediate sanction in western legal systems in lieu of actual imprisonment.135 The significance of paying a fine for humans is deprivation of some of their property, whether the property is money (a fine) or other property
(forfeiture).136 When a person fails to pay a fine, or has insufficient property to pay the fine, substitute penalties are imposed on the offender, particularly imprisonment.137 The imposition of a fine on a corporation is identical to the imposition of a fine on a person, since both people and corporations have property and bank accounts. Thus, the payment of a fine is identical whether the paying entity is human or a corporate entity. However, most AI robots have no money or

135

See generally GERHARDT GREBING, THE FINE IN COMPARATIVE LAW: A SURVEY OF 21
COUNTRIES (1982); NIGEL WALKER AND NICOLA PADFIELD, SENTENCING: THEORY, LAW AND
PRACTICE (1996); Manfred Zuleeg, Criminal Sanctions to be Imposed on Individuals as
Enforcement Instruments in European Competition Law, in EUROPEAN COMPETITION LAW
ANNUAL 2001: EFFECTIVE PRIVATE ENFORCEMENT OF EC ANTITRUST LAW 451 (Claus-Dieter
Ehlermann & Isabela Atanasiu eds., 2001); Judith A. Greene, Structuring Criminal Fines:
Making an ‘Intermediate Penalty’ More Useful and Equitable, 13 JUST. SYS. J. 37 (1988);
Manfred Zuleeg, Criminal Sanctions to be Imposed on Individuals as Enforcement Instruments in European Competition Law, in EUROPEAN COMPETITION LAW ANNUAL 2001: EFFECTIVE
PRIVATE ENFORCEMENT OF EC ANTITRUST LAW 451 (Claus-Dieter Ehlermann & Isabela
Atanasiu eds., 2001).
136

See generally DOUGLAS C. MCDONALD, JUDITH A. GREENE & CHARLES WORZELLA, DAYFINES IN AMERICAN COURTS: THE STATEN-ISLAND AND MILWAUKEE EXPERIMENTS (1992);
STEVE UGLOW, CRIMINAL JUSTICE (1995).
137

See generally Use of Short Sentences of Imprisonment by the Court, REPORT OF THE SCOTTISH
ADVISORY COUNCIL ON THE TREATMENT OF OFFENDERS (1960); FIORI RINALDI, IMPRISONMENT
FOR NON-PAYMENT OF FINES (1976).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

35

property of their own, nor have they any bank accounts. In effect, the imposition of fines on AI robots may be problematic.
Assuming, however, if an AI robot did have its own property or money, the imposition of a fine on it would be identical to the imposition of a fine on humans or corporations. For most humans and corporations, property is gained through labor.138 When paying a fine, such property resulting from labor is transferred to the state.139 That labor might be transferred to the state in the form of property or directly as labor. As a result, a fine imposed on an AI robot might be collected as money or property and as labor for the benefit of the community. When the fine is collected in the form of labor for the benefit of the community, it is not different from community service as described above.140
Most common punishments are applicable to AI robots. The imposition of specific penalties on AI robots does not negate the nature of these penalties in comparison with their imposition on humans. Of course, some general punishment adjustment considerations are necessary in order to apply these penalties, but still, the nature of these penalties remains the same relative to humans and to AI robots.
V. CONCLUSION
If all of its specific requirements are met, criminal liability may be imposed upon any entity—human, corporate or AI robot. Modern times warrant modern legal measures in order to resolve today’s legal problems. The rapid development of Artificial Intelligence technology requires current legal solutions in order to protect society from possible dangers inherent in
138

JOHN LOCKE, TWO TREATISES OF GOVERNMENT (1689).

139

See supra note 135.

140

See supra pp. 33-34.

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

36

technologies not subject to the law, especially criminal law. Criminal law has a very important social function—that of preserving social order for the benefit and welfare of society. The threats upon that social order may be posed by humans, corporations or AI robots.
Traditionally, humans have been subject to criminal law, except when otherwise decided by international consensus. Thus, minors and mentally ill persons are not subject to criminal law in most legal systems around the world.141 Although corporations in their modern form have existed since the 14th Century,142 it took hundreds of years to subordinate corporations to the law, especially to criminal law.143 For hundreds of years, the law stated that corporations are not subject to criminal law, as inspired by Roman law (societas delinquere non potest).144 It was only in 1635 that an English court dared to impose criminal liability on a corporation.145
Corporations participate fully in human life, and it was outrageous not to subject them to human laws since offenses are committed by corporations or through them. But, corporations have neither body nor soul. Legal solutions were developed so that in relation to criminal liability, they would be deemed capable of fulfilling all requirements of criminal liability, including

141

See discussion supra Part III.A.

142

WILLIAM SEARLE HOLDSWORTH, A HISTORY OF ENGLISH LAW 471-76 (1923).

143

Id.

144

See generally William Searle Holdsworth, English Corporation Law in the 16th and 17th
Centuries, 31 YALE L.J. 382 (1922); WILLIAM ROBERT SCOTT, THE CONSTITUTION AND FINANCE
OF ENGLISH, SCOTISH AND IRISH JOINT-STOCK COMPANIES TO 1720 462 (1912); BISHOP
CARLETON HUNT, THE DEVELOPMENT OF THE BUSINESS CORPORATION IN ENGLAND 1800-1867 6
(Harvard University Press 1963).
145

See, e.g., Case of Langforth Bridge, 79 Eng. Rep. 919 (K.B. 1635); R v. Inhabitants of
Clifton, 101 Eng. Rep. 280 (K.B. 1794); R v. Inhabitants of Great Broughton, 98 Eng. Rep. 418
(K.B. 1771); R v. Mayor of Stratford upon Avon, 104 Eng. Rep. 636 (K.B. 1811); R v. The
Mayor of Liverpool, 102 Eng. Rep. 529 (K.B. 1802); R v. Saintiff, 87 Eng. Rep. 1002 (K.B.
1705).

Vol. 22

SYRACUSE SCIENCE & TECHNOLOGY LAW REPORTER

37

factual and mental elements.146 These solutions were embodied in models of criminal liability and general punishment adjustment considerations.147 It worked. In fact, it is still working, and very successfully.148
Why should AI robots be different from corporations? AI robots are taking larger and larger parts in human activities, as do corporations. Offenses have already been committed by
AI robots or through them. AI robots have no soul. Thus, there is no substantive legal difference between the idea of criminal liability imposed on corporations and on AI robots. It would be outrageous not to subordinate them to human laws, as corporations have been. As proposed by this article, models of criminal liability and general paths to impose punishment do exist. What else is needed?

146

See generally Frederick Pollock, Has the Common Law Received the Fiction Theory of
Corporations?, 27 L. Q. REV. 219 (1911).

147

Id.

148

Id.