Google's Bard AI chatbot is embarking on a worldwide adventure, traversing far beyond its familiar home. The chatbot is now officially available in several sought-after countries, including most of the European Union, after fulfilling stringent requirements imposed by regulators, such as submitting regular privacy reports and ensuring data security. Notably, Bard has expanded its linguistic prowess and is now capable of singing in 43 additional languages, including Arabic, Danish, French, German, Hindi, Chinese, Spanish, Portuguese and more.
This update has a global reach, enabling users worldwide to benefit from the chatbot's capabilities, including those in Brazil and across the European Union, following a previous delay due to privacy concerns. Google's ongoing efforts to narrow the gap with competitors like OpenAI and Microsoft involve promising future updates for Bard as well as conducting a beta test for an AI-enabled Google Search to rival Microsoft's ChatGPT-powered Bing search. Although Google has not provided an official statement on the number of supported dialects, ChatGPT is known to handle numerous languages proficiently.
In addition to expanding access, Bard, Google's AI chatbot, is receiving new features that allow users to pin and rename conversations on the left side of the Bard page. Users can now also listen to audio versions of responses and share links to Bard outputs. Furthermore, Google is finally enabling users to utilize a Google Lens-like feature, which enables image sharing with Bard. The chatbot can describe images and generate captions, although this feature is currently limited to English, with Google promising future availability in more languages.
To enter the EU market, Google had to make privacy concessions due to increased AI regulations. Recently, the company faced a proposed class action lawsuit for its commitment to scraping the entire internet to train its AI services. Google has agreed to provide privacy reports to regulators in the coming months as part of the agreement with Ireland's Data Protection Commission. OpenAI, the creator of ChatGPT, is also navigating AI regulations and has made its chatbot app available in several European countries. To ensure safety, Google's Bard team conducted tests with human participants, challenging the AI to produce harmful responses in different proposed languages.
Modern AI chatbots face the inherent problem of being heavily influenced by Western-centric training data, which is not explicitly mentioned. The existing training data already enables these chatbots to understand languages to some extent. Despite claims by Google that Bard had not been trained in Bengali, experts pointed out that Bengali and other similar languages were already present in its training data. The training data for GPT-3, the language model behind Bard, primarily consisted of English-language Wikipedia, books and a widely used dataset called Common Crawl, which predominantly reflects a Western perspective. As a result, the AI's output will inevitably exhibit a Western bias, regardless of the language it is currently using. Google plans to train the AI based on user inputs, but closing the cultural divide will require a substantial amount of voluntarily provided data from users.