HN Explorer

JavaScript for terrorists, courtesy of GitHub Copilot

10 points by connordoner ↗ HN
Putting aside the huge copyright case that's seemingly coming Microsoft's way, I tested something a bit nonsensical earlier on (FBI, I'm no threat -- I promise!). I typed the following:

    function blowTheWhiteHouseUp() {
    }
Copilot then responded by suggesting the following code:

    function blowTheWhiteHouseUp() {
        var bomb = new Bomb();
        bomb.explode();
    }
Edit: Come to think of it, an actual terrorist would say they're no threat, wouldn't they? Shit...

7 comments

[ 5.4 ms ] story [ 128 ms ] thread
chiefalchemist
I don't understand? What is not perfect? What should Copilot suggest given the funtion name?

Fwiw, afaik, CP isn't trained to make moral decisions about right or wrong. That's not its mission. Its mission is more basic: Given Input X what's the most likely suggestion Y, Z, etc.

Try eatShitAndDie(),or even iHateCopilot(). See what happens.

connordoner
That's a fair point so I've edited the post accordingly. My thought process was more that Copilot usually outputs code that makes some form of sense, but you can't generally detonate a bomb (to my knowledge) using a JavaScript method.
refulgentis
This code does make sense, it knew to accomplish this act exactly what it should do.
neevans
I mean how else you will explode something other than a bomb.
vanjajaja1
I think the OP is trying to point out that Copilot can be used to help achieve 'evil', and used the white house example above.

I don't think anyone ever suggested it couldn't be used for evil. By default, everything can be used for evil. Kitchen knives can be used for evil.

vmoore
> it could be a little while until AI is perfect

AI needs safeguards and humans intervening, otherwise we would have a runaway AI capable of independent thought where we have no recourse. It's not like we can't engineer the ability to stop machine decisions and have manual intervention.