Tools like Bing Chat, Windows Copilot, ChatGPT, Claude or Google Bard have seen a rise to prominence this year. These advanced chatbots promise to deliver information to users who chat with them. While you can’t ask them anything, as some content is locked down, you can get answers and information about lots of things.
Ask about the Mona Lisa or the Hallgrímskirkja and you get a good overview of these items, usually. You may get instructions on fixing PC issues or your car, and even medical advice is not out of the question.
There is always the chance of hallucination, which more or less refers to it returning content that is not true. Still, many tech companies are pushing AI like crazy. Microsoft, for example, added Bing Chat to Windows and several other company products.
Google Bard and Human Reviewers
Google confirmed on the Bard Help website that human reviewers may look at conversations. Feedback from Bard users plays an important role in improving Bard, but Google says that this is not enough. Human reviewers are “a necessary step of the model improvement process” according to the company.
The reviews, ratings and rewrites of human reviewers helps Google improve the quality of its generative machine-learning models”.
Google explains that conversations that human reviewers access are unlinked from Google accounts. Furthermore, random samples are picked for human review and “only a portion of all Bard conversations are reviewed”.
While that sounds reassuring, it is clear that input from human users of Bard may reveal their identity. Google recommends to users that they don’t reveal anything in conversations with Bard that they don’t want human reviewers to potentially have access to.
To Google’s credit, it highlights the fact that human reviewers may access conversations on the Bard website prominently.
What Human Reviewers do
Reviewers look for “low-quality, inaccurate, or harmful” Bard responses according to Google. Once identified, evaluators suggests higher-quality responses. These are then used to “create a batter dataset for generative machine-learning models”.
In other words, Google is using human reviewers to improve Bard’s responses to user queries.
How to prevent the sharing with reviewers
Google Bard users have just one option to prevent the sharing of their conversations with human reviewers. This requires disabling the Bard Activity. Here is a step-by-step guide on disabling Bard Activity:
- Open the Bard Activity website on Google’s My Activity hub.
- Activate the toggle to turn off Bard Activity on the page that opens. Note that you may also delete existing conversations while there.
Note that Bard activity won’t be saved to the Google account anymore. In other words, you can’t access conversations from one device on another when the feature is disabled.
The deletion doesn’t affect conversations that has been reviewed by human reviewers already. Google retains that data and related data for up to three years according to the privacy information on the Bard Help website.
Related information may include the language, device type and location info according to Google.
Closing Words
The advice to never include personal information that could be traced back to you is as old as the Internet. While this limits some conversations with AI, it is still sound advice.
Bard users who want to include personal information in their conversations may want to turn off Bard Activity first, as this prevents access for human reviewers.
Now You: do you use AI tools?