Automations and 3 Laws or 3LU movement

[viziel]

Three Laws Unsafe!

If you are interested in the philosophical discussions regarding Asimov's Three Robotic Laws this is a great site.



It poses allot of questions regarding how a robot brain may interpret seemingly sensible laws posed by humans. But without the upbringing, understanding and genuine intelligence a robot may see these things as different. There are assumed values in the terms that we as humans understand implicitly but a robot may not.

[Rob Lang]

Ah, my speciality (I have a PhD in Artificial Intelligence, don't-ya-know).

In Icar, the "three laws" aren't laws but philosophical statements. Asimov himself proved the laws were inherently broken in I,Robot. An artificial intelligence is not programmed, nor copied but created inside an environment where it gains its own personality and can be trained (using far more powerful AIs) to understand the philosophy of pacifism towards humanity. From the philosophy forms an unshakeable believe and that becomes core to its programming.

Like Turing machines, Asimov's laws were there to prove a point - a single important point. Sadly, they've been taken to all extremes that they were never designed to.

[viziel]

I found it a fun read.

What I find most interesting is how they state given a situation to which a robot DOSN'T have programming for, and therefore cannot resolve through the given laws, it will most likely fall back to whatever is considered the (IF THEN ELSE) logic. In the documents case the robot would just stand there staring into space.

However, they mention that a human given the same situation will tend to act chaotically in an attempt to resolve the problem and learn from the outcome.

This ability to deal with situations in intuition and something that we always assume humans are better than AI at.

An interesting prospect. I, Robot clearly shows how the laws can be exaggerated and a robot could prevent a human from leaving their bed for fear of ANYTHING harming them. And innocently they suggest that robots would never know greed and therefore never intentionally steal a purse.

[Rob Lang]

Most modern AI doesn't collapse into simple if-then-else statements - even at the limit cases. AI is mostly a series of probability distributions. Even if you force it into some horrible stable state, an external input (as simple as time) can push the algorithm just enough to get it moving again.

That's the problem with philosophers of AI - they've never made any. As we say to people who quote Turing machines all the time: Boil me an egg. Turing machine have no concept of time so can't boil you an egg. Not much of a universal computer is it?

[Groffa]

[Rob Lang]

Groffa, if you have any other questions, then just email/throw them my way. I'm more a Neural Networks sort of person rather than anything else but we did have to do a fair amount of computational theory.

[Groffa]