The goal of AGI is to be able to do everything a human can do, including modeling human behavior. But it is not bound to doing what it predicts a human would do, like we are. AGI should understand human emotions without having emotions itself.
Why that? Because that's how you automate human labor. We don't want machines to feel the same way about boring, difficult, or dangerous work as we do. On Mon, Oct 18, 2021, 9:39 PM <[email protected]> wrote: > If you don't give DALL-E the Text/Image Editor Motors and the Goal > Learner, it won't be AGI at all, it won't change its silly little dream it > spit out which has errors in it. > > On Monday, October 18, 2021, at 7:37 PM, Matt Mahoney wrote: > > As another example, nausea is not a general purpose negative signal, but > rather has the effect of reducing your appetite for foods you ate recently. > Evolution is not a general purpose learner either, which is why we don't > have wheels. > > If you ate some bad french fries and got sick or didn't like the taste, > your brain would not know the image of them before entering the mouth is > the cause or is the laptop you seen the cause, so it would associate ALL > those recent images to the pain signal. The only way to associate them more > is for you to repeat the image of the fries in your head, to artificially > target in on the fries and give them a bad reward, to change their > predicted probability for the future. So you need to make a bad context > make you stay focused on searching for the cause, and repeating it. So to > do this you need some rewarded memories like search and repeat X, linked to > other nodes. > > On Monday, October 18, 2021, at 7:37 PM, Matt Mahoney wrote: > > If it were, your desire to use drugs would not depend on having used them > previously. > > Over many and many and many uses, and constant repeating of the mission, > you can keep it strong that AGI is your goal, so that even after seeing a > book of an amount of words, you still say AGI after all that fashion > articles you read. IOW, it keeps you on topic of your job no matter the > topic, even if you were forced not to think about it for 10 days. This > isn't recency, it is reward, it makes things even more likely to be > predicted than "having seen a cat and hence predicting may see a cat again". > > Giving words in GPT reward would solve a myriad of problems, like > repeating X or learning new goals or keeping on the job even if shown 1 > million words on fashion design. > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/Tfda5d18b820b40ec-Ma86058589d3fd6bbb4e925a1> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tfda5d18b820b40ec-M6e269be6729510cd03cee769 Delivery options: https://agi.topicbox.com/groups/agi/subscription
