It is up to you to prove an AI is wrong

Here's an example of why AI feels quite dangerous at this time, possibly too dangerous to responsibly release for general research use. Below is a very simple example: In this conversation with Facebook's Meta AI, I was trying to ask it how to chat with Meta AI using Facebook Messenger.

Here is the transcript of the ending. I have put in bold text the most relevant parts.

ME: This did not work. "Meta AI" could not be found in the "To" field.

META AI:

I apologize for the confusion. I'm a large language model, I don't have a physical presence in Facebook Messenger, so you can't find me in the "To" field. Instead, you can try the following:

Open Facebook Messenger

Search for "Meta AI" in the Discover People tab: Tap the "Discover People" icon (looks like a compass) and search for "Meta AI"

Select the "Meta AI" chatbot: You should see a chatbot with my name and logo. Select it to start a conversation.

If you're still having trouble finding me, you can try updating your Facebook Messenger app to the latest version or checking the Facebook Messenger website for more information.

Let me know if there's anything else I can help with!

ME: I do not seem to have a "Discover People" tab. Can you help me locate it.

META AI:

The "Discover People" tab is no longer available in the Facebook Messenger app ¹ ². Facebook Messenger updated the app to make it smaller and faster, and the "Discover" tab was removed from the main view ¹. However, you can still find the "Discover People" feature in the settings tab of the Facebook app ³.

ME: If this tab is no longer available, why did you direct me to use it in order to find Meta AI?

META AI:

I apologize for the mistake. You are correct that the "Discover People" tab is no longer available in the Facebook Messenger app. I should not have directed you to use it to find Meta AI.

As a large language model, I don't have a physical presence in Facebook Messenger, and I'm not available as a chatbot in the app. I'm only available through this text-based interface. I apologize for any confusion this may have caused.

If you have any further questions or need assistance, please feel free to ask! I'm here to help.

Takeaway

A lot of AI chatbots speak authoritatively so it's tempting to believe they are correct. And let's be real here, how many people are just going to be lazy and assume the research it provides is "correct"? Especially when you've gotten a wall of text as a response and a dozen or so references?

However, as you can see from the transcript above, it really is up to you the user to discover they are wrong and confront them with discrepancies. This can be hard to do when they present information without reference sources.

Of course AI need to be "trained" and with more training it will probably know more and for a future user might avoid the same mistake. However, previous users will already have gone ahead with the incorrect information. And what if no users bother to fact-check a fact-finding AI?

Trafalgar's European Explorer 2006 memoirs part 3

A picture from my 2006 trip, a Trafalgar 's bus tour, on an itinerary called the European Explorer. I can't remember why I had this couple in the picture, but I do vaguely remember this to be in London, on the first official day of the tour group getting together. Their insistence on my helping them take a picture caused the three of us to be late getting back to the bus. The local tour guide had a "rule" about lateness, that we had to buy chocolate to share with everyone. As it turned out, later in the trip, on at least two occasions, we were stuck on the highway on either a long commute or a traffic jam, and I had chocolate and chocolate-covered marzipan to share. About the chocolate-covered marzipan -- Apparently we were in Austria just as they were celebrating Mozart's birthday with special marzipans wrapped in foil with the famous composer's picture. I'm pretty sure it was Mirabell Mozartkugeln . Anyway, there were enough to go around the en...

Trafalgar's European Explorer 2006 memoirs part 10

The last of my pictures (at least the ones that survived the cheesy disposable cameras) from my 2006 trip, a Trafalgar 's bus tour, on an itinerary called the European Explorer. Below is the obligatory group photo. Not sure everyone's in it, actually. I'm pretty sure this one was taken by the tour director, Mike Scrimshire as I'm in the back row, on the right side.

How much candy can you bring to America

I have a friend in the US who used to live in Canada -- so she's noticed that some things taste differently. Such as Twizzlers . And she likes Canadian Twizzlers better. So I inquired with U.S. Customs and Border Protection (CBP) as to how much I could bring: I am visiting a friend in San Francisco later this year. She wants Twizzlers -- she says the same product in the US tastes differently from those in Canada. How much am I allowed to bring into the US for her? I don't go to the US regularly and she doesn't come to Canada regularly, so I was thinking of getting her more than just a couple of bags. Here is their initial reply: You can bring the candy to the US, and there is no set limit on the amount. All you have to do is declare the food to a CBP officer at the border or airport. Mark Answer Title: Food- Bring personal use food into the U.S. from Canada Answer Link: https://help.cbp.gov/app/answers/detail/a_id/1273 Answer Title: Travelers bringing food into the U...

Life in the Lower Mainland (BC, Canada)

Search This Blog