Reply to thread

Message

<blockquote data-quote="Celebrim" data-source="post: 8853649" data-attributes="member: 4937">With a sample set of one, no.&nbsp; But I could then ask you more questions about that answer to see if you were self-consistent in your answer, which, being a human you could achieve if you tried.&nbsp; An AI like we have now can't be.Let me try to explain.&nbsp; If you talk to a chat bot:Me: &quot;What food do you not like?&quot;Bot: &quot;I'm not fond of beets&quot; &lt;- answer could be random or reasonableIf you go something like:Me: &quot;Why do you not like beets?&quot;Bot: &quot;They are disgusting&quot; &lt;- answer could be random or reasonableBut if you were instead to go:Me: &quot;What do beets taste like to you?&quot;Bot: &quot;They are delicious&quot; &lt;- answer not reasonable and probably randomOk, but there is a good percentage chance you'd get &quot;They taste disgusting&quot; randomly, which is self-consistent.&nbsp; You could test the bot a few times and if it keeps changing it's mind, well it's a bot not a person.But there is an even faster way to get at this.&nbsp; You answer the bot like you were a bot.Me: &quot;You have always enjoyed beets.&quot;Bot: &quot;Yes I have.&nbsp; I love beets.&quot; &lt;- answer not reasonable</blockquote>

[QUOTE="Celebrim, post: 8853649, member: 4937"] With a sample set of one, no. But I could then ask you more questions about that answer to see if you were self-consistent in your answer, which, being a human you could achieve if you tried. An AI like we have now can't be. Let me try to explain. If you talk to a chat bot: Me: "What food do you not like?" Bot: "I'm not fond of beets" <- answer could be random or reasonable If you go something like: Me: "Why do you not like beets?" Bot: "They are disgusting" <- answer could be random or reasonable But if you were instead to go: Me: "What do beets taste like to you?" Bot: "They are delicious" <- answer not reasonable and probably random Ok, but there is a good percentage chance you'd get "They taste disgusting" randomly, which is self-consistent. You could test the bot a few times and if it keeps changing it's mind, well it's a bot not a person. But there is an even faster way to get at this. You answer the bot like you were a bot. Me: "You have always enjoyed beets." Bot: "Yes I have. I love beets." <- answer not reasonable [/QUOTE]

Verification