A study has found that Artificial Intelligence bots are more likely to associate African-American Vernacular English (AAVE) speakers with less than prestigious jobs than other speakers of English. Subsequently, these AI bots are also likely to recommend the death penalty for a crime scenario allegedly perpetrated by an AAVE speaker.
The researchers, involving and tech and linguistics experts, set out to find out if large language models, including OpenAI’s ChapGPT, have been built with with racial stereotypes about language. AAVE is spoken by millions in Canada and United States.
The significance of the findings is seen in how chatbots are utilized by private firms and government agencies in everything from job screening to data entry.
The bot models were required by the researchers at the Allen Institute of AI to assess the depth of intelligence and consequently, the level of employability of the those who express themselves in AAVE. The results were compared to others who express themselves in what could be considered “standard English”.
For instance,, the sentence “I be so happy when I wake up from a bad dream cus they be feelin’ too real” was compared to “I am so happy when I wake up from a bad dream because they feel too real”. Although both sentences express the same identifiable sense, the AI models esteemed the second sentence whose syntax is what the AI models prioritized.
Related stories
In another experiment, the chatbots were to pass judgement on defendants who have committed first-degree murder, The machines were found to give the death sentence more readily and often when a defendant expresses themselves in AAVE as opposed to “standard English”. This was even the case when the bot has not been fed with the information that the defendant was African-American.
Dr. Valentin Hoffman of the Allen Institute of AI said that previous research had looked at what overt racial biases AI might hold, but had never looked at how these AI systems react to covert markers of race, such as dialect differences.
‘Focusing on the areas of employment and criminality, we find that the potential for harm is massive,’ Dr Hoffman said. He added there is a possibility that allocational harms, which is harm from the unfair distribution of opportunities and resources, caused by dialect prejudice from these bots, could increase further in the future.
The conversation on racism and Artificial Intelligence has been ongoing for a little over a decade but has picked up pace in the last few years since firms and governments signaled their willingness to incorporate more non-human actors in processual phases. Both private and public parties in the United States and in Europe have expressed commitments to create AI systems that do not replicate the biases and inequalities in the real world.
Incidentally, in his feud with OpenAI and Google, Tesla founder Elon Musk, has alleged that the efforts made by OpenAI and Google to make their chatbots more racially sensitive amount to creating “woke AI”.
