As technology grows and the field of artificial intelligence grows more mature every day, people and and more more clos closee to thes thesee new new inven inventi tions ons in thei theirr daily daily life life.. Whil Whilee peopl peoplee exper experie ienc ncee the the advantages the technologies bring along, they also are afraid the “dooms day” the technology could bring upon them. In some movies, robots have exceeding intelligent and bring trouble. Yet is it possible to really build AIs that can intimidate human mind !an scientists ma"e machines that can thin" in the same way as human do Intelligence machines li"e #uring machine can imitate human thin"ing but only superficially. #he Idea of “Artificial Intelligence” was first put forward by the great scientist Alan $athison #uring. As a great scientist and innovator, #uring designed a machine, which he hopes can imitate human. A #uring machine ta"es an income statement and %udge whether it is true or false &'ar"ada"is &'ar"ada"is ())*. #his machine can theoreticall theoretically y calculate calculate logical logical answers answers mathematical mathematically ly &'ar"ada"is ())*. Ideally, a #uring+s machine can ta"e any income message and process an outcome eventually. Yet in reality, there will be a point that the #uring machine always have a new true or false uestion after the last one and it repeats forever &'ar"ada"is ())*. #hen the machine cannot tell whether the statement is true or not &'ar"ada"is ())*. #herefore, the dream of the “ultimate statement answerer”, in this situation, is not practical &'ar"ada"is ())*. #here is also an idea of a “sonnet writing machine” that #uring imagines, which goes li"e this Interrogator: Interrogator: In the first line of your sonnet which reads "Shall I compare thee to a summer's day," would not "a spring day" do as well or better? Witness: Witness: It wouldn't scan. Interrogator: Interrogator: How about "a winter's day," day," hat hat would scan all right. Witness: !es, but nobody wants to be compared to a winter's day. Interrogator: Interrogator: Would Would you say r. #ic$wic$ reminded reminded you of of %hristmas? %hristmas? Witness: In a way. Interrogator: Interrogator: !et !et %hristmas is is a winter's day, and I do not thin$ thin$ r. r. #ic$wic$ #ic$wic$ would would mind the comparison. comparison. Witness Witness:: I don't thin$ you're serious. serious. &y a winter's winter's day one means a typical typical winter's winter's day, rather rather than a special one li$e %hristmas. &ey %hristmas. &ey /0)*
Yet this machine is a pure imagination and successfully ma"ing it does not means it can thin" as human do. 1eside 1eside the machine, #uring #uring also offered offered the famous “#uring “#uring test” which focuses focuses on whether or not a machines can act li"e a human by imitating human behavior &enderson /23*. It is said that the day AI passes the #uring test will be the doomsday of human being. owever, a computer program posing as a (45year5old 6"rainian boy passed the #uring #est in /7(3 &ulic" 02*. Yet its success is not admitted by all people, because when 8ugene was as"ed, “9id you see :ubilee”, he responded, “#ry to guess; Actually, I don+t understand why you are interested. I "now you are supposed to tric" me.” which can be the answer to any uestion it cannot answer directly &ulic" 02*. #hus some scientists argue that the #uring test tests AI too narrowly, so instead, instead, they prefer identifying identifying AI through its logic &enderson &enderson /<7*. It is unarguable, unarguable, that the #uring test cannot help scientists tell whether a machine can thin" in the exact same process as human do. #herefore, #herefore, the #uring=s #uring=s machine machine and #uring=s #uring=s test are too superficia superficiall for creating an intelligence machine that can thin" as same as human do. 9espite huge efforts, scientists have not successfully build a machine that reaches the height of human intelligence. >cientists have been trying to ma"e AI smarter and smarter, but fast and “smart” does not necessarily means that the AI thin"s the same way human do. #a"e the chess playing machines &li"e 9eepblue* as an example. #he machines do not thin" the same way a human player does &enderson 4/*. ?o human player would calculate every possible step before they move and no one can do that &enderson 4/*. Yet this is the exact way computer does and
beats the human players &enderson 4/*. #here is also a famous thought experiment @!hinese oom@. #his experiment loc"s a person, who does not spea" !hinese at all, into a room with a database of !hinese symbols and instructions &ey 0<*. It says that by following the instructions, the person will be able to give right answers to uestions in !hinese, which are given from a person out of the room &ey 0<*. In this case, one can pass a !hinese understanding #uring #est #est without understanding !hinese at all &ey 0<*. With larger and larger memories, computers can store more and more data in them as "nowledge &ey )2*. #herefore, letting the computer programs to be more capable of imitating the decision ma"ing process of human &ey )2*. In /772, Berrucci=s team started building a machine for the game :eopardy; &ey /C0*. Yet it is hard for them to let the computer recogniDe where to search &ey /C0*. #he uestions have all the different points to start with and the computer do not "now how to %udge which one is the one it should search with &ey /C0*. #herefore, the computer has to search though a lot of things for possible answers &ey /C0*. /C 0*. Although searching through from all a ll the different starts ta"es a lot of time &ey /C0*. #he team finally found a way to reduce the time, which is to let the computer do all the different search in the same time &ey /C0*. #he machine Watson won the game against human, used more than (77 different technologies to understand every single clue, decide how to find the answer, and list all the answers &ulic" <*. Watson used more than (77,777 sample uestions to train for its :eopardy; competition &ulic" <*. Yet this is not how human thin". >o some scientists loo"ed into exactly how human thin". ?eurons, being in charge the human thin"ing process, handle information by deciding whether the incoming signals pass a certain reuirement or not &ey timulating neurons in human brain, Artificial neural networ"s &A??s* ta"es input and form output from them &ulic" (2*. #here are more than a billion connections in the biggest A??s in AI today &ulic" (2*. 1y processing huge amount of data, it can find patterns and combine them together into higher level meaning &ulic" (2*. 1ut does any an y human need a ton of data to learn to ma"e a simple decision Erobably not. #herefore, even though scientists try to ma"e their machines thin" li"e human by every conceivable means, the current existing machines still cannot reach the height of h uman intelligence. $achines do not learn as efficiently or effectively as humans. What ma"es machines intelligence unable to reach human intelligence If scientists want to build machines that can thin" the same as human do, first they need to "now how human thin" &>tuart /*. If people can understand a human mind through introspection, psychological experiments, or brain imaging, it will be possible to compare the input and output of the machines &>tuart /*. Yet Yet it is is hard for machines to thin" rationally since the information they obtain is not (77 F certain as well as formal &>tuart /*. Also, they difference between @principle@ and @real life@ ma"es it even harder to solve uestion rationally &>tuart /*. $ins"y offered that human thin"ing can be divided into ) levels“Inborn, Instinctive eactions”, “Gearned eactions”, “9eliberative #hin"ing, eflective #hin"ing, >elf5eflective #hin"ing”, and “>elf5!onscicous 8motion” &(47*. Animals, including human, of course, are born with @instincts@ li"e dodging and see"ing food in order to survive &$ins"y (44*. Yet Yet if the environment changes, they may need to adopt new habits and change chang e their reactions to certain things &$ins"y (44*. And when there is a bran5new thing happening, animals try random actions, and then they the y "now which is the “right” reactions for this situation, thus this reaction get “reinforced” &$ins"y (44*. #herefore, when the same thing happens again, it is li"ely the animal will have the same reaction, again &$ins"y (44*. #hen the mind rethin" what it has done or what it is going to do &$ins"y (3/5(34*. #hus letting the individual h ave the
idea of if he or she is doing a right or wrong thing &$ins"y (3/5(34*. Am I confused &$ins"y (3/5(34* Am I on the right trac" &$ins"y (3/5(34* 8ven further, when people are as"ing themselves even more complicated uestions li"e @What would he have thought of me after I do that”, they are in the sixth level leve l of the mind, the level of >elf5!onscious eflection &$ins"y (3)*. #hey set a goal of what they @should be@, did what they have done, and then examine and reflect on whether they met the goal &$ins"y (3)*. As uman beings, people can thin" bac" an d forth about what they were thin"ing earlierH they can ma"e random decision which they who decided it cannot even tell why. owever, every brain activity people go through daily, everything they ta"e normally, are not so easy for a machine to accomplish. As Ga"e says that “Bor most interesting "inds of natural and man5made categories, people can learn a new concept from %ust one or a handful of examples, whereas standard algorithms in machine learning reuire tens or hundreds of examples to perform similarly &(44/*.” 8ven a children can learn a bran5new concept by comparing and generalising the new idea with "nown ones &Ga"e (44/*. Yet Yet man5 made machines, especially the most frontier ones li"e the @deep5learning@ ones, reuires a ton of of data to learn new things &Ga"e (44/*. Burthermore, people learn a wider range of things than machines do when they are given the same material to learn from &Ga"e (44/*. uman can create new ideas based on the given one, and probably some other information they gathered in their mind through life &Ga"e (44/*. uman has the ability to learn a rich number of information from only small amount of data &Ga"e (444*. (44 4*. owever, learning a more complicated model, in any learning theory, reuires more instead of less data &Ga"e (444*. $achines under current technology can do no more than storing new concepts, and deal with them exactly as same as programmed. #herefore, as Ga"e mentioned in his article, “A central challenge is to explain these two aspects of human5level concept learning- ow do people learn new concepts co ncepts from %ust one or a few examples And how do people learn such abstract, rich, and flexible representations An even greater challenge arises when putting them together- ow can learning succeed from such sparse data yet also produce such rich representations &(444*” Eeople always ta"e their ability of @thin"ing smoothly@ for granted &$ins"y /()*. Yet Yet they overloo"ed the past they the y have been through &$ins"y /()*. 1ac" when they were infants, people learnt how to pic" up things, recogniDe what is edible, how to tal", and how to ma"e decisions &$ins"y /()*. #heir ability of thin"ing humanly is built bit by bit in their infancy &$ins"y /()*. !omparing to human mind, the man5made machines are too far behind on the depth and efficiency. #herefore, building machines that can achieve achiev e human thin"ing is highly impossible. It is true, that there is still chance that machines can beat human using their vast space of memory and incredible speed in calculation and acuiring ac uiring information. owever, human mind wor"s in a far too delicate way for man5made machines to imitate. ?o matter if it is the way human learn new "nowledge, process them, or ma"e decisions and do further thin"ing based on the "nowledge, there is no chance that it is possible for machines to wor" the same way.