Who is responsible if an AI system makes a mistake?

Liability in case of error committed by AI systems is spread among different parties including developers, manufactures, organizations using the AI and end-users. Responsibility is subject to the context and reason behind the mistake, can it be a failure in the design of AI, used incorrectly by an individual, or the organization failed to apply it properly. Even when AI is driven autonomously the human actors are still legally and ethically liable, because AI cannot itself be liable.

Key Responsible Parties

Developers and Manufacturers: The developers and manufacturers are responsible in case of the errors that may occur due to the defects in the design of the AI, the programming, or the biased training data. Their task is to ensure accuracy, ethical design and the mitigation of bias.

Organizations Implementing AI: Companies that implement AI in their processes should make sure that their AI is well integrated and monitored. These organizations can be liable in case AI is used improperly or in cases where risks are predictable but not taken care of.

End-Users: It is the responsibility of those users who use the outputs of AI to be judgmental and practice due diligence. Nobody can use AI wisely or neglect their limitations, and this can place end-users in part of the blame.

Regulators and Policymakers: They set the limits and guidelines of AI implementations but they frequently have difficulties to keep up with technological changes.

Factors and complexity that affect Liability.

• The type of the error (design fault, abuse, weakness in the learning phase).
• Human control and intervention.
• Were there any guidelines to use AI, or were they used unintentionally?
• The implication of open-source or third-party association.
• Laws in existence regarding the use of AI and liability.