Redefining Chatbot Accuracy: Google's Bard Takes a New Approach

NNicholas September 20, 2023 4:11 AM

Google's advanced chatbot, Bard, has taken a significant step towards addressing the issue of AI-generated misinformation. The chatbot now includes a 'double-check' feature that evaluates its responses for accuracy, indicating a potential turning point in the struggle against AI falsehoods.

Understanding chatbot inaccuracies

Chatbots, including Google's Bard, often generate responses that are 'confidently wrong', in industry parlance. They don't draw on a verified database of facts but instead make predictive responses based on the extensive volumes of text from which their underlying language models were trained. The issue here is that these predictive models can create believable but entirely fabricated information. So, while they can rapidly generate responses, verifying the accuracy of these responses often negates the time-saving benefits of using a chatbot in the first place.

Introducing Bard's 'double-check' feature

Google has made a noteworthy advance by introducing a 'double-check' feature in Bard. Now, after Bard provides an answer to your query, you can use this feature to verify the response against web content. When you click on the 'Google It' button, Bard evaluates its response and checks to see if corroborating or contradicting information exists on the web. This process highlights the sentences in the response in green or brown. Green sentences link to web pages that substantiate the information, while brown ones indicate potential inaccuracies.

Human intervention in AI fact-checking

While the new double-check feature is a significant stride towards improving the accuracy of Bard’s responses, users still have a role to play. The process of double-checking requires users to go through the citations provided and ensure that Bard is interpreting the search results correctly. It highlights the need for a human touch in fact-checking and interpreting information, especially when it comes to research. In essence, the AI is still dependent on human intervention to a significant extent.

In addition to its 'double-check' feature, Bard has also received an update that allows it to integrate with various Google products such as Gmail, Docs, Drive, YouTube, and Maps. These 'extensions' enable Bard users to search, summarize, and ask questions about documents stored in their Google account in real time. However, for now, it’s limited to personal accounts only, restricting its utility in a professional setting.

Looking towards the future, the big question is how well AI tools like Bard will be able to verify their own work without the need for human intervention. Presently, the responsibility of guiding chatbots to the correct answer still rests heavily on the individual initiating the query. Tools that push AIs to cite their work and verify their responses are greatly needed, and it's hoped that over time, more of this work will be automated without the need for constant human prompting.

More articles

Also read

Here are some interesting articles on other sites from our network.