EFTA02509326.pdf

From: Joscha Bach < Sent: Saturday, March 10, 2018 1:52 AM To: Jeffrey Epstein Subject: Re: Attachments: signature.asc Hmm. I'd say that there is a multidimensional space in which =nderstanding is projected. Understanding is the creation of a mapping =etween the features of a domain and a function that I already know how =o compute, so I can simulate the domain. Shallow understanding involves =apping of a particular feature configuration, deeper understanding =xplores the latent variables of the feature set. There is usually more =han one way of creating such a mapping, and when we have found several, =e can also identify relationships between the mappings. Category theory =ystematizes that. When you describe your understandings (such as the path of light through =pace), it seems to me that your perspective is descriptive, i.e. you =ook at the emergent pattern that is generated by your understanding, =ithout looking at the structure of the generator itself. I try to =nderstand that generator, i.e. how to create a structure that can =roduce the desired pattern. This constructive perspective is what =omputationalism is all about. Btw, Google has just announced that they think they might be getting =loser to quantum supremacy: =ttps://www.technologyreview.com/s/610274/google-thinks-its-close-to-quant=m-supremacy-heres-what-that-really- means/ If they ever get there, I will be forced to revise a part of my current =reliminary model of how the world works, which would be very exciting. =t would probably mean that digital physics must be wrong, and finite =utomaton computationalism must be only treated as a theory about models =uilt on constructionist formal languages, and I might get converted to =cott Aaronson's views. > On Mar 9, 2018, at 12:08, jeffrey E. <[email protected]> wrote: > understanding is a multi dimensional space the language is a =rojection in that space. or an arrow in category theory. the =ocal point has history . so like the play appears different from =very seat in the theatre the integaration over each point projects =is understanding on the language. > On Fri, Mar 9, 2018 at 5:33 PM, Joscha Bach < > =rote: > What do you think of as space/field effects? The universe or learning? > Btw., did you ever come across Schmidhuber's idea of a Goedel Machine? > On Mar 9, 2018, at 05:39, jeffrey E. <[email protected]> wrote: >> » I would think of it more of a space / field effects , Not =ecursive algorithm s >> > On Fri, Mar 9, 2018 at 6:06 AM Joscha Bach < > =rote: EFTA_R1_01640591 EFTA02509326 » Last week I got to know Steve Hyman, Daniel Kahneman and Bob > =orvitz. Telefonica invited all of us to a two day workshop with Pablo =odriguez, Ken Morse and a few others, where we were meant to advise =hem on how to use Al for health applications. I told them that I think =he goal of therapeutic invention is not to increase happiness, but =ntegrity. Happiness is merely an indicator, not the benchmark. Current =pps tend to subvert the motivation of people, but I don't think that =his is necessary or the best strategy. Humans are meant to be =rogrammable, not subverted. They perceive their programming as "higher =urpose". If we can come from the top, supporting purpose, instead of =rom the bottom, subverting attention, we might be more successful. =Downside might be that we create cults.) Of the bunch, Hyman managed to be the most interesting (Kahneman was =ery charismatic but mostly tried to see if he could identify an =pplication for his system one/system two theory). Gary Marcus was =here, too, but annoyed everyone by being too insecure to deal with his =ncompetence. >> » Did I tell you that I discovered that Deep Learning might be best =nderstood as Second order AI? >> » First order Al was the classical Al that was started by Marvin =insky in the 1950ies, and it worked by figuring out how we (or an =bstract system) can perform a task that requires intelligence, and then =mplementing that algorithm directly. It yielded most of the progress we =aw until recently: chess programs, data bases, language parsers etc. > Second order Al does not implement the functionality directly, but =e write the algorithms that figure out the functionality by themselves. =econd order Al is automated function approximation. Learning has =xisted for a long time in Al of course, but Deep Learning means =ompositional function approximation. > Our current approximator paradigm is mostly the neural network, i.e. =hained normalized weighted sums of real values that we adapt by =hanging the weights with stochastic gradient descent, using the chain =ule. This works well for linear algebra and the fat end of compact =olynomials, but it does not work well for conditional loops, recursion =nd many other constructs that we might want to learn. Ultimately, we =ant to learn any kind of algorithm that runs efficiently on the =vailable hardware. » Neural network learning is very slow. The different learning =lgorithms are quite similar in the amount of structure they can squeeze =ut of the same training data, but they need far more passes over the =ata than our nervous system. > The solution might be meta learning: we write algorithms that learn =ow to create learning algorithms. Evolution is meta learning. Meta =earning is going to be third order Al and perhaps trigger a similar =ave as deep learning. >> » I intend to visit NYC for a workshop at NYU on the weekend of the =6th. >> > We just moved into a new apartment; the previous one had only two =edrooms and this one has three, so I can have a study. It seems that we =re as lucky with the new landlords as with the previous ones. >> » Bests, and thank you for everything! >> > Joscha >> >> >> > > > On Mar 8, 2018, at 16:37, jeffrey E. <[email protected]> =rote: >> > > > > progress? >> > >>> >>> please note > > The information contained in this communication is confidential, > > may be attorney-client privileged, may constitute inside > » information, and is intended only for the use of the addressee. It > > is the property of JEE Unauthorized use, disclosure or copying of > > this communication or any part thereof is strictly prohibited and > > > may be unlawful. If you have received this communication in error, 2 EFTA_R1_01640592 EFTA02509327 > > please notify us immediately by return e-mail or by e-mail to > > [email protected], and destroy this communication and all > » copies thereof, including all attachments. copyright -all rights > » reserved >> > > -- » please note > The information contained in this communication is confidential, may > be attorney-client privileged, may constitute inside information, > and is intended only for the use of the addressee. It is the > property of JEE Unauthorized use, disclosure or copying of this » communication or any part thereof is strictly prohibited and may be » unlawful. If you have received this communication in error, please » notify us immediately by return e-mail or by e-mail to > [email protected], and destroy this communication and all copies > > thereof, including all attachments. copyright -all rights reserved > -- > please note > The information contained in this communication is confidential, may > be attorney-client privileged, may constitute inside information, and > is intended only for the use of the addressee. It is the property of > JEE Unauthorized use, disclosure or copying of this communication or > any part thereof is strictly prohibited and may be unlawful. If you > have received this communication in error, please notify us > immediately by return e-mail or by e-mail to [email protected], > and destroy this communication and all copies thereof, including all > attachments. copyright -all rights reserved 3 EFTA_R1_01640593 EFTA02509328