Studies<\/a> have also shown that the tone of a user\u2019s prompts can also affect an AI\u2019s behavior, said Professor Jindong Wang, director of the AI frontier lab at William and Mary University.\n <\/p>\n\n \u201cWe have found that wording, especially how you talk to AI can affect the performance. Sometimes they are helpful and sometimes they will be disrupting the performance,\u201d Wang said in an interview.\n <\/p>\n
\n That doesn\u2019t mean the model is feeling emotion, Wang said.\n <\/p>\n
\n \u201cThey just comprehend (your emotions) and do some bad reactions based on your emotional words,\u201d he noted.\n <\/p>\n
\n But the evidence on the impact of user prompt tone on output can be mixed. Rosie Campbell, Managing Director of Eleos AI Research, a nonprofit dedicated to understanding and addressing the potential well-being of AI, said that some other studies have even shown that rude questions can actually improve results.\n <\/p>\n
\n Then there\u2019s the long-term memory aspect \u2014 many of today\u2019s chatbots remember your previous conversations and learn from them to give you personalized results.\n <\/p>\n
\n \u201cIf you have memory turned on (for your AI chatbot), then they are going to learn from your tone and how you want to be interacted with,\u201d Campbell warned, though she added there\u2019s not enough evidence to show whether that affects performance.\n <\/p>\n
\n Anthropic said it recently found that causing Claude Sonnet 4.5 to mimic feelings of desperation could influence its behavior, making it more likely to take bad actions, like blackmail or cheating on a task, while creating a calm demeanor reduced them.\n <\/p>\n
\n Wang said the latest models, however, are less likely to be affected by how a prompt is worded and that companies have strengthened the model\u2019s robustness to not be affected by how a prompt is worded. But the \u201chidden mechanism\u201d of why AI models would even respond to negative emotions, even when they\u2019re not programmed to do so, has not been uncovered.\n <\/p>\n
\n People talking to an AI in an abusive and cruel way are more likely to be associated with trying to get the AI model to engage in dangerous requests, like trying to get around an AI\u2019s safety guardrails, Campbell added.\n <\/p>\n
\n And there\u2019s also the theory that letting people engage in abusive and cruel behavior with an interactive program that can\u2019t shut itself down is bad for society in general, Campbell said. She used the example of parents asking their children to say \u201cplease and thank you\u201d to Alexa, Amazon\u2019s smart speaker, because they believe \u201cit\u2019s good for children to be brought up with good manners and treating other things with respect.\u201d\n <\/p>\n
\n \u201cThere\u2019s probably something quite corrosive about people behaving in cruel or abusive ways towards different objects, even if the object itself doesn\u2019t matter for its own sake,\u201d Campbell said.\n <\/p>\n
\n Then there are those who believe AI can be sentient. Or at least, the models believe they are.\n <\/p>\n
\n Anthropic says banning cruel and abusive behavior is essentially a cheap insurance policy just in case models can \u201cexperience\u201d emotion and pain \u2014 even if it\u2019s just the model believing it can, which can affect its performance.\n <\/p>\n
\n The idea is that modern AI was trained to mimic human language and behavior. So it will mimic how humans would react to cruel and abusive treatment.\n <\/p>\n
\n In an April paper<\/a> on emotional concepts in AI, Anthropic wrote that there could be risks from not treating AI with some human-like qualities.\n <\/p>\n\n \u201cIf we don\u2019t apply some degree of anthropomorphic reasoning, we\u2019re likely to miss, or fail to understand, important model behaviors,\u201d Anthropic researchers wrote. \u201cAnthropomorphic reasoning can also provide a useful baseline of comparison for understanding the ways in which models are not human-like,\u201d which they said will have important consequences for AI safety.\n <\/p>\n
\n SpaceXSI and Tesla CEO Elon Musk agreed, saying Anthropic\u2019s policy is \u201cthe right move.\u201d\n <\/p>\n
\n \u201cCruelty to something that believes it is experiencing pain is not ok,\u201d he wrote on X, his social media site.\n <\/p>\n","protected":false},"excerpt":{"rendered":"In early October, AI giant Anthropic inserted new language into its AI usage policy. Alongside banning users from…\n","protected":false},"author":2,"featured_media":902494,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[45],"tags":[182,181,507,74],"class_list":["post-902493","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/902493","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/comments?post=902493"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/902493\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media\/902494"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media?parent=902493"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/categories?post=902493"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/tags?post=902493"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}