What is intelligence?

Intelligence is a human’s ability to learn from experience, adapt to new situations, use knowledge to manipulate one’s environment, and lastly, understand and handle abstract concepts.

So, how to create an artificial intelligence?

There are many definitions of artificial intelligence, but here we will talk about 4 approaches to artificial intelligence from Russel & Norvig in their book, Artificial Intelligence, a modern approach, third edition.

The first approach is thinking rationally; this approach uses the law of thought developed by Aristotle in his syllogisms, a form of reasoning that provided patterns for argument structures that always yielded correct conclusions when given correct premises. The second approach involves thinking humanly; the objective is to imitate the actual workings of human minds, for example, the communication between neurons when we get a stimulus. The third approach is acting rationally; we used the rational agent approach, whose objective is to achieve the best outcome, or the best expected outcome, when there is uncertainty. The fourth approach is acting humanly; to call a machine acting humanly, it should pass the Turing Test, when a human interrogator can’t tell whether the responses come from a person or from a machine. There are some capabilities to pass the Turing test: first, natural language processing to enable the machine to communicate like a human; knowledge representation to store what it knows or hears; automated reasoning to use the stored information to answer questions and draw new conclusions; and lastly, machine learning to adapt to new circumstances and detect and extrapolate patterns.

We have mentioned machine learning earlier as an ability of machines to adapt to new circumstances and to detect and extrapolate patterns. There are 3 types of machine learning, 1st is supervised learning, 2nd is unsupervised learning, and 3rd is reinforcement learning. So, for supervised learning, here we have data in the table; there are 4 columns here: sky, air temperature, humidity, wind, and play (Fig 1). Assume we want to predict the value of the column “play,” whether yes or no, based on the value in other columns. So, we called column “play” as target and other columns as a predictor. And we refer to this table as training data. Next, we use the training data as input in the learning process to build a trained model. We call this process training. After we get the trained model, we can make a prediction of the target from the testing data. The second type of machine learning is unsupervised learning, where the difference between data in supervised and unsupervised learning is that we don’t have a target in our data. Look at these points. We have the x and y axes as our attributes. What we do in clustering is make a categorization of these data based on their similarity and difference. This is the example if we make a categorization based on x and y values. We have three classes: the blue one, the red one, and the green one (Fig 2). The last is reinforcement learning. Assume the biggest rectangle is our environment, and inside this environment we have our agent in a state.

Pelajari lebih lengkap disini.