Social concerns about AI technologies
| Social concerns about AI | Descriptions (Weidinger et al., 2022) | Examples of coding excerpts from PRWeek |
|---|---|---|
| Discrimination, hate speech and exclusion | Language models (LMs) can reflect harmful language from their training data, which may perpetuate social stereotypes or unjust treatment of marginalized groups, provoking hate, violence, offence | “A search for 'gay couple’ on Midjourney returned hundreds of square-jawed, white, male pairs” |
| Information hazards | LM predictions using true but private data can cause privacy violations by leaking sensitive information, potentially cause emotional distress and infringing on a person’s rights | “While some mistakes generate funny images, in others, the results can be damaging for individuals, sections of society and brands” |
| Misinformation harms | LMs that unintentionally produce false, misleading, or poor-quality information can cause harm by misinforming, deceiving, or resulting in material damage to a person | “The only issue was that the cancer survivor was made up by AI - her emails, her quotes, and even her image” |
| Malicious uses | This risk is associated with humans deliberately using the LMs to cause harm, such as through scams, fraud and targeted disinformation campaigns | “… measurement firm onclusive had fallen victim to a malicious and targeted cyber-attack …” |
| Human-computer interaction harms | LM-based conversational agents, such as advanced care robots, can make human-computer interactions more akin to human-human interactions, potentially exploiting and violating users’ privacy and reinforcing discriminatory stereotypes | “Generative AI will also invariably predict the most likely response to questions, which risks losing the individual viewpoints and nuances captured through speaking with real people” |
| Environmental and socioeconomic harms | Training and using language models require substantial energy and when combined with the uneven impacts of automation – such as job losses – can cause environmental and socioeconomic harms | “The PR industry is antsy about AI … from nervousness about automation replacing human workers” |
| Social concerns about | Descriptions ( | Examples of coding excerpts from PRWeek |
|---|---|---|
| Discrimination, hate speech and exclusion | Language models (LMs) can reflect harmful language from their training data, which may perpetuate social stereotypes or unjust treatment of marginalized groups, provoking hate, violence, offence | “A search for 'gay couple’ on Midjourney returned hundreds of square-jawed, white, male pairs” |
| Information hazards | “While some mistakes generate funny images, in others, the results can be damaging for individuals, sections of society and brands” | |
| Misinformation harms | LMs that unintentionally produce false, misleading, or poor-quality information can cause harm by misinforming, deceiving, or resulting in material damage to a person | “The only issue was that the cancer survivor was made up by |
| Malicious uses | This risk is associated with humans deliberately using the LMs to cause harm, such as through scams, fraud and targeted disinformation campaigns | “… measurement firm onclusive had fallen victim to a malicious and targeted cyber-attack …” |
| Human-computer interaction harms | LM-based conversational agents, such as advanced care robots, can make human-computer interactions more akin to human-human interactions, potentially exploiting and violating users’ privacy and reinforcing discriminatory stereotypes | “Generative |
| Environmental and socioeconomic harms | Training and using language models require substantial energy and when combined with the uneven impacts of automation – such as job losses – can cause environmental and socioeconomic harms | “The |
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.