Over the weekend, this tweet appeared:
First things first, Joanna Stern is one of the founders of The Verge and now writes a widely read tech column for the Wall Street Journal.
Here’s Joanna’s delighted response:
> That’s it, I finally love Siri. Apple finally did it.
Here’s a link to Joanna’s Wikipedia page. This turns out to be important since this is the page Siri quotes in the response to the question, “How old is Joanna Stern?”, as shown in the embedded tweet.
I went to the Wikipedia page and did not find any reference to a birthdate. Birthdates are common for most biographical Wikipedia pages. So how did Siri come up with an age as a response to this request?
And to make matters even more confusing, when I posed the same question to Siri, I get the same link, but this response:
> Joanna Stern is 32
Siri also helpfully provides a date of birth of December 5, 1985.
According to Joanna’s response on Twitter, both of these are wrong.
This anecdote made me think about the source of Siri’s information and how vulnerable Siri, Alexa, Cortana and Google-person are to misinformation. I’m also interested in how Siri came up with that birthdate, given the link was to a page with no embedded birthdate field.
I don’t find it amusing at all.
Another day, another reason why I don’t use Siri…
‟…
This anecdote made me think about the source of Siri’s information and how vulnerable Siri, Alexa, Cortana and Google-person are to misinformation. I’m also interested in how Siri came up with that birthdate, given the link was to a page with no embedded birthdate field.‟
Indeed!
The error is on Siri for making a strong claim on something that it has low certainty of. If you type, “How old is Joanna Stern” in Google, you get links. If you type, “How old is Donald Trump” you get the big box with the correct answer. The key is making sure you artificial intelligence is intelligent enough to know how certain it is about an answer and wait until it hits a high threshold.