Siri is a fairly good AI, but the fact that it stumbled on some basics is a source of disappointment for me. In this test, Alexa stumbled as well.
Siri is very cool. Apple has put a lot of work into it. Siri is useful to me on an ongoing basis. But…
An AI is designed to parse speech and do its best to respond. But when it stumbles on very basic concepts of speech, rich in context, it can be frustrating. For example, I asked Siri this question on my Apple Watch S4.
Just to be sure, I checked on my iPhone. Same answer. It’s almost like what the physicist Dr. Wolfgang Pauli (1900-1958) was fond of saying “That’s not even wrong.”
Of course, if one leaves off “in pounds,” Siri (iPhone) presents a list of some website links. You’ll have to select one and hope it provides the right answer in plain sight. As an aside, here’s the correct answer, right out of Google search, very simply. I double checked myself, and 8.34 pounds is close to the correct. 8.3454. (It wasn’t rounded up correctly.)
If one wants to drop back and let let Siri dictate the units of measure, then some progress is possible. I’m back to my AWS4/watchOS 5.0.1 here.
Why does Siri report “3.79” and Google/Quora “3.78”? It’s because a more precise answer is 3.78541. Again, Google/Quora didn’t round up correctly. Siri did.
Alexa Up to Bat
I asked Alexa the identical question. “How much does a gallon of water weigh in pounds?” Alexa was better at figuring out that I wanted an answer in pounds. The answer I got was:
“9.92 pounds or 4.5 kilograms.”
I got that same answer three times. And that, as Dr. Pauli would have said, is indeed wrong.
What I Learned
- Siri couldn’t pick up the context of a desired answer in pounds. Alexa could.
- When Siri was given a simpler verbal context, it picked its own units of measure. At least the answer was right.
- Little attention was paid to rounding rules here by some sources.
- Alexa picked up on the context of desired units but presented a very wrong answer. This is alarming. Amazon?
Of course, one could do a lot more testing, and I’m sure all kinds of interesting effects would be noticed. But my simple goal here was to report some very odd things that I noticed right away. And wonder why these AIs both stumbled on something so simple.