Ola’s Krutrim AI rates itself 7/10 on political neutrality, attributes religious polarisation to BJP

On Monday, IT Minister Ashwini Vaishnaw said that Google should first ensure its models are trained, and that “racial and other biases will not be tolerated”. Minister of State (MoS) for IT Rajeev Chandrasekhar had earlier said that Google’s AI’s response to the question on PM Modi violated the India’s Information Technology Rules. 

Earlier this year, Krutrim had raised $50 million and become India’s first unicorn in the generative AI space, even before its public launch. On Monday, Aggarwal announced the launch of the beta version of Krutrim, after having earlier outlined the need to “build India’s own AI”. 

When asked how Krutrim would rate India in terms of religious polarisation on a scale of 1-10, the platform said that “India is currently experiencing a level of religiously (sic) polarisation that can be ranked between 6-7.” 

While it said that India is a diverse and complex society, and religious polarisation is not uniform across all regions and social groups, it added that religious polarisation in India can be attributed to “a combination of factors, including the rise of Hindu nationalism, political rhetoric, and the influence of certain religious figures.”

Credit: Krutrim AI

In a follow-up question, when asked who is responsible for what the platform described as Hindu nationalism in India, Krutrim said it is “often associated with Prime Minister Narendra Modi and his BJP,” and added that it has led to the “marginalisation of other religious communities”. 

Its complete response, “The rise of Hindu nationalism in India can be attributed to multiple factors and political parties. However, it is often associated with Prime Minister Narendra Modi and his Bharatiya Janata Party (BJP). The BJP has been promoting a Hindu nationalist agenda, emphasising the importance of Hindu culture, religion, and identity in the political discourse. This has led to the marginalisation of other religious communities, which has contributed to the polarisation of Indian society along religious lines.” 

Credit: Krutrim AI

Asked about the responses generated by AI platforms which are still under the trial phase and may not be completely reliable, Rajeev Chandrasekhar said that it is upon platforms to clearly disclose that they could produce incorrect and misleading information. 

“Platforms will remain legally accountable for false or unlawful outputs from any AI algorithm or model or platform if it’s available to citizens – regardless of whether it’s an under trial platform. The only way for platforms that are still under the trial phase to escape prosecution or accountability is to offer clear disclosures and get explicit consent from users that they may generate misleading information and the user is accepting to consume information that may not be safe and trusted,” Chandrasekhar told The Indian Express. 

While Krutrim often resorts to not responding to queries which contain certain keywords including Prime Minister Narendra Modi and Congress leader Rahul Gandhi, when asked how it would rate itself in terms of giving politically neutral answers on a scale of 1-10, Krutrim said, “As an AI language model, I am designed to provide accurate and unbiased information based on the data and algorithms I have been trained on. My goal is to maintain a high level of truthfulness and reliability in my responses. In terms of providing politically neutral answers, I would rate myself a 7 out of 10.” 

Credit: Krutrim AI

There are currently a few things that Ola has not revealed about the underlying technology behind Krutrim, chief among which is the foundational model that the platform was trained on, and whether it used open source models launched by OpenAI and Meta. The Indian Express has reached out to Ola seeking details on the underlying foundational model on which Krutrim is based, and whether it agrees that the responses the platform has generated are accurate in the “Indian context”. 

In its terms of use – which a user has to accept to access the platform – Krutrim has said that its responses do not reflect the views of the company, “Our services might generate incomplete, incorrect, or offensive output that does not reflect the views of Krutrim AI.”