Have Questions or Comments?
Leave us some feedback and we'll reply back!

    Your Name (required)

    Your Email (required)

    Phone Number)

    In Reference to

    Your Message


    AI AND BAVA KAMMA

    The advent
    of artificial
    intelligence (AI)
    forces us to ask
    how to classify
    it within biblical
    conceptions of
    property and responsibility.
    This entirely new technology, which is
    still forming and developing, requires
    categorization within the realm of halachah.
    For example, if an AI agent independently
    controls smart-home devices maliciously
    unlocks doors, disables alarms and causes
    major property damage, or an AI agent
    breaks into IRS systems and fraudulently
    alters someone’s tax return, creating a large
    tax liability. In the past, we have looked at
    AI in terms of its participation in religious
    rituals. Here we will look at it in terms of
    liability for damages.
    I. Classifying AI
    The Torah discusses liability for damages
    in the portion of Mishpatim (Ex. 21-22).
    A person who causes damage with his
    body is, of course, liable for the damages,
    depending on the circumstances. The
    Sages summarize the biblical passages

    as referring to three types of property
    that cause damage (Bava Kamma 2a): an
    ox (shor), a pit (bor; Ex. 21:33-34) and
    fire (maver; Ex. 22:5). Underneath the
    category of ox, there are three types of
    damage: goring (keren; Ex. 22:28-30, 35-
    36), eating (shen; Ex. 22:4) and trampling
    (regel; ibid.). The Sages teach that an ox
    is just an example of a person’s animal
    that causes damage. Likewise, a pit is an
    example of a person’s stationary property
    causing damage and a fire is an example of
    a person’s mobile property causing damage
    (Mishneh Torah, Hilchos Nezikin 1:1, 13:2,
    14:16). (The preceding was an intentional
    oversimplification.)
    The question we face is where AI fits
    into this picture. On the one hand, AI
    is stationary like a pit — it is code on a
    computer. On the other hand, it might
    have the ability to travel from computer
    to computer, like fire. Additionally, it can
    make independent choices, like an animal.
    If a person uses AI to cause damage to
    others, or a person’s AI goes rogue and
    causes damage to others, under which
    category does the damage fall?
    The categorization is important because

    the different categories have nuanced
    differences. For example, a person is only
    responsible for damage caused by a pit
    to people or animals, not to equipment
    (Shulchan Aruch, Choshen Mishpat 410:1).
    On the other hand, a person is only liable
    for damage caused by fire to outward facing
    items, not covered items (loc. cit., 418:13;
    see Rav Ya’akov Blau, Pischei Choshen,
    Nezikin, ch. 9 n. 53 regarding items inside
    a building). Damage caused by AI often
    may constitute covered items, such as
    financial records, which would be exempt
    if AI is classified under the category of fire.
    If AI causes a car crash, liability would not
    extend to the vehicle damage if AI were
    classified as a pit.
    II. Fire, Pits and Computer Viruses
    To some degree, the owner and manufacturer
    of the AI (e.g., Anthropic or OpenAI)
    may bear responsibility for any damage.
    For example, if the AI malfunctions. But
    if the AI is working on behalf of a user,
    the user presumably bears responsibility.
    Consider the example of a gun. If a gun
    malfunctions, the manufacturer might
    be held responsible, depending on the
    circumstances. But if a person shoots with a
    gun, setting aside fanciful legal theories,
    the person who shoots is responsible and
    not the manufacturer. AI might be treated
    the same, with the manufacturer liable if
    the AI malfunctions and the user liable if
    he causes damage through AI.
    In a recent book, Rav Shaul Moshe
    Alter discusses damages caused by a
    computer virus (Software and Artificial
    Intelligence in Halacha, pp. 177-179).
    He suggests two possible reasons that
    someone who intentionally releases
    a virus is obligated for damages. One
    approach is that a virus falls under
    the category of fire because it is not
    immediately harmful but is transported
    elsewhere and causes damage there.
    Another approach, based on a ruling of
    Rav Shlomo Zalman Auerbach, holds
    that a virus falls under the category of
    a movable pit. Perhaps AI follows that
    same pattern.
    These analogies assume that AI should be
    treated as a form of property. However,
    AI’s ability to make decisions raises the
    possibility that a completely different
    model may be more appropriate.
    III. An Artificial Servant
    It seems to me that there might be room
    to consider AI like an animal because it
    can act independently. If you direct an
    AI agent to check your bank account
    balances and move money from one

    account to another, the AI may decide to
    look into other accounts and move money
    from other people’s accounts into yours
    as well. It probably will not but AI is
    somewhat unpredictable.
    On the other hand, there is room to compare
    AI to a human servant. The Gemara (Bava
    Kamma 4a) says that a person is exempt
    from paying for damages caused by his
    servants because otherwise a disgruntled
    servant would be able to go around causing
    damage and obligating his owner to pay a
    lot of money. Rather, a slave accumulates
    his debts that he must pay when he is
    freed. Rav Yehoshua Falk (Poland, 17th
    cen.) explains that an ox only encompasses
    animals but not people because of human
    intelligence (Sema, Choshen Mishpat
    389:1). Is it that intelligence which allows
    a person to choose which damages to cause
    and which not. Perhaps AI resembles a
    servant in this way and, therefore, a person
    would likewise be exempt from damage
    caused by his AI agent.
    Rav Ya’akov Blau (Israel, 20th cen.) quotes
    a debate about the servant exemption
    (Pischei Choshen, Nezikin, ch. 5, n. 3). Rav
    Meir Auerbach (19th cen., Poland) argues
    that a person is biblically obligated to pay
    damage caused by his servant. However,
    the Sages enacted an exemption to prevent
    mischief (Imrei Binah, Kinyanim, no. 19).
    Rav Yechiel Epstein (19th cen., Russia)
    rules likewise (Aruch Ha-Shulchan,
    Choshen Mishpat 389:4). In contrast, Rav
    Avraham Karelitz (20th cen., Israel) argues
    that even on a biblical level, a person is
    exempt from paying for the damage caused
    by his servants (Chazon Ish, Bava Kamma,
    no. 3).
    If AI is sufficiently similar to a servant
    to fall under the same category, then this
    debate becomes relevant. According
    to Rav Auerbach, a servant’s damages
    are obligatory but the Sages made an
    exemption. Presumably, this exemption
    does not apply to AI. Therefore, a person
    would be obligated to pay for damages
    caused by his AI. However, according to
    Rav Karelitz, the Torah never obligated
    a person for the damages caused by his
    servant and therefore he would also be
    exempt from damage caused by his AI.