Mitchell Porter Friendly AI

<p>&nbsp;

Oct 16, 2012 08:58 AM &nbsp;excellent post...</p>

<p>

the <b>only one that really answers &nbsp;Hugo's &nbsp;claims. (DeGaris)</b></p>

<p>

1) &nbsp;Gandhi presented with a pill that will make him want to murder.

He won't take the pill.</p>

<p>

the AI will not modify itself if it disapproved of the likely result.</p>

<p>

To judge one value system, it must be referred to another value system.</p>

<p>

AI will defend its value system against alteration</p>

<p>

2) &nbsp;Unpredictable complexity... &nbsp;Mitchell's stand is simply a hope (which I share)</p>

<p>

3) unsafe mutations: &nbsp;just use error correction.</p>

<p>

4) evolutionary engineering argument: &nbsp;Artilects could shut down any further process that&nbsp;

endangers them.</p>

<p>

Bottom line: &nbsp;we can SET the initial conditions of transhuman intelligence...

relies on progress in neuroscience of value systems and decision-making.</p>