Re: [sl4] How much do languages matter for AGI programming?

From: Matt Mahoney (
Date: Fri Nov 28 2008 - 08:49:37 MST

--- On Fri, 11/28/08, Edward Miller <> wrote:

> MapReduce does sound like a good idea. It is open source,
> and this has allowed people to port it to Java. Thus, I
> wouldn't feel tied down to C++, or even Java. We need to
> find the best language. Would you want an AGI that is prone
> to buffer overflows? These sorts of problems can make the
> AGI very vulnerable to hacking, malfunction, or complete
> failure. I am assuming it would be best to implement the
> most rock-solid language(s) possible. If such a language
> doesn't exist to fit our precise requirements, perhaps
> creating a new language or modifying an existing language
> will be necessary.

I think this emphasis on programming languages is leading us astray from looking at the hard parts of the problem. Just because a program is written in Java instead of C++ doesn't make it more secure. Low level security vulnerabilities like buffer overflows are the easiest kind to fix. (For example, by setting the no-execute bit for the stack segment in newer x86 processors). The real security problem is intelligent worms, as I described in my proposal. When a worm can understand and use natural language, it can convince the user to do just about anything. (Please enter your administrative password to install 27 updates). Software is already too complex for people to know what their computers are doing. (Your computer is protected). People don't care if they give up part of their computing resources to botnets as long as it lets them get their work done.

As a practical matter, AGI has to have distributed ownership because it is too expensive for anyone to develop or own more than a tiny part of it. Developers are just going to use their favorite languages.

-- Matt Mahoney,

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:03 MDT