<html><head><meta http-equiv="content-type" content="text/html; charset=utf-8"></head><body dir="auto"><div dir="ltr"></div><div dir="ltr">Dear Connectionists, </div><div dir="ltr"><br></div><div dir="ltr">Richard really hits the nail on the head here. The problem with so many discussions of AI, particularly within the machine learning community, is that the perspective that is commonly taking is a <i>behaviorist</i> one, looking only at output, and largely ignoring matters of internal representations. You simply can’t tell whether a broken clock has a representation of time, by noticing that it is correct twice a day.</div><div dir="ltr"><br></div><div dir="ltr">In the 1950s, Cognitive psychologists figured out that behaviorism, — focusing largely or entirely on behavior while dismissing what was going on inside was a bad idea. Behaviorism was such a bad idea that the entire field of behaviorist psychology (once dominant throughout the United States) largely collapsed within a decade.</div><div dir="ltr"><br></div><div dir="ltr">There is an old saying: <i>you can save two hours in the library by spending six months in the lab </i>(painfully reinventing what was already known). By continuing to ignore the cognitive revolution, black box approaches to machine learning may ultimately turn out to have wasted year or even decades. </div><div dir="ltr"><br></div><div dir="ltr">Conversations about the “understanding” of machines that are essentially unreliable black boxes bound to increasingly opaque data sets, seem to me to be a waste of time.</div><div dir="ltr"><br></div><div dir="ltr">Gary </div><div dir="ltr"><br></div><div dir="ltr"><br></div><div dir="ltr"><br><blockquote type="cite">On Mar 16, 2023, at 07:47, Richard Loosemore <rloosemore@susaro.com> wrote:<br><br></blockquote></div><blockquote type="cite"><div dir="ltr"><span></span><br><span>Geoff,</span><br><span></span><br><span>Clever deflection. ;-) The learning disabled young adult you mentioned does NOT use a rote-memorized copy of the entire contents of the internet when he tries to answer the question "how many legs does the rear left side of a cat have?"</span><br><span></span><br><span>What matters is *both* the bad performance and what is going on inside, because that tells us that the cause of the failure is completely different in the two cases.</span><br><span></span><br><span>Richard</span><br><span></span><br><span>On 3/10/23 1:03 PM, Geoffrey Hinton wrote:</span><br><blockquote type="cite"><span>...</span><br></blockquote><blockquote type="cite"><span>Do you really want to use the fact that [a learning disabled young adult] misunderstood this question to say that he has no understanding at all?</span><br></blockquote><blockquote type="cite"><span>Are you really happy with using the fact that chatGPT sometimes misunderstands to claim that it never understands?</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><blockquote type="cite"><span>Geoff</span><br></blockquote><blockquote type="cite"><span></span><br></blockquote><span></span><br></div></blockquote></body></html>