Week 10 Lecture (3).pptx

Artificial Intelligen ce Week 10, BU 1173, Miss Amanda Paananen Review: What is Artificial Intelligence? There are a variety of definitions for Artificial Intelligence. Generally, AI is a program or robot performing a task that requires some form of intelligence. AI also refers to how a computer m...

Artificial Intelligen ce Week 10, BU 1173, Miss Amanda Paananen Review: What is Artificial Intelligence? There are a variety of definitions for Artificial Intelligence. Generally, AI is a program or robot performing a task that requires some form of intelligence. AI also refers to how a computer might develop the ability to learn or think on its own. Of course, there are various ways computer scientists define what it means to think The Turing Test In the 1950’s, British computer scientist Alan Turing proposed a test that went on to be named the Turing Test. The Turing Test is where a computer can mimic human responses under certain conditions. In the test, there are three separated stations. At one station is a human asking questions. At the second station is a computer, and at the third station is a human answering questions. The human asking the questions gives a series of questions to both the computer and the other human who is answering the questions. If the questioner can not tell the difference between the computer and the other human answering the questions, then the computer is said to have intelligence. Other Definitions… Another definition of AI is any task that a computer does that a human cannot tell if it was performed by a computer or a human. AI could also be described as the ability to perceive and act upon its environment. AI is already being used in many ways, from Amazon’s “Alexa” and Apple’s “Siri” answering your questions, to programs that can recognize what is in a picture, as well as bots on twitter deleting hateful posts Different Types of AI Artificial Intelligence is generally broken into two types: 1) artificial narrow intelligence (weak AI) 2) artificial general intelligence (strong AI) Weak AI Weak AI is the kind that completes specific tasks without being programmed to do so. Examples AI winning on the game show Jeopardy! or beating a grandmaster at chess. It is what allows Maps to give you the best route to a destination based on current traffic, and it is what allows “Siri” to answer basic questions on your phone. It can help doctors spot tumors in X-Rays, identify inappropriate content online and respond to customer service questions. All of these tasks are completed by the AI using the data it has been fed. It then uses this data to give a response that appears human like. Weak AI is the only type of AI that exists in any real way today in 2022. Strong AI Strong AI is the type computer scientists are working towards developing that could think like a human. Strong AI would be able to learn on its own and upgrade itself. Strong AI would not have pre-programmed responses to questions and would make independent decisions based on the situation. Strong AI would be able to interact with the world around it and learn from it. There would hypothetically be no limit to the amount of knowledge it would eventually gain. This is often called the singularity, where a superintelligence would have unknown limits that would create rapid technological advances with unimaginable changes for humans. Would strong AI eventually take over from the human race? Or would it create an ideal world for all of humankind? How is AI Created? Weak AI responds to what it is programmed to do. Siri’s responses are programmed. It is programmed to recognize the words, “hey Siri” and “play music”. When it detects these words together, it is programmed to play music. Weak AI then, is a program that has been trained to do specific tasks, such as play music when you say the words “play music”. How is AI Created? So how do you “train” or create an AI program? Well, it’s complicated. However, in basic terms, you feed the program large amounts of data to do a specific task. For example, if you wanted to train an AI program to detect tigers in a photograph, you would feed it thousands of pictures of tigers from different angles and in different light. The AI would then create a pattern of what the pixels of a tiger photograph should look like. Then when the AI program is used to examine photos of tigers, it will compare the pixels of that photo with what it knows a tiger should look like. AI learns from examining large amounts of data, or examples, and then does a pattern match to make a decision. Yes, this matches what a tiger looks like. No, this does not match what a tiger looks like. To train an AI program, you require huge amounts of computing power. How does Weak AI work? “You are what you eat!” This is true for Artificial Intelligence too. Artificial Intelligence makes decisions based on the huge amounts of data samples it is feed. If you want to make an AI that can delete hateful speech from Twitter, then you would start by feeding the AI program huge amounts of samples of what hateful speech is and then the AI would use those sample sets to delete speech that matches what you fed it. You train the system to recognize patterns and examples of what you want it to recognize and do. You are training your AI to do a specific, defined task such as remove certain words from social media or identify what is in a photograph. How does Weak AI work? (cont.) With weak AI, you are creating a very specialized system. If you train an AI to play chess, it will only be able to play chess. The goal is to use Artificial Intelligence to do a specific task faster and more accurately than a human could do it. It would be very inefficient to have to hire thousands of humans to search through Twitter for offensive language as opposed to having an Artificial Intelligence be trained to do the same job. Once an Artificial Intelligence has been fed a sufficiently large number of examples or data of what you want it to “learn”, it can now apply that data to completing a task. This is why it is called “weak” AI. It is operating based on the data a human feeds it and it only does that specific task it was designed for. Examples of Weak AI #1 Health Care: AI can be trained to analyze X-rays and recognize tumors in patients. By having a large data set to analyze, the AI could make more accurate decisions about the X-ray and whether the patient has a tumor. AI could save a patient from having to have unnecessary surgery to remove a growth that was not actually cancer. This not only saves the patient from having to do surgery but frees up the doctor’s time to see other patients and saves the medical system money. Examples of Weak AI #2 Google Maps: Google is using AI to improve Maps by using data to create a real time street view map overlaid onto the real world. For example, as you hold your phone up to a downtown city block, a map would be overlaid, showing which route to take next. It would also have pop ups describing the businesses you are looking as well as any other information that might be useful such as menus or ratings for restaurants. It could also tell you about services provided by other businesses you are looking at on your phone. Example s of Weak AI #3 Classifying images: One example of using AI to classify images would be to feed it large amounts of photos of different plant species. In time the AI would be able to quickly and accurately help a botanist identify various plants. Another example would be using AI to identify people in photos. This could be used at a large workplace to enhance security or make communications more efficient. Why Doesn’t Strong AI Exist? Strong AI, the kind that can think for itself, be self-aware, have empathy, upgrade itself and have no limit to its knowledge, resulting in a singularity, does not exist at this point in time. Strong AI would have to replicate the human brain which has roughly 500 trillion synapses in it. There are more connections in our brain than there are stars in our Galaxy. The computing power required to replicate the human brain does not yet exist. Human Intelligence vs. Artificial Intelligence There is much debate as to what human intelligence actually is. What does it mean to think and understand? What does it mean to be self-aware? What does it mean to sense the world around you and react in an intelligent way? These questions don’t have one single answer, which makes defining and creating strong AI difficult. What’s holding us back? Many scientists believe that to truly create a strong AI, you would have to connect it to a body and have it go through the entire range of human development from child to adult. The AI would have to be “born” as a “child” and develop to adulthood, acquiring the necessary experiences gained over a lifetime of interactions. We are still just beginning to understand what it would mean to create intelligence and just how complicated it would be.