| The primary focus of this essay is the future of | | | | speeding up tasks still performed by people such |
| Artificial Intelligence (AI). In order to better | | | | as the rule based AI systems used in accounting |
| understand how AI is likely to grow I intend to | | | | and tax software, enhance automated tasks such |
| first explore the history and current state of AI. | | | | as searching algorithms and enhance mechanical |
| By showing how its role in our lives has changed | | | | systems such as braking and fuel injection in a |
| and expanded so far, I will be better able to | | | | car. Curiously the most successful examples of |
| predict its future trends. | | | | artificial intelligent systems are those that are |
| John McCarthy first coined the term artificial | | | | almost invisible to the people using them. Very |
| intelligence in 1956 at Dartmouth College. At this | | | | few people thank AI for saving their lives when |
| time electronic computers, the obvious platform | | | | they narrowly avoid crashing their car because of |
| for such a technology were still less than thirty | | | | the computer controlled braking system. |
| years old, the size of lecture halls and had storage | | | | One of the main issues in modern AI is how to |
| systems and processing systems that were too | | | | simulate the common sense people pick up in their |
| slow to do the concept justice. It wasn't until the | | | | early years. There is a project currently |
| digital boom of the 80's and 90's that the | | | | underway that was started in 1990 called the |
| hardware to build the systems on began to gain | | | | CYC project. The aim of the project is to provide |
| ground on the ambitions of the AI theorists and | | | | a common sense database that AI systems can |
| the field really started to pick up. If artificial | | | | query to allow them to make more human sense |
| intelligence can match the advances made last | | | | of the data they hold. Search engines such as |
| decade in the decade to come it is set to be as | | | | Google are already starting to make use of the |
| common a part of our daily lives as computers | | | | information compiled in this project to improve |
| have in our lifetimes. Artificial intelligence has had | | | | their service. For example consider the word |
| many different descriptions put to it since its birth | | | | mouse or string, a mouse could be either a |
| and the most important shift it's made in its | | | | computer input device or a rodent and string |
| history so far is in how it has defined its aims. | | | | could mean an array of ASCII characters or a |
| When AI was young its aims were limited to | | | | length of string. In the sort of search facilities |
| replicating the function of the human mind, as the | | | | we're used to if you typed in either of these |
| research developed new intelligent things to | | | | words you would be presented with a list of links |
| replicate such as insects or genetic material | | | | to every document found with the specified |
| became apparent. The limitations of the field were | | | | search term in them. By using artificially intelligent |
| also becoming clear and out of this AI as we | | | | system with access to the CYC common sense |
| understand it today emerged. The first AI | | | | database when the search engine is given the |
| systems followed a purely symbolic approach. | | | | word 'mouse' it could then ask you whether you |
| Classic AI's approach was to build intelligences on | | | | mean the electronic or furry variety. It could then |
| a set of symbols and rules for manipulating them. | | | | filter out any search result that contains the word |
| One of the main problems with such a system is | | | | outside of the desired context. Such a common |
| that of symbol grounding. If every bit of | | | | sense database would also be invaluable in helping |
| knowledge in a system is represented by a set | | | | an AI pass the Turing test. |
| of symbol and a particular set of symbols ("Dog" | | | | So far I have only discussed artificial systems |
| for example) has a definition made up of a set of | | | | that interact with a very closed world. A search |
| symbols ("Canine mammal") then the definition | | | | engine always gets its search terms as a list of |
| needs a definition ("mammal: creature with four | | | | characters, grammatical parsers only have to deal |
| limbs, and a constant internal temperature") and | | | | with strings of characters that form sentences in |
| this definition needs a definition and so on. When | | | | one language and voice recognition systems |
| does this symbolically represented knowledge get | | | | customise themselves for the voice and language |
| described in a manner that doesn't need further | | | | their user speaks in. This is because in order for |
| definition to be complete? These symbols need to | | | | current artificial intelligence methods to be |
| be defined outside of the symbolic world to avoid | | | | successful the function and the environment have |
| an eternal recursion of definitions. The way the | | | | to be carefully defined. In the future AI systems |
| human mind does this is to link symbols with | | | | will to be able to operate without knowing their |
| stimulation. For example when we think dog we | | | | environment first. For example you can now use |
| don't think canine mammal, we remember what a | | | | Google search to search for pictures by inputting |
| dog looks like, smells like, feels like etc. This is | | | | text. Imagine if you could search for anything |
| known as sensorimotor categorization. By allowing | | | | using any means of search description, you could |
| an AI system access to senses beyond a typed | | | | instead go to Google and give it a picture of a |
| message it could ground the knowledge it has in | | | | cat, if could recognise that its been given a picture |
| sensory input in the same manner we do. That's | | | | and try to assess what it's a picture of, it would |
| not to say that classic AI was a completely | | | | isolate the focus of the picture and recognise that |
| flawed strategy as it turned out to be successful | | | | it's a cat, look at what it knows about cats and |
| for a lot of its applications. Chess playing | | | | recognise that it's a Persian cat. It could then |
| algorithms can beat grand masters, expert | | | | separate the search results into categories |
| systems can diagnose diseases with greater | | | | relevant to Persian cats such as grooming, where |
| accuracy than doctors in controlled situations and | | | | to buy them, pictures etc. This is just an example |
| guidance systems can fly planes better than pilots. | | | | and I don't know if there is currently any |
| This model of AI developed in a time when the | | | | research being done in this direction, what I am |
| understanding of the brain wasn't as complete as | | | | trying to emphasise in it is that the future of AI |
| it is today. Early AI theorists believed that the | | | | lies in the merging existing techniques and |
| classic AI approach could achieve the goals set | | | | methods of representing knowledge in order to |
| out in AI because computational theory supported | | | | make use of the strengths of each idea. The |
| it. Computation is largely based on symbol | | | | example I gave would require image analysis in |
| manipulation, and according to the Church/Turing | | | | order to recognise the cat, intelligent data |
| thesis computation can potentially simulate | | | | classification in order to choose the right |
| anything symbolically. However, classic AI's | | | | categories to sub divide the search results into |
| methods don't scale up well to more complex | | | | and a strong element of common sense such as |
| tasks. Turing also proposed a test to judge the | | | | that which is offered by the CYC database. It |
| worth of an artificial intelligent system known as | | | | would also have to deal with data from a lot of |
| the Turing test. In the Turing test two rooms | | | | separate databases which different methods of |
| with terminals capable of communicating with each | | | | representing the knowledge they contain. By |
| other are set up. The person judging the test sits | | | | 'representing the knowledge' I mean the data |
| in one room. In the second room there is either | | | | structure used to map the knowledge. Each |
| another person or an AI system designed to | | | | method of representing knowledge has different |
| emulate a person. The judge communicates with | | | | strengths and weaknesses for different |
| the person or system in the second room and if | | | | applications. Logical mapping is an ideal choice for |
| he eventually cannot distinguish between the | | | | applications such as expert systems to assist |
| person and the system then the test has been | | | | doctors or accountants where there is a clearly |
| passed. However, this test isn't broad enough (or | | | | defined set of rules, but it is often too inflexible in |
| is too broad...) to be applied to modern AI | | | | areas such as the robotic navigation performed |
| systems. The philosopher Searle made the | | | | by the Mars Pathfinder probe. For this application a |
| Chinese room argument in 1980 stating that if a | | | | neural network might be more suitable as it could |
| computer system passed the Turing test for | | | | be trained across a range of terrains before |
| speaking and understanding Chinese this doesn't | | | | landing on Mars. However for other applications |
| necessarily mean that it understands Chinese | | | | such as voice recognition or on the fly language |
| because Searle himself could execute the same | | | | translation neural networks would be too inflexible, |
| program thus giving the impression that he | | | | as they require all the knowledge they contain to |
| understand Chinese, he wouldn't actually be | | | | be broken down into numbers and sums. Other |
| understanding the language, just manipulating | | | | methods of representing knowledge include |
| symbols in a system. If he could give the | | | | semantic networks, formal logic, statistics, |
| impression that he understood Chinese while not | | | | qualitative reasoning or fuzzy logic to name a few. |
| actually understanding a single word then the true | | | | Any one of these methods might be more |
| test of intelligence must go beyond what this test | | | | suitable for a particular AI application depending on |
| lays out. | | | | how precise the effects of the system have to |
| Today artificial intelligence is already a major part | | | | be, how much is already known about the |
| of our lives. For example there are several | | | | operating environment and the range of different |
| separate AI based systems just in Microsoft | | | | inputs the system is likely to have to deal with. |
| Word. The little paper clip that advises us on how | | | | In recent times there has also been a marked |
| to use office tools is built on a Bayesian belief | | | | increase in investment for research in AI. This is |
| network and the red and green squiggles that tell | | | | because business is realising the time and labour |
| us when we've misspelled a word or poorly | | | | saving potential of these tools. AI can make |
| phrased a sentence grew out of research into | | | | existing applications easier to use, more intuitive |
| natural language. However, you could argue that | | | | to user behaviour and more aware of changes in |
| this hasn't made a positive difference to our lives, | | | | the environment they run in. In the early day of |
| such tools have just replaced good spelling and | | | | AI research the field failed to meet its goals as |
| grammar with a labour saving device that results | | | | quickly as investors believed it would, and this led |
| in the same outcome. For example I compulsively | | | | to a slump in new capital. However, it is beyond |
| spell the word 'successfully' and a number of | | | | doubt that AI has more than paid back its thirty |
| other word with multiple double letters wrong | | | | years of investment in saved labour hours and |
| every time I type them, this doesn't matter of | | | | more efficient software. AI is now a top |
| course because the software I use automatically | | | | investment priority, with benefactors from the |
| corrects my work for me thus taking the | | | | military, commercial and government worlds. The |
| pressure off me to improve. The end result is | | | | pentagon has recently invested $29m in an AI |
| that these tools have damaged rather than | | | | based system to assist officers in the same way |
| improved my written English skills. Speech | | | | as a personal assistant normally would. |
| recognition is another product that has emerged | | | | Since AI's birth in the fifties it has expanded out |
| from natural language research that has had a | | | | of maths and physics into evolutionary biology, |
| much more dramatic effect on people's lives. The | | | | psychology and cognitive studies in the hope of |
| progress made in the accuracy of speech | | | | getting a more complete understanding of what |
| recognition software has allowed a friend of mine | | | | makes a system, whether it be organic or |
| with an incredible mind who two years ago lost | | | | electronic, an intelligent system. AI has already |
| her sight and limbs to septicaemia to go to | | | | made a big difference to our lives in leisure |
| Cambridge University. Speech recognition had a | | | | pursuits, communications, transportation, sciences |
| very poor start, as the success rate when using | | | | and space exploration. It can be used as a tool to |
| it was too poor to be useful unless you have | | | | make more efficient use of our time in designing |
| perfect and predictable spoken English, but now | | | | complex things such as microprocessors or even |
| its progressed to the point where its possible to | | | | other AI's. In the near future it is set to become |
| do on the fly language translation. The system in | | | | as big a part of our lives as computer and |
| development now is a telephone system with real | | | | automobiles did before it and may well begin to |
| time English to Japanese translation. These AI | | | | replace people in the same way the automation |
| systems are successful because they don't try to | | | | of steel mills did in the 60's and 70's. Many of its |
| emulate the entire human mind the way a | | | | applications sound incredible, robot toys that help |
| system that might undergo the Turing test does. | | | | children to learn, intelligent pill boxes that nag you |
| They instead emulate very specific parts of our | | | | when you forget to take your medication, alarm |
| intelligence. Microsoft Words grammar systems | | | | clocks that learn your sleeping habits or personal |
| emulate the part of our intelligence that judges | | | | assistants that can constantly learn via the |
| the grammatical correctness of a sentence. It | | | | internet. However many of its applications sound |
| doesn't know the meaning of the words, as this is | | | | like they could lead to something terrible. The |
| not necessary to make a judgement. The voice | | | | pentagon is one of the largest investors in artificial |
| recognition system emulates another distinct | | | | intelligence research worldwide. There is currently |
| subset of our intelligence, the ability to deduce the | | | | much progressed research into AI soldier robots |
| symbolic meaning of speech. And the 'on the fly | | | | that look like small tanks and assess their targets |
| translator' extends voice recognitions systems | | | | automatically without human intervention. Such a |
| with voice synthesis. This shows that by being | | | | device could also be re-applied as cheap domestic |
| more accurate with the function of an artificially | | | | policing. Fortunately the dark future of AI is still a |
| intelligent system it can be more accurate in its | | | | Hollywood fantasy and the most we need to |
| operation. | | | | worry about for the near future is being beaten |
| Artificial intelligence has reached the point now | | | | at chess by a children's toy. |
| where it can provide invaluable assistance in | | | | |