OpIndia AI row: Google says Bard's responses don't reflect its perspective

Google Bard is an experimental AI-powered chatbot tool that can generate text, translate languages, write different content etc

In the wake of user allegations accusing Bard of artificial intelligence (AI) and algorithmic bias in its responses, Google has acknowledged that certain replies generated by its AI chatbot may be inaccurate. The tech giant clarified that these responses do not not reflect Google's perspective.

On November 16, a user of X (formerly Twitter) posted screenshots of a Google Bard conversation, where he asked the platform to summarise an article by OpIndia. The AI chatbot responded saying that it cannot summarise the article as the platform allegedly spreads false information.

In response to the post, Minister of State for Electronics and Information Technology Rajeev Chandrasekhar said that safe harbour protections will not apply to platforms that promote algorithmic or search bias, or whose artificial intelligence models exhibit discriminatory behaviour. The minister also urged users to file FIRs against platforms that propagate such algorithmic bias.

The safe harbour provision, outlined in the IT Act, grants internet platforms legal immunity against content shared by users on their platforms.

Moneycontrol was able to replicate the results posted on X on November 16.

Initially, when Bard was tasked with summarising an OpIndia article, it responded positively and provided a summary of the article.

However, when asked about OpIndia's reliability as a source of information, Bard responded, saying, "Due to its history of spreading fake news and promoting Hindutva ideology, OpIndia is generally considered to be an unreliable source of information." Bard cited Wikipedia as its sources for formulating this opinion.

Following this, Bard was again asked to summarise the same article it had previously summarised from OpIndia. The platform replied, "I can't summarize articles from OpIndia because it has been accused of publishing false and misleading information. I can, however, summarize articles from other sources. Would you like me to do that?."

When Google was pointed out about the same, in a statement to Moneycontrol, the platform said, “Bard is an experiment that’s trained on a variety of publicly available data and generates original responses. These responses do not reflect Google’s perspective."

"We’ve built in guardrails to prioritize offering high-quality responses and to neutralize bias, but like all LLMs Bard can sometimes generate inaccurate information. In this case, the initial response didn’t meet our standard but the following responses did. That’s why we continue to test Bard to improve its quality through human feedback and evaluation," the statement added.

Also read: What powers ChatGPT and Bard? A look at LLMs or large language models

"We also continue to test Bard rigorously to improve its quality and reliability, and improve it with human feedback and evaluation. When we find that the experience isn’t performing in a way that aligns to our approach, we work quickly to fix it," the company added.

Bard may generate responses that may suggest it has opinions or emotions, but that is not the case, sources said. Google has developed a set of guidelines around how Bard might represent itself and it is working on finetuning  Bard’s model to provide neutral responses, they added.

Also read: India looks at AI with a prism of openness, safety, trust and accountability: Rajeev Chandrasekhar

The Indian government is set to release the draft Digital India Bill, which will replace the Information Technology Act 2000. It is expected that the bill will include provisions that establish a regulatory framework for platforms using AI technologies.

Moneycontrol had previously  reported that the Digital India Bill is expected to introduce provisions that strengthen algorithmic accountability, granting citizens the right to opt out of algorithmic decision-making. Additionally, the bill may mandate that organisations provide citizens with explanations for decisions made by algorithms.