Balentine's Brave New World

Been plowing through Bruce Balentine's new Book, "It's Better to be a Good Machine than a Bad Person," and I must say that so far I am enjoying it.

The basic idea of the book is that it's about time that we gave up trying to have machines try to behave like human beings (and do an awful job at it -- bad persons) and started having machines tackle those problems that they can solve well (good machines), and in the process interact with human beings in their capacity as machines rather than pretend to be human beings.

So, as you can imagine, Balentine doesn't like it when a machine tries to act or sound like a human: saying that it's sorry, expressing gratitude, giving compliments, etc. For Balentine, such anthropomorphism not only adds little value to the interaction but in fact confuses and ultimately leads to frustration and disappointment when the the system does not live up to the intelligence the surface anthropomorphism implied.

I am completely sympathetic to the idea of building machines that are true to their identity. Yes: humans do not interact with machines the way they interact with other humans. My only concern is that spoken and heard language are so suffused with the human that expecting a human being to somehow find a way to use it and yet strip away the layers of emotional and cognitive meanings that are fully enmeshed in it is a difficult endeavor, to say the least.

Imagine being interrupted in mid-sentence by a machine: would you help not feeling irritated? Or how would you feel if the system were to order you around with "Give me your contact ID" or "Say that again." As things stand, I wouldn't like it and -- and this is the main point -- I wouldn't be able to help not liking it.

But Balentine feels that sooner or later, we will get to that brave new world where we wouldn't react emotionally when we are talking to a machine. We would know that this is the way to interact with a machine and we would turn our emotional sensors off.

I think what makes this tricky is the fact that the interaction is verbal in both ways. We have no problems shouting orders at machines in a way that we would not a human being; and we have no way accepting cryptic responses from machine when not spoken (ATM menus, Boarding passes). But as soon as we are engaged in a two-way verbal dialog (even when not spoken), we are overwhelmed with the anthropomorphic illusion.

But let's grant that someday we will somehow get ourselves to that point. The question is: how are we to cross the chasm from where we are today to that time when humans will talk to machines in that special human-to-machine way? Will one killer product (iPhone?) or application help us make the quantum leap? Or will we gradually evolve into such a new standard?

Will ramble on some more in future posts as I read and think about this....

Content vs. Form -- more on Balentine's book....

Continuing with my reading of Bruce Balentine's book. I am now in the thick of his argument against building "bad humans" and for building "good machines" -- this is articulated explicitly in essays #94-#97 (pp. 265-279).

I've expressed in a previous post my sympathy with the core idea of moving away from the paradigm of building machines that mimic humans and towards a more grounded paradigm where the focus is on helping people solve problems as efficiently as possible.

There are several tensions that have come up in this proposition, though, which I will tackle in the next few posts.

Here is the first: I am not comfortable with the neat dichotomy that Balentine seems to rely on between form and content. He seems to suggest that there is a clean way of slicing off "useless fluff" (my expression) and getting to the heart of the matter that is just as effective as language with "the fluff."

Example: should a system say "thanks" after you give it a piece of information it asked for? From what I can tell, that's a bit of the machine trying to sound human, as far as Balentine is concerned. But in my view, the "thanks" is not "just" an expression of gratitude (which, granted, is primarily a social transaction). It also serves the functional role of alerting you that the system has heard you and that it has accepted your input, and that the interaction was completed successfully.

How about this one: the system is retrieving information from a back-end and things are taking a bit longer than usual, and so it says, "Sorry for the delay. Looks like the system is slow today."

Is that too much anthropomorphism? Too much chit-chatting? I don't think so. It's very useful information: tells me that the system is still working on things and that it is indeed slower than usual. And you know what, I don't mind the fact that the system "realizes" or acknowledges that I don't like waiting. Not that I am fooled into thinking that the machine has feelings. And yet, I like the "sorry" message. It is useful.

More tensions tomorrow...