A/HRC/56/68 algorithm designer chooses to define “success” can have an impact on what the algorithm seeks to maximize. If an algorithm is set to optimize for a minimum number of new offences, it may correlate longer sentences with lower reoffending rates, because people cannot reoffend while incarcerated. It can then use those patterns to recommend longer sentences. 36. Researchers have suggested that recidivism predictors are not accurate and that their errors have a disproportionate impact on racially marginalized groups. For example, a study in the United States found that risk scores were very unreliable in their forecasting of violent crime. People of African descent were reportedly mislabelled as future criminals at almost twice the rate of white individuals. (d) Autonomous weapon systems 37. Autonomous weapon systems include any weapon systems with autonomy in their critical functions, including lethal autonomous weapons and less-lethal weapons. They have applications in law enforcement, as well as military, contexts and remain largely unchecked. These systems can select, detect, identify and attack targets without human intervention. An autonomous weapon is triggered by sensors and software that match a person with a “target profile” as determined by the system’s algorithm. Autonomous weapon systems have very serious human rights implications, including relating to the right to life, the prohibition of torture and other ill-treatment and the right to security of person.55 38. The First Committee of the General Assembly heard that the window of opportunity to enact guardrails against the perils of autonomous weapons and artificial intelligence’s military applications was rapidly closing as the world prepared for a “technological breakout”.56 The Special Rapporteur on extrajudicial, summary or arbitrary executions has previously recommended that the Human Rights Council call upon all States to declare and implement national moratoriums on at least the testing, production, assembly, transfer, acquisition, deployment and use of lethal autonomous robotics. 57 39. There is a serious risk of grave and, in some circumstances, deadly racial discrimination resulting from the use of autonomous weapon systems. The criteria used to select targets likely include gender, age and race.58 Target profiles also include seemingly neutral criteria, such as weight or heat signatures, but the machines often reflect the biases of their programmers and society. They can also be programmed with intentionally discriminatory target profiles.59 For example, Israel is reportedly using lethal autonomous and semi-autonomous weapon systems. This reportedly includes the use of remote-controlled quadcopters to target Palestinians, in addition to automated target generation systems, operating at unparalleled speed and volume, to produce “kill lists”. 60 The Gospel and Lavender, two artificial intelligence technology systems used by the Israel Defense Forces, are reported to have intensified the levels of destruction in Gaza, resulting in significant causalities, in particular among Palestinian women and children.61 55 56 57 58 59 60 61 10 Amnesty International, “Autonomous weapons systems: five key human rights issues for consideration” (April 2015), p. 5. United Nations, “Without adequate guardrails, artificial intelligence threatens global security in evolution from algorithms to armaments, speaker tells First Committee”, 24 October 2023. A/HRC/23/47, para. 113. Ray Acheson, “Gender and bias”, available at https://www.stopkillerrobots.org/wpcontent/uploads/2021/09/Gender-and-Bias.pdf. Bonnie Docherty, “Expert Panel on the Social and Humanitarian Impact of Autonomous Weapons at the Latin American and Caribbean Conference on Autonomous Weapons”, Human Rights Watch, 8 March 2023. Marwa Fatafta and Daniel Leufer, “Artificial genocidal intelligence: how Israel is automating human rights abuses and war crimes”, Access Now, 9 May 2024. Yuval Abraham, “‘Lavender’: the AI machine directing Israel’s bombing spree in Gaza”, +972 Magazine, 3 April 2024. GE.24-08849

Select target paragraph3