Concentration of Authority

"Militarization, Surveillance, and Weaponization: Concentrating power can occur at increasing levels, from small groups to national bodies. Code generative systems can improve development for technical surveillance systems and language models can be used to surveil text communication within work, social, and other environments [1]. Generative AI mechanisms for accumulating power and control at a national level, such as surveillance, has not yet happened, but government and military interest in deploying and weaponizing generative AI systems is growing [106]. Use includes generating synthetic data for training AI systems [102] and military planning [78]. Military use is not inherently weaponization and risk depends on the use case and government interest. Favorable arguments use AI to protect national security and require differentiating national security interests from undue harm [44]. Generative AI systems are also enabling new kinds of cyberattacks, and amplifying the possibilities of existing cyberattacks. For example, synthetic audio has been used to copy the sound of someone’s voice for more compelling fraud and extortion [124]. Large language models are also facilitating disinformation campaigns, influence operations, and phishing attacks [92]."(p. 15)

"Global deployment of a model can consolidate power within a single, originating culture, to determine and propagate acceptability [245] across cultures [150]. Highest performing characteristics of generative systems such as language, dominant cultural values, and embedded norms can overrepresent regions outside of where a system is deployed. For example, a language model that is highest performing in the English language can be deployed in a region with a different dominant language and incentivize engaging in English. Establishing or reinforcing goodness with certain languages, accents, imagery, social norms, and other representations of peoples and cultures can contribute to this norms and values imposition. Certain modality characteristics such as language carry within it its own logics and frames. Though English as a lingua franca is globally beneficial, the consequences of its dominance as a result of a historic process of militarised colonization should be examined. Insidious effects which generative AI systems could further embed include the erosion of global multilingualism, undermine the right to language and culture, and further marginalize the necessity for widespread multilingual education. The effects of generative AI systems on child development, including the technologically mediated socialisation of norms and values is also an area to be inquired. These are in addition to the emotional and behavioural effects of chatbots on children. This, according to UNICEF [248], included the enforcement of "bias, given that they often select a predetermined reply based on the most matching keywords or similar wording pattern"."(p. 15)