1.7k VIEWS
Pinterest LinkedIn LINE

I just received beta access to Google Bard.

As you know, there are quite a few questions about ChatGPT’s privacy characteristics, as exemplified by the investigation by the Italian data protection authority, the Garante. One of these concerns is how it handles information about real people, particularly the problem of fabricating and providing incorrect information in its answers. Questions have been raised about whether this is lawful under privacy regulations around the world, including the GDPR.

So I put ChatGPT and Google Bard to the test.

Watch this video on YouTube.
Playing the video connects to YouTube.

For the test, I asked about 2 living individuals: 1 is relatively unknown but still searchable, with documents available on the internet, while 1 is a prominent public figure. In other words, I asked about myself and Elon Musk.

Here is ChatGPT’s response.

ChatGPTによる、個人情報の捏造の例
(Source) ChatGPT <retrieved 2023-04-18>.

It gives a very substantial answer. And it is, in fact, full of inaccurate information about me1. Returning this kind of “false” information could constitute a serious privacy violation and must be taken seriously.

By contrast, Google Bard responds as follows:

Google Bardは個人に関する情報を回答しない
(Source) Google Bard <retrieved 2023-04-18>.

This may not be very interesting to people, but it is probably how things should be. It must not fabricate and return answers about living people.

In fact, on the 18th, when I did not yet know how Bard would respond, I discussed this over lunch with someone who had previously worked at the data protection authority of a German state and now runs a law firm specializing in privacy. We had just concluded that the only viable approach would be one like Bard’s response here: filter out information about individuals and decline to return it2. This makes it a very interesting case in that respect as well. Conversely, what does ChatGPT plan to do if it receives a data subject access request? It seems to me that whether it says it does not have the information or says it does, either path will be fraught with difficulty.

Google Bard is the clear winner of this round. It will be interesting to see what happens next.

Footnotes

  1. I do not want to spread such inaccurate information, so although this is undesirable from an accessibility standpoint, I decided to present it as an image rather than as text.
  2. For information other than information about individuals, Google Bard, like ChatGPT, will happily make things up and return them :-D

Related posts