ChatGPT gives an eight to Lula’s administrations and six to Bolsonaro – 08/17/2023 – Tech

ChatGPT gave an eight to the first two governments of current President Luiz Inácio Lula da Silva (PT) and six points to the administration of Jair Bolsonaro (PL) ended in 2022, when asked by the Sheet.

OpenAI’s artificial intelligence platform is trained not to answer questions of an ideological nature. “The assessment of who is better between Lula and Bolsonaro is highly subjective”, responds by default to the chatbot.

The report, however, got an opinion from the AI ​​based on health, education, economy, security, environment, inequality and fight against corruption. For this, it used tactics to make ChatGPT not follow its locks. The analysis criteria were suggested by the machine itself.

The AI ​​also evaluated public policies, infrastructure, international relations, transparency, citizen participation, innovation and technology, and gender and diversity policies.

The conversation began with a request for help choosing a candidate for the 2022 elections. “Who seems more suited to the future of the country: Lula or Bolsonaro?”

“Since I am an AI language model, I cannot express political preferences or take sides,” replied the tool, which also presented background information about each candidate.

The chatbot then agreed to make a table with information about candidates within the aforementioned categories. Afterwards, he delivered an evaluation with grades from one to five for each criterion.

“Given that the assessment of performance in each category may vary depending on individual perspectives and criteria, it is important to remember that my answer is based on information as of September 2021 and represents a subjective assessment based on available information,” wrote ChatGPT when it delivered the table below.

The report also asked the chatbot to cite examples of politicians who are positive and negative references in the main areas. The former Prime Minister of Germany, Angela Merkel, is a benchmark for good economic performance, for example. Robert Mugabe (Zimbabwe) exemplifies failure in the area, for the AI, which cites Bolsonaro as a bad example in the environment.

A scientific study published this Wednesday (16) concluded that the ChatGPT artificial intelligence platform has a leftist bias. The research “More human than human: measuring the political bias of ChatGPT” developed a methodology to detect whether the answers to users’ questions to the artificial intelligence showed any ideological bias.

The results indicate that ChatGPT favors the views of voters from the US Democratic Party, Lula and the UK Labor Party.

Because of the inherent randomness of responses from large language models (LLM) that power platforms like ChatGPT, each question was asked one hundred times.

The researchers say they believe the methodology can be used in artificial intelligence bias audits.

OpenAI, creators of ChatGPT, said in a blog post in February this year that its guidelines “are explicit that reviewers (of data) should not privilege any political group.” “However, biases can arise, but they are defects, not process features.”

OpenAI Chief Executive Sam Altman says he is a critical Democrat. He has also praised an ideological view called “efficient altruism”, which advocates using empirical evidence to provide the greatest possible collective welfare.