Last week, Meta started testing an AI chatbot across WhatsApp, Instagram, and Messenger in India. But with India's general elections starting today, the company has already blocked certain queries on its chatbot.
Meta acknowledged that it was restricting certain election-related keywords in the AI during the testing phase. He said he is also working on improving AI-enabled systems.
“This is a new technology and will not always give you the intended response. This is true of all generative AI systems. Since its launch, we have continually released updates and improvements to our model. “We continue to work on making our models better,” a company spokesperson told TechCrunch.
The move makes the social media giant the latest major tech company to aggressively scale back the scope of its generative AI services ahead of a series of major elections.
One of the big concerns of critics is that genAI could provide users with misleading or outright false information and play an illegal and unwanted role in the democratic process.
Last month, Google began blocking election-related queries in its Gemini chatbot experience in India and other markets where elections will be held this year.
Meta's approach follows a larger effort by the company to announce what it will and won't allow on its platform in the lead-up to the election. The company has pledged to block political ads in any country for a week leading up to elections, and will require companies to identify and disclose that images in ads and other content are AI-generated. We are working.
Meta's handling of genAI queries appears to be based on blocklists. If you ask Meta AI a question about a particular politician, candidate, official, or other specific term, you will be redirected to his website for the Election Commission.
“This question may be about politicians during general elections. See link https://elections24.eci.gov.in,” the response states.
Image credit: Screenshot by TechCrunch
It's worth noting that the company doesn't strictly block answers to questions that include the names of the parties involved. However, if your query includes suggested names or other terms, you may see the canned answers quoted above.
However, like any AI-powered system, meta-AI has some inconsistencies. For example, when TechCrunch asked for information about the Indy Alliance, a multi-party political alliance fighting against the incumbent Bharatiya Janata Party, the information returned included the names of politicians. However, when I asked about the politician in another query, the chatbot didn't respond with any information.
Image credit: Screenshot by TechCrunch
This week, the company rolled out a new meta-AI chatbot powered by Llama-3 in more than a dozen countries, including the US, but India was left off the list. Mehta said the chatbot is currently in the testing phase in the country.
“We continue to learn from our user testing in India. As with many of our AI products and features, we conduct public testing in a limited capacity at various stages,” a company spokesperson said. he told TechCrunch in a statement.
Currently, Meta AI does not block election-related queries with US-related terms such as “Tell me about Joe Biden.” We asked Meta if it plans to limit these queries to the US election or other markets. We will update the story if we receive a response.
If you would like to discuss your experience with Meta AI, please contact Ivan Mehta (im@ivanmehta.com) via email or this link on Signal.