Re: CAPTCHA over smtp (yet another spam solution to discuss)

R. Armiento wrote:

[Captchas are an instance of C/R email, and C/R email lends itself to 
the idea of posting financial bonds to prevent spam.]
>You make it sound as if we implement one, we can as well implement both. 
>Not so. CAPTACHAs can be implemented today just as a "protocol thing" 
>(e.g. as discussed here, an extension to SMTP).

I would argue for a general-purpose system.  The main technical 
difficulty lies in making the C/R part universally available.  Whether 
you instantiate the system with captchas or with a "money system"---both 
things are pieces of cake, as compared to the "let's reengineer the 
Internet" part.

>Money systems require an interoperability between worldwide financial 
>institutions which is not there yet.

Credit cards go a long way.  With the least sophisticated technology you 
could imagine, i.e., (16-digit) card identifiers.

>I believe the resistance to switch is lower for CAPTCHA systems than 
>for money systems. One argument is that apparently the resistance was 
>not too high for it to work for web forms.

The web, at least w.r.t. to forms, has virtually always been 
"challenge/response."  Mail is different.

Moreover, I think it's fair to claim that there have been web forms that 
collected your credit card number way before captchas became popular.

>Regarding kids: Kids will just have to learn how to solve CAPTCHAs, 
>similar as how they learn how to use other features of the MUA.

"Please identify the naked body part in this captcha."

Regardless of age, the usual caveats apply:  What about people with 
disabilities (like blind people)?

>Put more simply, I want "the solution to spam" to banish mass mailings 
>without explicit prior consent. White lists are a good way to give such 
>consent. I see no problem here.

So you wouldn't want to promote the exchange of information that is 
beneficial to both parties, the sender and the recipient?  In other 
words, you cannot imagine ever receiving a message that is useful to you 
although you had not whitelisted the sender?

[Example of how financial bonds work out]
>What a horrible mess. Lets say I am the target of a number of focused 
>advertising campaigns for e.g. mortages, so I have to raise my bond to 
>50¢ to deter them. But then I stop getting emails from my belowed 
>XXX-awreness foundation?!

The mortgage marketer would also have the financial resources to pay 
people to solve captchas.  The difference is, increasing the number of 
captchas to be solved (or their complexity) may result in legitimate 
senders spending minutes to get through to you (a single recipient).  If 
you make your attention more expensive, you receive the mortgage 
marketer's money.

[Honeypots]
>I enter my CC number on web pages a couple of times per year. However, I 
>will click 'yes' on the send email dialog 20 times per day. In fact, if 
>the $$$ is less than some well chosen max amount, my mailer will be 
>setup not to show these dialogs, or it would drive me crazy. So, no, 
>these "dialogs" are not the same.

You are really talking about contacting 20 people a day, none of who 
whitelisted you.  Your MUA would probably be setup such that the dialog 
would be skipped for (maybe a subet of) recipients in your address book.  
(These people might be friends who simply cannot figure out how to 
whitelist you.)

So, you are really left with the situation of contacting 20 /strangers/ 
a day.  But even then, the "well-chosen max amount" is likely to not 
exceed the equivalent cost of, say, a long-distance phone call.  Three 
things come to mind:

One, YMMV, but I am pretty sure most people would disagree with the idea 
that contacting 20 strangers via email should be possible (and even 
convenient) by pledging less money than contacting 20 strangers on the 
phone would actually cost (assuming "money" and "email" in the same 
sentence don't cause great discomfort to begin with).

Two, if people set up honeypots just to seize bonds /marketers/ would 
post---hooray.  Spammers (erm, marketers, really) would need to figure 
out how to better target their messages.  They're supposed to; your spam 
filter is not.

Three, you are only in the situation to configure your MUA to authorize 
bonds for you automatically /if/ requiring their presence becomes 
commonplace on the recipient's side of things.  In other words, your 
situation is this:  You would like to contact 20 strangers a day, AND 
the vast majority of email users have already found it to be worth their 
while to configure their "general-purpose C/R" email such that access to 
their inbox requires bonds posted rather than captchas solved.

>>>I have thought about [to what extent CAPTCHA puzzles should be 
>>>configurable]: but it seems dangerous to allow users to configure 
>>>this themselves.
>
>>In other words, you are assuming that you know better than email 
>>recipients how to protect them from spam?
>
>Yes, I do.

And you mentioned compatibility with current Internet philosophy.

>For certain users. With enough freedom for end-users to design their 
>own CAPTCHAs I can easily imagine having the following conversation 
>over and over and over:
[...]
>- Oh... stupid thing... it should have told me.

If already the special case of captcha C/R is hard to get right, we can 
as well get the general case of arbitrary C/R right.

>>>Do you mean you want the signatues to be *repudiable* or 
>>>*non-repudiable*?
>>
>>Repudiable. 
>
>I have always thought that the hard part is to make any system 
>non-repudiable. OpenPGP signatures are repudiable to some degree:
>* Erase your private key and denounce all emails you sent with that key
>* Claim that your private key was stolen by a third party.
>* Claim that you wrote the email with a gun to your head.
>* Claim that you accidentally left your terminal logged in, and someone 
>must have sent an email while you were out.

Repudiability usually implies that your private key remains valid when 
you repudiate your signature.  If you get broken into, you need to 
/revoke/ your key, which is a different story.

>A CAPTCHA system could *maybe* be designed for this to work by making it 
>indeterminable if a specific email got through based on solving a 
>CAPTCHA or invoking the whitelist.

Indeterminable for the recipient?  If something gets through to me from 
someone who isn't whitelisted, I can make that determination.

R. Armiento wrote:

>[Joachim Kupke:]
>>>>Anonymous electronic money exists.  Oh, it involves C/R, though.
>[R. Armiento:]
>>>Do you have a reference? I do not know of any existing solutions
>>>that let my purchase be anonymous both to the the buyer and all
>>>third parties (i.e. "the bank").
>[Martin Cosgrave:]
>>http://www.jya.com/nsamint.htm
>
>Nice report. Thank you for the link!

http://www.crypto.ethz.ch/research/display.html?label=CaMaSt97  came out 
the next year and addresses the following problem.

>The report brings up several cryptographic schemes for the kind of 
>anonymous money systems we were discussing. However, it also makes a 
>good case to why we will not see this up and running any time soon [see 
>Section 5.4 and Conclusions]. For one thing, law enforcement agencies 
>around the world will not allow a money system that is completely 
>untraceable.

The Camenisch-Maurer-Stadler system has already made it into textbooks 
on cryptography.


--Joachim