[conspire] More on robocalls (was: Terrible customer)

[Rick Moen]

----- Forwarded message from Rick Moen via Tech <tech at golug.org> -----

Date: Fri, 6 May 2016 17:28:27 -0700
From: Rick Moen via Tech <tech at golug.org>
To: tech at golug.org
Subject: Re: [GoLugTech] Terrible customer
Organization: If you lived here, you'd be $HOME already.

Quoting Patrick via Tech (tech at golug.org):

> OK, on my cell phone, that is registered as DO NOT CALL on the FCC
> website and registry, I received a call at 1340 May 6, 2016 asking
> how my computer is doing, with a very strong Hindi language accent.
[...]
> But, the FCC will get the report form filled out by me, anyway.

One cannot help noticing that all of these calls arrive via VoIP 
transport.  E.g., one on my cellular two days ago claimed a (bogus,
synthetic, defined by software at the calling end) caller-ID number of
1-888-888-8888.  So, here's one question I ponder in these matters: 
What can FCC do about it?

I'm intending that as a genuine rather than rhetorical question, but
cannot see offhand that FCC has the mandate to police VoIP or any tools
for either investigation or enforcement.  So, I suspect that all you've
achieved is adding another record to FCC's virtual filing cabinet.

Lately, I've been attempting to classify incoming automated robocalls by
category, out of sheer curiosity.  First top-level distinction is
between the very crude incoming calls that _don't_ bother to detect an
answering human, and those that do.

Classic example of the former type is the 'Rachel from Card Services'
call.  (There are several other names and voice actors.)  This automated
script launches immediately upon answer, unconditionally.  I'm
_guessing_ that these are not logging intelligence about the called
number, and they aren't mining any data from it (except perhaps the fact
of line pickup and datestamp) unless and until someone presses '1' 
to find out how to lower credit card interest rates, tra la.

The latter type, the one that does attempt to wait for, and detect and
note the existence of an answering human voice, is more interesting.
Again, I divide this into two [sub-]types:  Either there is a recorded
voice payload once an answering human has been detected, or there isn't.

The latter puzzled me for a while.  My (wrong) guess for about a year
was that it was autodialer information running in a defective mode where
the handoff to the function that normally would play the outgoing voice
message.  On reflection, and some Internet reading about robocalls, this 
seems unlikely.

Instead, I'm pretty sure that most if not all of those are very narrow
attempts to probe numbers to find live humans.  That is the sole purpose
of that call, and somewhere a computer is recording a database record
with date stamp, number called, and human yes or no.  Why?  Because
those are then the valuable numbers to concentrate on with a barrage of
_further_ robocalls that then have outgoing voice payload.

This is analogous to 'web beacons' in spamvertising mails.  Both voice
numbers and e-mail addresses are now exactly alike in that reachable
numbers / mailboxes are a dime a dozen but the real prize to be sought
is live human eyeballs / ears.

In the case of e-mail, what made the world change was the prominence of
free webmail providers.  Before that, you could have reasonable
semi-confidence that lack of Non-Delivery Report meant you reached a
human's current (or at least recent) address -- because people had mail
accounts because of paying for them.  Instances of mailboxes going
completely unattended happened, but were the exception rather than the
rule.  The dominance of GMail, Yahoo Mail, Fastmail and the rest means
there's no longer any perceived cost to leaving a mailbox valid after
you ceased using it -- and cheap storage means 'mailbox full' seldom
happens any more, either.  So, the world changed and deliverability of
mail tells you nothing about whether you've reached a person.

In the case of telephone numbers, the rise of first answering machines,
then centralised answering services, then dedicated fax lines and lines
otherwise computer-managed has had a similar effect, and the industry
abusing voice lines _desperately_ wants to know which currently are
being answered by humans.

I figure the key to the problem is to prevent that.  So:

My household has a new answering strategy, recommended by our friend
Duncan.  He picks up the telephone on incoming calls (impliedly, except
where it's a known number) and says nothing.  He waits.  If it's a
Rachel from Card Services -type robocall, the pitch starts immediately.
If it's the other major class of robocall (that attempts to detect an
answering human), there's dead air until the calling robodial machine
hangs up.
 
If it's a calling human, the caller seems to reliably say 'Hello?
Anyone there?' or such after a few seconds.  After all, it's common
enough for the answering person to have said 'Hello' before the handset 
properly initialised, or some other glitch prevented hearing the
answering greeting.  Human callers know this and instinctively
compensate.

To avoid even the chance of that sort of awkward moment, I've given
Duncan's tactic a further tweak.  I low-whistle some sort of tune 
(or almost-tune).  Frequently, it's 'She'll Be Coming Round the Mountain
When She Comes', your choice of either a sly joke about callers and/or a
'Close Encounters of the Third Kind' reference.  This _seems_ to work
even better than Duncan's complete silence tactic.

If you think about it, the Type 2 robodialers have a significant problem
to solve:  They must reliably detect humans answering, but must not
trigger on answers from, say, answering machines with recorded human
voices in the caller greeting.  This calls for somewhat fragile
heuristics.  I figure they require sounds in the human voice pitch range
at a minimum volume for at least a second, but must rule out detected
speech longer than about five seconds (as that's more likely an
answering machine).

This is why either Duncan's greet-with-silence or my
greet-with-very-low-whistling tactic seems likely to win (in the sense
of foiling live-human detection software), because it avoids triggering
robodialer heuristical checks.  My version has the further advantage 
that a calling human can recognise someone whistling 'She'll Be Coming
Round the Mountain When She Comes' as being an answering person, even
though the robodialers cannot.

All in my opinion and in my experience.  Other perspectives welcomed.

_______________________________________________
Tech mailing list
Tech at golug.org
http://lists.golug.org/listinfo/tech

----- End forwarded message -----