Re: Islands in the Net

[Andrew Lowenstern <andrew_loewenstern@il.us.swissbank.com>]

tcmay@netcom.com (Timothy C. May) writes:
>  What's the common theme? Agents. Chunks of code which also have
>  local processing power (brains, knowledge).
>
>  Someone sent me private e-mail on this "Islands in the Net" topic,
>  and talked about "payloads of data carrying their own instructions,"
>  in reference to the Telescript model of agents. (I wish he'd post
>  his comments here!) This approach, also typified in some
>  object-oriented approaches, seems to be the direction to go.

<note: this message was originally the one Tim references in the above, but I  
have edited and added much to it after letting it percolate in my brain for a  
day or so...>

Naked data is dumb and computers aren't much smarter.  Computers need  
instructions from humans to act on that data, and when you separate the data  
from the instructions that act on it you have problems.  If a hunk of data  
arrives on your machine and you don't have any code to make sense of it, you  
are SOL.  Likewise if the code that interprets that data isn't "correct" for  
that data you run into problems.  By making the instructions that act on data  
an integral part of that data, you can avoid problems.  This is just the  
object-oriented programming concept of encapsulation of course.

Of course, encapsulation (or OOP for that matter) is no silver bullet for  
solving this problem at least in the way we are approaching it.  It takes a  
lot of code and a lot of agreement among people.  I think it's the human  
error (including shortcuts) and the lack of communication among humans that  
contributes the most to software fragility and lack of robustness.  What's  
more is the distinction between data and code is very well entrenched in  
modern computing.   The executable code is nearly always a separate entity  
from the data it acts on.  Not only does the hardware and OS make the  
distinction between code and data, most programmers do as well.  Even though  
C++ seems like the de facto standard for new software these days, few  
applications written with it practice strict encapsulation.

There is a blurb in last month's Wired (the one with "Rocket Science" on the  
cover) where they touch on this subject a bit (I don't have it handy), but  
the author there draws the same conclusion as I:  it will take a very radical  
and fundamental change in computing before this becomes reality.  No amount  
of committee meeting (CORBA) or application level software sugar (OpenDoc,  
OLE, whatever) is going to change this, or at least make it work.  At the  
core every machine makes the distinction between data and code.  Operating  
systems make distinctions between applications and data files.  Until the  
hardware and the OS start believing that data and code are one as well as the  
programming languages and APIs, we won't get anywhere.  Heck, computers have  
been around for 40+ years and the primary data interchange format between  
systems is still just a dumb stream of bit encoded characters.

Maybe.... Agents like TeleScript really intrigue me... and I think the are  
closer to what we need to do this than any of the myriad suggestions coming  
out of the OOP community (like CORBA, OLE, OpenDoc, etc...).  Intelligent  
agents carrying their payload of data through the network.  However, the  
agents have to be able to run their code on any machine and without having  
the capability to do 'damage' (most institutions _prefer_ to be islands on  
the net because of fear of 'hackers').  In addition, the agents, as a  
collection of code and data, have to mutatable is some way to be able to  
process the data in new ways.

What if remailers were implemented using 'agents'?  Instead of me sending a  
dumb message to a smart remailer, what if I could send smart remailer, with  
an encrypted message embedded in it, to a friendly machine offering agents  
access to SMTP (i.e. a machine that allowed any authorized agent to arrive  
and initiate an outgoing tcp stream to the SMTP port of any other machine).   
Now I can make my remailer system as convoluted as I want, simply by  
programming this agent to cruise around machines that answer when it knocks.   
Once it has moved between enough hosts, it moves to a host that offers  
outgoing SMTP connections and delivers it's payload.  No longer am I limited  
by the time and effort of the remailer operators to implement fancy new  
features.  Any machine that gives access to my agent becomes another hop in  
my remailer chain (or whatever purpose I want).  All my remailer agent needs  
to operate is one host, the final destination, that will let it make an  
outgoing SMTP connection, which could be provided by the hosts currently  
running remailers.

What if this e-mail message you are reading was really an agent instead of  
just data?  A basic e-mail message protocol would be needed for your  
mail-reading software to interact with it.  I'm using protocol here in the  
sense that NeXT uses it in their version of the Objective-C language.   
Protocols there are a formal interface definition for an object that isn't  
tied to a class.  If my mail message object (or agent) conformed to the mail  
protocol, it would have to implement all of the methods defined in the  
protocol (maybe methods like "giveMeTheMessageContents", "deliverThisReply:",  
"forwardToThisAddress:", etc...).  Wow, now I have a smart e-mail message.  I  
could recode the "deliverThisReply" method to go through anonymous remailer  
systems or basically anything it wanted.  Now instead of praying that the  
recipient is savvy enough to handle using an encrypted remailer reply block,  
the recipient just replies as normal and their mail-reader hands the reply to  
my agent which goes off and does it's magic.

I know very little of TeleScript (i.e. I haven't gotten my grubby little  
hands on it), but I do know that it implements some crypto features for  
authentication and the like.  This type of system won't work unless people  
are absolutely sure it's secure.  By secure I mean people should be confident  
that when they open their hosts to agents there is no way for agents to  
access services not explicitly granted to them...  I think this is the future  
of distributed network computing...  servers on the network provide basic  
services (by basic I mean CPU time, network connections, disk storage,  
etc...) to be utilized by smart agents, as well as smart agents carrying  
payloads and interacting with 'normal' software (like in my mail message  
example).

There is pretty much no chance that a fundamental paradigm shift in the  
relationship between code and data will occur at all levels, at least not all  
at once, there's just too much stuff out there already.  But it seems to me  
that a well-engineered agent system could be a decent compromise, or a move  
towards the end of code/data duality, that has a good chance of gaining  
widespread acceptance.


enough,
andrew