<html>
<head>
<meta content="text/html; charset=windows-1252"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
On 2016-04-06 12:04, Dan TheBookMan wrote:<br>
<blockquote
cite="mid:8D5B19D5-0A2F-4CEF-80B4-08A5A0A8908B@gmail.com"
type="cite">
<meta http-equiv="Content-Type" content="text/html;
charset=windows-1252">
<div>Was this prediction based on anything more than gut feel?</div>
</blockquote>
<br>
Those predictions were all gut feel. <br>
<br>
It is well known (among forecasting experts) that experts on a
domain rarely make good forecasts about the domain - it is a
different kind of expertise. <br>
<br>
<blockquote
cite="mid:8D5B19D5-0A2F-4CEF-80B4-08A5A0A8908B@gmail.com"
type="cite">
<div id="AppleMailSignature">Also, isn't progress often nonlinear
and perhaps many of the predictions are linear? My guess is in
this field more rapid advances are happening now where past
progress was more fits and starts.
</div>
</blockquote>
<br>
Fits and starts are less predictable than exponential growth,
although people of course do tend to underestimate that because of
linear predictions (or the availability heuristic). <br>
<br>
The big problem IMHO is the lack of ways of measuring progress in
most AI domains. We have nice results for speech recognition,
character recognition and image captioning, but far less good
measurements on actual smarts. <br>
<br>
Shane Legg's AIQ ( <a class="moz-txt-link-freetext" href="http://www.vetta.org/2011/11/aiq/">http://www.vetta.org/2011/11/aiq/</a> ) seems to be
fairly hard to apply, even if it is general (and the only serious
application so far of the BrainFuck language!). <br>
<br>
<pre class="moz-signature" cols="72">--
Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University</pre>
</body>
</html>