<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Mon, Dec 21, 2015 at 12:51 PM, Will Steinberg <span dir="ltr"><<a href="mailto:steinberg.will@gmail.com" target="_blank">steinberg.will@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><p dir="ltr">Yes, but in your case, Adrian, I assume you used a program to create new data that *you* then had insight from.<br></p></blockquote><div>Well, that's the trick. If I created the code, which of us - the code or I - is having the insight?<br><br></div><div>That said, the program did not create new data. The one (technically, a series of similar programs) I speak of performed a brute force analysis of more combinations of existing data than I could go through in a reasonable (relative to the task) time, and showed me the patterns that it found. As a result of using my creation, I possessed new insights about said existing data, regardless of whether it was the code or I that actually created those insights. Said insights proved to be correct, and I employed them in a way that resulted in a quantifiable benefit to me.<br><br>I then generalized this program and provided it to others who were encountering similar problems, albeit with slightly different sets of data, and they wound up with equivalent insights to my own (and equivalent benefits to what I reaped) without my specific involvement for each particular set of data. Did I create each one of those insights, without spending time (even one second) on each individual case? Did the other people create those insights, even though all they did was input data and run the program, ignorant to the logic and processes in the code? Or did the program generate those insights?<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><p dir="ltr"></p>
<p dir="ltr">The problem with the insight program is that it has to be able to, from data, make conclusions whose form of statement is not provided for in the original code. Does that make sense?<br></p></blockquote><div>Given the example you cite, this seems to be a nonsense request. <br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><p dir="ltr">An insight program, in theory, could be given a set of geological data and derive the laws of calculus.</p></blockquote><div>If it could derive said laws from geological data, it could derive said laws from no data. Calculus simply is not contained within observations of geology.<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><p dir="ltr">Or state the principal of natural selection in English.<br></p></blockquote><div>That could perhaps be done given sufficient biological data...and, of course, sufficient capability with the English language. However, then the insight is within the data.<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><p dir="ltr">It needs to go "outside the box", which is the whole thing that's confusing about human insight.</p></blockquote><div>Human insight does not work the way you have described either. Humans did not derive the laws of calculus from arbitrary geological data, and there does not appear to be reason to suspect that an AI should be able to either.<br></div></div></div></div>