Times and times again, here we are for the new exciting yet challenging semester. This time I am taking AI. I set my mind up for the challenge, to redress some of the problem I had encountered last semester.
1. Sleeping in class, lazy off class
In algo class last semester( I did not blog about it much ), my usual routine is snoring half way through the lecture ( maybe thats why there is no writing.. I have no recollection. I am in ception ) and boy it was embarrassed : “Asian Sam is looking at you, Duc”
This time around, I am planning to fix it. What I found works is, I would found a nice place close to the class and take a nap before the class begins so I would come to class feeling less drowsy. You could not really blame it on me tho : algo class is ~ 3 hours from 6 pm. AI is 1h20, more manageable plus the professor seems to be not that excel in speaking english( I don’t think he is a native speaker ) and for that fact I spent most of the time trying to figure out what he is trying to say( I am no native speaker either so I have this kind of trouble. He apts to talk really fast. Maybe he’s French ) — I got slides so that’s not too bad. kudos to him for teasing the snapshot crowd: “Don’t take pictures. It’s a waste of memory. Go on the course website to download the slides”( Like I did. I make my $ worth )
Is like no requirements at all : as long as you know how to code one of the majors programming language you would do well. Most of the projects/homework requires one to implement algo from scratch and no library or framework are allowed to use. Editing code and compiling will make use of the online platform www.vocareum.com . So you submit code there and they will test your codes ? This is remained to be seen as currently the setup is in progress.
Also, it always amazes that professor somehow has managed to formularize grading. All classes I have taken thus far have some form of formula for student to turn in, calculate grades. Maybe it is mathematics tradition: here’s the crank, turn it, you will know your grade: oracle…..sort of ( since number could be regarded as some form of world building, model-ish creating space where it works its magic in ). The formula is as follows :
grade = points * ( 1 – n * 0.2 ), n = #days late
That is, you lose %20 per day late. In light of algo, let’s look at the function growth rate. This formula scales the points by day that is whether you submit 1 second late or 23:59 hour late it is the same and don’t bother submitting 5 days late because you got 0 anyway. Wuff. seems hard. I’d better submit early and the code should work perfectly on all cases. I don’t see this class offering curve…it’s easy ? It does not seem so.
This reminds me of my old undergrad prof. he’s ambitious enough to devise a curve of his name to try to fix student class curve fairness:
And here’s the incentive : Student code will get run against one another and the winner get a chance to have an interview with Google. My god. dream job. But with the class of this size where in bright students cockblocking everywhere I don’t think I stand a chance. Not even close. Still. Nobody is taxing your dream. Dream on.
3. So Why AI
( I found that most of the link attached to the slides are dead. He probably points it to his local drive. )
Many applications. Here are some fun ones:
Auto-playing Mario. It seems to have some sort of collision detection rays to all objects moving to guess which action to respond to.
ASIMO. I remember when I was still in Vietnam, Honda brought the dude here for showing. He could walk around, gesticulate friendly greeting but pretty much the selling point is navigating.
Siri meets Siri. Feeding Siri into itself. Look silly. Somewhat relevant is what Google has done with Electric Sheep Neural Network[ feeding input back into itself as a process of perpetual learning]( I still have trouble understand it so Im just gonna leave a link here to review later ). LINK .Also Do Android dream of Electric Sheep is a good novel to check out as people recommend.
DARPA robotic challenge.
2015 Winning team : KAIST from Korea. It has wheel on knee as to easily maneuver against terrain as well as retina scan(?) to detect objects….
Self Drive car. I believe I heard somewhere Sebastian Thrun was involved somehow in a forerunner of Google self driving car Governmental project wherein the competition requires competitor to self drive from LA to Vegas.
It even permeates to the IoT, wearable devices. Imagine a personal assistant like J.A.R.V.I.S. Does Google Glass have that potential ? But still “how do you draw black?” Take a look at Google Glass Project.
With all those cool applications, one cannot help but wonder what is this that makes of AI or …
4. What is AI ?
A is for Artificial or Fake or … manmade. I is for Intelligence, one’s ability to pursue and learn, reason, deduce about many complex ideas, that is the characteristic that us human possessed or at least on average on the same level( of intelligence compared to animal or others — hmm, does plant think ? ) : I think therefore I am, right ?
Combined it together, AI is something that is inhuman ( possibly made by human ) having an ability to think like human and therefore act like one. That says under some circumstances can we figure out the one we are talking to is human or not ?( I don’t know….what if human body is possessed by Aliens ???? ). Alan Turing was pondering the same questions before us >>50 years ago ( maybe after the Enigma game seeing what machine could do and thinking someday it has enough power to fake being human ):
Alan Turing devised a test called Turing test as a way for one to tell if computer behave human like. Turing approach is classified as “Thinking human” approach among other approaches : “acting humanly“, “thinking rationally” and “acting rationally“. Our class main focus is on the last approach which is defined in the book by Peter Norvig , AI, modern approach, 3rd ed( teacher said 2nd ed is fine [ for those like me having no money for textbook] ). The guiding principle is based on the idea that rationality be well defined mathematically and thus provable and general enough( Turing digresses on being general; instead focusing on the topic the evaluator would like to ask ???). Therefore it should be easy to create agents for each specific tasks(components) to later combine them to create a human-like AI machine ?
Nowadays, to pass Turing test, an AI agent general needs to have the following knowledge:
+ Natural Language Processing : to communicate with the examiner. ASIMO (above) could understand language to some extents ( even though it still got trouble when ppl feeding sentences to it in unison ).
+ Knowledge Representation : a medium faster store/retrieve info. How does Watson store it knowledge ? A database ? Throwing link out there to read later.
+ Automated Reasoning : To infer new results, come up with new conclusions.
+ Machine Learning: pattern recognition, adapt to new environment.
+ Vision : to view the surrounding, include examiner(subject) and his presenting objects.
+ Motor control : to maneuver the environment, act on requested objects.
+ Other senses : audition, smell, touch, etcs.
>> Intelligent Agent = Sensors( to perceive environment ) + Effectors( to act upon the environment ) => maximize progress toward goals
Let’s take a look at Wall-E as an example for agent:
See Wall-E in actions ( Note that it does have human emotion breathed-in by animators of Pixar ):
It could be summarized by the Acronym : P.A.G.E(Percepts, Actions, Goal, Environment ). Agent is designed for specific task & well-defined : known goal & environment, a tool to analyze system, not new programming or hardware( even though some requires specialized hardware and new programming approach or methodology such as deep learning for instance )
4*. State of the Art AI. (EDIT : 08/27/2016)
Textbook Q 1.14 asks student to peruse literature on whether AI achieve the following( I think it’s cool so I added this mini section . 4*. 4-addendum ):
a. Playing a decent game of table tennis? NO. Could hit and swing but not at human lv.
b. Driving in Cairo, Egypt? NO. Seem to be hectic
c. Driving in Victor Ville, California ? YES. Not that crowded, only testing on urban setting( sign, lights, etc, not during traffic with human subjects ).
d. Buying a week’s worth of groceries at the market? NO. Lacking perception( could do OCR on price and item recognition but how to know to check out for e.g ) & adaptability( stuck in lines or avoid getting cheated ). LINK
e. Buying a week’s worth of groceries on the Web? YES. Inputs( online data/ markets ) are in favorite format( digital ) and easy to interact environment.
f. Playing a decent game of bridge at a competitive level ? YES. Card games have less possibilities than Chess or Go who computer has been really good at recently( Deep Blue and Alpha Go ).
g. Discovering and proving new mathematical theorems ? –. I know it could prove theorem automatically but don’t know about discover new theorem( with the rise of deep learning and neural net–actually giggle at this :“Is there anything risible? Biggus Dickus^^”). According to this, computer already did.
h. Writing an intentionally funny story ? NO. Generating sentence is a start . Increasing input data( lack… ), output are still not human-like writing ( unpublishable ), not to mention maintaining genre like funny: “…since the machine can not directly experience the emotions you’re trying to convey. It’s difficult to fathom how you could generate genuine creative writing in that sense”
i. Giving competent legal advice in a specialized area of law ? NO . This is a magnitude harder than (h) and First Order Logic is fragile(I think). Seems cool
j. Translating spoken English into spoken Swedish in real time ? YES. But how stylish and aesthetic is it ?
k. Performing a complex surgical operation ? NO. A high quality help in high-precision operation
…….I think that is enough for first day of class. I left the class thinking about us human and our intelligence : are we God ourself? because we are created by some guy in his image( according to my Christian friends ) but we don’t have a mind of God. To create an AI is to play God ourself isn’t it ? What we can ask about AI, can God ask about human ? “Hey dude, you are not behaving like God.”…..Just some shower thought. I really don’t know……..
Anyway, I am traveling back to the 80s for paranormal settings where I don’t need to think about these AI things…….
Hmm, limit to oner episode per day because binge watch leads to binge procrastination, luckily Casey is on hiatus for a week now :