EFTA01790457.pdf

From: Joscha Bach a Sent: Sunday, October 27, 2013 1:48 AM To: Greg Borenstein Cc: Sebastian Seung; Joi Ito; takashi ikegami; An Gesher; Kevin Slavin; Martin Nowak; Jeffrey Epstein Subject: Re: MDF Am 24.10.2013 urn 02:56 schrieb Greg Borenstein > I think this issue of the changing definition of intelligence being a roving goal post is absolutely critical, Joscha. And it's one that =ong-predates 20th century digital computation-based Al efforts. > Recently, I've been reading the work of Jessica Riskin, a Stanford =istorian who studies the long history of Al and Artificial Life. =pecifically, Riskin's been writing about a strange phase in the history =f mechanical automatons that happened in the second half of the 18th =entury. Previously, automatons had always been built with their =echanism in one place (i.e. in a hidden box or platform) that then =rove their figures via a series of rods or connectors. The figures, the =epresentative part of the automaton were like the birds in a cuckoo =lock with no relation to the mechanism that made them move. > Then, suddenly, in the second half of the 18th century, a series of > =utomaton makers started to produce automatons that were built in a > way =hat was analogous to the thing they represented. (...) Yes! This represents a shift from a depiction to some kind of =essentialist" mechanist world view. This essentialist thinking is =till alive today, even if the notion of the essence has changed. Searle =elieves that thinking is impossible without the "intrinsic properties =f biological neurons", Penrose wants the intrinsic properties of =uantum systems, and a lot of people seem to believe that minds are =hemical or biological or environment-interaction phenomena. The common =enominator of these positions is that we need to recreate the actual =akeup to produce the property (intelligence, mind, cognition). At the moment, the most fuitful position is probably functionalism, i.e. =he idea that the property is the result of the underlying functionality =with respect to that property), and that there is no essence. If we =ould replace all biological neurons with little machines that perform =xactly the same functions and interact with the rest of the body in the =ame way, the resulting mind would be unchanged. Contemporary functionalism, however, is mostly still built on a the =oundation of mechanism, in the form of materialist physicalism. We are =sking ourselves: what kind of mechanism is the mind? What are the =echanics of neurons? Epistemologically, I think that matter and mechanics cannot be primary. =11 we have is information (discernible differences at our systemic =oundaries), and our ideas of matter, or causal mechanics, are encodings =ver these patterns of information. Information is primary, the universe =s a pattern generator, minds are a class of systems that (in very =articular and distinct ways) identifies and manipulates structure in =hat information. "Computation" means, simply put: meddling with =nformation. Minds, in this computationalist perspective, are certain =ypes of information processing systems. (Ones that implement functions =or integrating, representing, interpreting, combining, anticipating... =nformation, and that have a motivational system to set goals that =irect these functions.) Most computer scientists are computationalists by instinct: to us, =verything is a computer program in some sense. (Physics, for instance, =s the endeavor to find a possible implementation that could produce all =nown observable phenomena.) Most other people on the planet, including =uite a few philosophers, are not. To them, the idea of EFTA_R1_00118618 EFTA01790457 "reducing" =ind and universe to regular and stochastic changes in patterns of =nformation (aka computation) might even sound offensive. By the way, the first Al optimist was probably LaMettrie. His small, =itty and much maligned book "L'homme machine" (1747) is full of =odern insights, such as the continuum between humans and great apes, =he futility of the scholastic method (building on authority instead of =xperiment), the nonsense of dualism, the idea that machines need not be =hysical but can be mathematical, and so on. When he predicts that =aucanson's automatons herald the imminent arrival of machines that =ill actually speak and understand, he sounds almost like Ray Kurzweil =-) LaMettrie had relatively little impact in his time; his open atheism and =ejection of souls, vitalism, etc. made him a persona non grata. Perhaps not entirely unlike Minsky among continental philosophers of =ind... And the impact of the Lighthill report has probably a lot to do =ith the offense that people take at the notion of mind as machine. > This strikes me as very much like the process we go through with the =efining Al tasks like chess or Jeopardy or car driving. We start off "relieving that these are tasks that only human intelligence can achieve. =hen we build computational systems that can do them. Those systems are =ften inspired by the way humans achieve the tasks, but in the end work =n extremely non-human ways. Google's self-driving car uses massive =atellite data and laser scanning to drive, Deep Blue doesn't play chess =ike a human does. Applied Al has the big benefit that it does not offend anyone. The =pplications you mentioned are usually straightforward engineering, i.e. =hey do not even attempt to mimic human intelligence, but only look for =ays to solve the task at hand in the best possible way. If we want these applications to teach us something about the mind, we =eed to impose additional constraints. For instance, in robotic soccer, =e tended to brute-force problems with sensing the robot's environment =ith hardware (laser scanners, sonar, additional cameras, better CPUs). =he exception was the Aibo league, which had to make do with Sony's =og robots. With only a terrible, shaky camera, little memory and a slow =PU, the programmers had to come up with attentional processing, anytime =lgorithms that deliver better results over time, resource allocation =tc. I found that the most constrained robots enforced the most =nteresting solutions (of course, without getting much credit for it). Another problem with the typical applications is that they usually =eplace or reproduce human performance, i.e. they need to start out with =he abilities of a trained adult. Instead, we might want to look at =eproducing the path that human children take towards intelligence, the =utonomous process by which children learn to make sense of the world, =cquire a language, visual and conceptual grammars and so on. After =bout 3.5 years of swimming in massive flows of data, a child won't do =ell at Jeopardy, traffic navigation or chess yet, but can already watch =he first Star Wars movie and afterwards explain that Darth Vader =estroyed princess Leia's planet, and that he needs a laser sword. All =ognitive development afterwards is probably trivial ;-) I think this is =he kind of performance we should be looking for when we try to build =1. Cheers, Joscha <?xml version=.0" encoding=TF-8"?> <!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd"> <plist version=.0"> <dict> <key>date-last-viewed</key> <integer>0</integer> <key>date-receivedgkey> <integer>1382838458</integer> <key>flags</key> <integer>8623750161</integer> 2 EFTA_R1_00118619 EFTA01790458  <key>gmail-label-ids</key> <array> <Integer>22</integer> <integer>2</integer> </array> <key>remote-id</key> <string>355708</string> </dict> </plist> 3 EFTA_R1_00118620 EFTA01790459