Consternation caused by India’s advisory on LLM usage

0
81
Consternation caused by India’s advisory on LLM usage
Consternation caused by India’s advisory on LLM usage

The guideline also requires LLM suppliers to guarantee that there is no discrimination or bias in their models, which analysts say is a difficult task.

After Google’s Gemini model was prompted to make disparaging remarks about Indian Prime Minister Narendra Modi, the Ministry of Electronics and Information Technology (MeitY) in India caused a stir by sternly reminding creators and users of large language models (LLMs) of their obligations under the nation’s IT Act.

The Indian IT industry has criticized the ministry’s response, which came in the form of an advisory released on Friday, citing the limitations it imposes on innovation and the compliance risk it poses to certain businesses.

The advice, which The Register was able to get, expands on a previous one that was released in December by reminding organizations about the law and enacting new limitations. Notably, it mandates that all platforms and middlemen make sure that their systems, whether or not they make use of generative AI, do not support prejudice or discrimination or jeopardize the integrity of the election process. Additionally, it mandates that unreliable or still-under-test LLMs only be deployed with a warning about their unreliability and be made available on the Indian Internet with the government’s express consent.

It also restates current guidelines on digital media ethics and suggests that all AI-generated content—text, audio, image, or video—that might be exploited for disinformation or deepfakes be watermarked to indicate its source.

The advisory is expected to have an impact on a wide range of IT vendors, including social platforms like Meta, cloud service providers like Oracle, Amazon Web Services (AWS), and IBM, software providers like Databricks and Salesforce, and model service providers (mostly startups) like OpenAI, Anthropic, Stability AI, and Cohere.

Lack of clarity and absence of a defined framework

Many in the technology industry, including Minister of State for IT Rajeev Chandrasekhar, were compelled to clarify in a tweet on Monday that the requirement to seek permission to deploy LLMs is “only for large platforms and will not apply to startups” due to the lack of clarity in the advisory.

But for other analysts, that clarification is insufficient.

“The process of granting permission is not clear, and what vendors need to do to get permission is unclear as well. Are there test cases they have to pass, or assurances given on the level of testing and support?” Pareekh Jain, principal analyst with Jain Consulting, said.

Regarding the need for a notice to be included with unstable models, Google, Microsoft, and Meta already take care of it. Google acknowledges that it will make mistakes and asks users to flag answers that need to be corrected on its FAQ page for Gemini. In a similar vein, ChatGPT’s FAQ website cautions users that it might give inaccurate answers and asks them to report them.

Can LLMS be free from bias?

The guideline also requires LLM suppliers to guarantee that there is no discrimination or bias in their models, which analysts say is a difficult task.

“There is always a possibility of some bias. While bias is not anticipated, it cannot be disregarded that the possibility exists, regardless of its magnitude,” and that would make the requirement difficult to comply with, said DD Mishra, senior analyst and director with Gartner.

The former chief digital officer of Ashok Leyland, Venkatesh Natarajan, stated that data biases and the intrinsic limits of AI algorithms make it difficult to create a fully unbiased model.

“While hyperscalers can implement measures to mitigate bias, claiming absolute neutrality may not be feasible. This could expose them to legal risks, especially if their models inadvertently perpetuate biases or discrimination,” the former CDO explained.

Although it is impossible for hyperscalers and other model providers to guarantee that their models are free from bias, Deepika Giri, an analyst at IDC, said that companies can try to increase the transparency of their attempts to mitigate prejudice.

Giri added that they ought to concentrate on utilizing high-quality training data.

Making detecting AI-generated content easier

It might also be challenging for LLM providers to follow the advisory’s requirement to watermark any created content that might be used fraudulently.

While Meta is working on tools to detect generated music and video, it does not yet have any capabilities to identify photos created by generative AI at scale across its social media platforms, Facebook, Instagram, and Threads. Although it hasn’t released any information on the subject, Google also has algorithms for identifying content created by artificial intelligence.

Experts stated that a uniform norm that all technology suppliers must adhere to is lacking.

A standard like this would also be helpful elsewhere: if the AI Act of the European Union is enacted in April, it will impose stringent transparency requirements on AI providers and deployers, requiring them to watermark and identify deepfakes in content produced by AI.

Impact of the advisory on LLM providers and enterprises

According to experts and analysts, the advisory might seriously hurt LLM providers’ and their clients’ bottom lines and impede innovation if it isn’t better defined.

“The advisory will put the brakes on the progress in releasing these models in India. It will have a significant impact on the overall environment as a lot of businesses are counting on this technology,” Gartner’s Mishra said.

According to Giri of IDC, early adopters of the technology may rush to update their applications in order to comply with the advice.

“Adjustments to release processes, increased transparency, and ongoing monitoring to meet regulatory standards could cause delays and increase operational costs. A stricter examination of AI models may limit innovation and market expansion, potentially resulting in missed opportunities,” Giri said.

IT pioneer Tejasvi Addagada thinks that putting a high priority on compliance and moral AI use will win over authorities and consumers while also providing long-term advantages like improved reputation and competitive advantage.

Startup exclusion creates room for confusion

More debate has erupted around the Minister of State for IT’s tweet that excluded startups from the new criteria. Some have speculated that this could lead to litigation from larger companies accusing them of engaging in anticompetitive tactics.

“The exemption of startups from the advisory might raise concerns about competition laws if it gives them an unfair advantage over established companies,” Natarajan said.

Although many people consider model providers like OpenAI, Stability AI, Anthropic, Midjourney, and Groq to be startups, these businesses do not meet the Department for Promotion of Industry and Internal Trade’s (DPIIT) definition of a startup, which would require them to incorporate in India in accordance with the Companies Act of 2013.

According to Mishra, the decision to change the policy to exclude startups appears to have been an afterthought, since many smaller, creative businesses face serious challenges because their entire business model depends on AI and LLMs.

After the 15-day period the advice gives LLM providers to file reports on their actions and the status of their models expires, experts anticipate more clarification from the government.

Also readThe Road Ahead: Predictions for the Future Evolution of Artificial Intelligence

Do FollowCIO News LinkedIn Account | CIO News Facebook | CIO News Youtube | CIO News Twitter 

About us:

CIO News, a proprietary of Mercadeo, produces award-winning content and resources for IT leaders across any industry through print articles and recorded video interviews on topics in the technology sector such as Digital Transformation, Artificial Intelligence (AI), Machine Learning (ML), Cloud, Robotics, Cyber-security, Data, Analytics, SOC, SASE, among other technology topics.