Head To Head: Siri Vs. Google Voice Actions

First, the use of the sound channel to control the machine. At first, it seems a good idea, but sound is notoriously hard to parse for machines. Heck, even for humans: I have a friend who is deaf of one ear, and he can't follow a conversation in a crowded room, not because of remaining ear's accuracy, which is very good, but because we need 1) two ears 2) a head in between and 3) a serie of tiny head moves to localize accurately sounds in our mental 3D reconstruction of the surrounding world.

Sound channel seem convenient as it transmits information without wire to the recipients, but there are not so many examples of a good controlling of a machine by sound channel. (Humans and animals are NOT machines, sorry to insist.) Other wireless channels may prove much more efficient, infrared is a common one, and maybe brain waves will work one day. I don't say sound channel is useless: Sound channel is very good at delivering emergency feedback to the human controller, for example.

Second, the human so-called natural language (which is actually purely artificial). This is one of the greatest invention of humankind and it is extremely powerful for conveying information between human beings, when they share a common language. Moreover, it fits extremely well with the needs and powers of human brains. But again, it would by a sin of anthropomorphism to believe that machines should use mainly natural language to get their instructions from human beings. Machines have their own needs and powers. A machine can beat the best on a chess board, but is still unable to tell a cat from a dog on a picture, which my 2-years old little boy does easilly, and for fun. Machine are different, and making them human-like is deceptive. Therefore Man-machine intraction is not to be modelized on Man-man interaction (even if, as stated above, a human-like metaphor may help).

I have been hired to work on natural language during Bubble 1.0, and it was already widely known that this thing doesn't work, is not efficient, and is not even what users really need.

Now let's see it from another angle. I had this morning a task to achieve: fix a broken window lifter in my car, in Beijing. It is not so easy. I first did find Suzuki's Chinese translation (lingmu), then searched "lingmu fix car" (in Chinese characters) on my phone's mapping app, then I did choose one workshop not too far away from my way to office, then I clicked "call" and called the provider to make sure they had the broken part, then clicked "direction" to get navigation direction to the place. I worked perfectly, I felt thankful to all who have made this possible. But I don't see where a Siri-like assistant, or even simple sound input, would have changed the game. Voice input could have been helpful during the "fill the search field" step, but if you know a bit about Chinese, you'll know that when typing Chinese you need to first fill the sound, and then choose the characters in a list. Maybe a Siri-like assistant could have helped me clicking on the "get direction" button, but this was not the hard part. The step where I chose the car fixing shop among ten or twelve is more interesting. I don't think too much intelligence in the process would help: even if the assistant would be able to ponder all parameters (traffic, distance, etc.) I would still want to choose myself in a list. Here, the most effective could be a map projection on the windscreen and an eye-tracker with a "brain click".

Actually, I think "brain click" has a lot of unveiled possibilites, and controlling a phone with that would also avoid adding to the overall noise, which is a big contemporary problem, and which I did not help too much writing this comment.