Artificial Intelligence

When people hear the word artificial intelligence, they only think of robots who are able to be fully integrated into human society. However, there are many types of artificial intelligence and types that are utilized on a daily basis that go unrecognized. Topics such as machine learning, deep learning, and expert systems are talked about in great detail. In this essay I will not only touch on the widely accepted version of artificial intelligence but also the intelligence that is needed during modern day to day life.

Artificial intelligence is constantly in the media in all forms. From movies to devices such as Alexa, it has taken over the world as we know it and it is nearly impossible to live without it. First coined by John McCarthy, artificial intelligence or AI refers to the process of machines replicating humanlike processes such as thinking and understanding. The concept of AI is not new at all, the 1950s is when someone first began hypothesizing about the capabilities for machines.

Alan Turing, a computer scientist and mathematician amongst many other things had a question that defined the goals of all artificial intelligence following after, is it possible for computers or machines think? The Turing Test was created soon after and the object of the test was for a machine to pass off as a human. It was based off of a game where there would be three people and one person would have to guess which one of the people he was talking to was a woman and which one was a man. The hypothetical test for the artificial intelligence was conducted as such, there are three individuals A,B, and C. They all have to be in separate rooms and person C has to figure which of the entities he is talking to is human and which is computer. All of the responses are through a computer or a typewriter, something where speech is not factored into whether the entity that they are talking to is human or machines.The central goal is for the computer is “make small talk with a human and understand context” (Brian McGuire, History of Artificial Intelligence). To Turing, if a machine could pass this imitation game, then computers had the capability of thinking for themselves. This test has its faults though, just because a computer can imitate a human does not truly mean that it is a sentient being. There is no real way to tell if a machine has consciousness because you are not the machine. The whole point of it is to see whether a computer can trick a human which can be measured it is not possible to measure if something like a machine can have emotions.What Turing was hoping for was that a machine who was able to pass the test would be able to enjoy human things such as art and be able to comprehend poetry which are things that are innately human and that only we do. Though the Turing test has its flaws, people are still trying to solve this theoretical question by building programs that have the goal of passing as human. Some machines are able to pass a certain percentage of the time, but no one machine has passed a majority of the time. There are alternatives to the turing test such as the Feigenbaum Test, which eliminates the task of the computer having to speak in a casual manner as if it were human. This test just requires that the machine is a complete expert on one subject and has to answer questions as well as any expert could. Another proposed alternative was that the human and the machine have to work together to have the human reach a specified goal. The point is to have the machine help the human as a human would help a fellow human. The level of difficulty of these three tests vary and because this is such a difficult topic there is still a lot about this subsect of artificial intelligence to be explored. This however is not the only form of artificial intelligence, as there are other types.

Currently we interact with more AI than we realize because it has been integrated into our society smoothly. There are different forms of artificial intelligence: weak AI and strong AI. Strong AI is what was discussed in the above paragraph, the goal is that the machines intellect and functionality is on par with that of a human beings. It should be able to learn and communicate but also hypothetically have thoughts and emotions that go along with it too.This type of AI is more philosophical than anything because it is the idea that these machines would be near human or human adjacent which is all just perception. Weak AI, or narrow AI, is a system that was created for a particular task. Take a look at Siri, Apple’s virtual assistant on their products. Siri is able to set reminders, take down notes, play any song that you request, and have simple conversation. However it is weak AI because it is only programmed for one task truly and that is to be a personal phone assistant, whilst strong AI is capable of doing anything that a human can. Weak AI is the type we tend to interact with more on a day to day basis. These machine are meant to seem as though they can think but have no true consciousness within them, they can never do anything for themselves and are bound by the rules that are given to them. There is also one last type of intelligence that is not even close to perfecting: superintelligence. Super intelligence is when machines improves at a rapid amount in short amounts of time so that its intelligence surpassess even the most intelligent human beings. It is seen mostly in science fiction when robots are able to take over the world. This also poses most of the current ethics discussions about artificial intelligence. Even though the three seem worlds apart they are both types of intelligence that have their own importance.

The 1950s through the 1970s were the age of neural networks or the thinking machines. Computer scientists, rather than putting more effort in creating intelligent machines have focused on letting machines learn for themselves when given information, recognize patterns and come up with solutions without much human interference. This branch of artificial intelligence is called machine learning and it is literally everywhere. The idea came about when researchers realized that having computers teach themselves would be way more efficient than having to program them to know everything about the world and carry out the specific tasks that they need. Neural networks are a way of programming that allows a computer to learn through observation like a human would. The engineers then began to code the computers to think as if they were human and use the internet as their resource, since it has is so vast and has most of the information in the world stored in it. There are different learning styles that machine learning utilizes: unsupervised and supervised. Unsupervised machine learning is when the computer has to understand patterns without without references and supervised is when that find patterns given some information to guide its searches. Machine learning is used in many different fields for different reasons. For example, healthcare ML is used for devices that monitor health of people in real time and help understand data and find trends for diagnoses. Banking uses ML to help notice fraud and point out investments that would make the company money. Even the education system utilizes it by grading and adapting to each students needs by putting problems catered to them. One of the most interesting products of machine learning is the self driving car. Though its not available for mass production, it uses concepts connected with deep learning to get around. Deep learning, a subset of machine learning, imitates how human beings try and obtain different sorts of knowledge. The technology that the self driving cars use from deep learning is image recognition. Deep learning can also be used for applications such as human speech recognition.

As mentioned in the previous paragraph, human speech recognition is also a section of artificial intelligence. It is called natural language processing, and because language is another part that makes us human, artificial intelligence must be able to understand all of the nuances of it. All of the rules of language have to be made into smaller pieces so that a computer can understand the data since, at least in english, words can have different meanings and can change parts of speech. Natural language processing is seen in again, Apple’s virtual assistant Siri. A key part of NLP is that the machine has to recognize relationships with other topics. An example is google’s knowledge graph, it enhances searches by suggesting other topics that could be related

to the one that is being googled.This is important because it sets up for chatbots, artificial intelligence that conducts conversation via text, to be able to more accurately communicate and connect with the human speaker. With actual speech, computers use a spectrogram and how much waves are displaced when certain letters or words are being said which helps the computer recognize the different words being said. This as well as everything else mentioned needs a lot of data for it to truly work out.

Machine perception goes hand and hand with Natural Language processing. Machine perception is when computers attempt to simulate how humans experience the world around them, and make sense of it. It is noted that MP references any sense that a human can have that a machine replicates, though a lot of the time it refers to sight. Computer vision is the actual term for when a machine has “high level” understanding from pictures and videos. What that means is that machines are able to look at these images and identify what is going on in it. Because computers make images through millions of pixels, it could be hard for the machine to interpret what it is faced with. In order for a computer to recognize what it “sees” programmers train the computer with hundreds of the same image to look at. So if hypothetically, if a machine was trying to interpret a picture of a dog, it would not try to figure it out by the specific features that a dig has, it would be through all of the hundreds of pictures of what a dog looks like that draws upon. We can see computer vision being used with apple products, face recognition to enter a phone and taking all the photos of specific people and grouping them together.

Fuzzy logic attempts to emulate human reasoning with a machine and it imitates how humans are able to make decisions without definite answers. Computers usually only answer with surefire yes or no answers but with fuzzy logic there are a range of answers in between those two. Rather than making exact measures for things humans use partial truths until we are definite about something.Fuzzy logic is good to use because it doesn’t need exact data For example fuzzy logic utilizes responses such as certainly yes, possibly yes, maybe, possibly no and certainly no, they are all on spectrums rather than yes or no. Since it has its basis in natural language, it sounds more like human reasoning than a machines.

Machines have the capability of being “experts” in certain fields. Mentioned earlier with the Feigenbaum test, they are expected to perform tasks at the level that a human expert would be able to. These systems have to be specialists in the topic that is being researched, have to be able to learn on its own, has to be able to justify all of the conclusions it comes to, and know how reliable the answer that its giving is. All expert systems have two major components: knowledge base and interference engine. The knowledge base is the data and skills which are at the same level as where a humans would be. The interface engine is the portion that uses all the data to learn new things or find new patterns with the system. These expert systems are usually needed when human experts are not readily available, the problem is not too big for the computer to handle, and when regular compting fails, amongst other reasons.

In conclusion, artificial intelligence does not only refer to one thing. There is weak, strong and super intelligence, all at various different levels of actions they can conduct. Machine learning and neural networks make it so that machines are able to learn like humans do and natural language processing help machines understand language like a human can. Machine perception makes it so that machines can experience senses like human can, fuzzy logic gives the machine a spectrum of answers rather than definite ones, and lastly expert systems are experts in certain topics. All of these could be put together to finally make the type of intelligence Alan Turing was speaking about when he asked, can machines think?