Introduction
The creation of artificial life, artificial life that clearly represents an improvement over life as it has evolved. That is how most science fiction authors and movie-makers view AI. For example, one movie contains the typical Frankenstein AI scenario. In that movie, self-aware robots rebel against their makers and decide to challenge their right to determine who lives and dies. More recently, another movie depicts a future with self-aware machines that are locked into a bitter battle for survival with humans. Popular science fiction books depict AI as not only independent but, in many ways, superior to conventional human beings
But these portrayals of AI software are, for the most part, fantasy.
At the other end of the scale, we have all the very real yet very specific AI programs that have been written and are available now. These should be viewed more as amazing tools that engineers have built rather than tools that have the ability to determine their own destiny. Desirable behavior from an AI program would be some kind of behavior that one would expect from a human as a result of a reasoning process. All AI systems, to some extent, mimic human behavior in order to solve problems that require intelligence. But most of the AI systems that we have today are users of a very small set of very specific techniques, more an engineering marvel than a psychological artifact
1. Fundamentals of machine learning
Traditional software systems follow a predetermined sequence of instructions to reach the solution. These systems are designed to perform tasks that have well-defined steps. They cannot generalize the answer across data points that were not designed to handle. Consequently, when a part of the task changes, the software needs to be reprogrammed to handle that task. Machine learning (ML) can address this limitation. It is a technique to give computers the ability to learn without being explicitly programmed. A new example can cause the program to change its processing rules as it learns more about any given task.
An ML model understands patterns and insights from data and then uses them to take appropriate actions. It can quickly process and analyze large volumes of data and automatically learn and adapt from past decisions. ML models can take the form of a rule, decision tree, linear or nonlinear function that maps input data to an output prediction. ML prediction can be a score, time series, category, etc. The model can be an ensemble of models, a neural network, or an output from a decision tree. The data can be structured or unstructured, labeled or unlabeled. It can also consist of features used to predict and allocate target data.
ML prediction analysis has two types: supervised and unsupervised. In the supervised model, the data is pre-labeled, while in the unsupervised case, the data is not; the model is not explicitly trained; it discovers the patterns. The key difference between ML and traditional software is that ML uses more data with fewer clear rules, while traditional software incorporates a high amount of explicitly programmed rules that use data
2. Definition and scope
Artificial intelligence (AI) falls within the field of information technology (IT). In the past few years, AI has become one of the trendiest fields in all areas due to the impressive advances experienced. In particular, within AI, the topic of machine learning (ML) is of great relevance. However, it is not simple to put a boundary between AI and ML. In this book, we will not use the term ‘AI’ too much, and when we do, we will consider it as a synonym for ML, or else we will make a clear distinction. However, we must know that AI is much more than ML, and in this chapter, we provide an overview of what is within the scope of AI, together with a general introduction to the evolution of AI in its history, addressing some relevant topics.
The field of AI is large and consists of a large number of subfields: knowledge representation and reasoning; machine learning; constraints satisfaction; natural language processing; planning; environmental perception; multi-agent systems; scheduling; game theory and combinatorial extensions; robotics; AI and creativity; emotions and AI; agents and AI-based simulation systems; AI and computer vision; search; and evaluation of AI techniques. We link each of these fields to specific problems (and sometimes to general problems), which can be solved with AI techniques. Each of these fields usually has a separate learned command, but the combination of AI subfields allows progress to be made in a problem that is very complex for one field to solve. For example, currently, most successful AI applications make specific reference to building a model: the model making use of a machine learning approach may be part of a more complex process that includes both search techniques and an automated assistant.
3. Deep learning and neural networks
Deep learning is an approach to artificial intelligence in which artificial neural networks with many layers are used to imitate the human brain. These systems can be trained to recognize patterns from billions of data points and learn to make human-like decisions in specific tasks.
Deep learning uses a large amount of data, many layers, and simple algorithms. Neural networks have more than one linear layer or processing layer and are built in the form of neurons. Deep learning is also good at building deep representations from unsupervised pre-training. A special backpropagation algorithm is used to make learning feasible in these networks. When enough training has been performed, a neural network performs a particular task without being explicitly programmed. This is called unsupervised learning. Neural networks used in deep learning have many hidden layers.
Artificial neural networks are inspired by the human brain and consist of a number of neurons. Deep learning focuses on the adaptive learning capability of a network. Deep learning operates by algorithm and model learning and is based on trial and error. When an artificial neural network observes a task enough times, it can learn to perform it either more quickly or more accurately. The trained system uses the results to improve its understanding of the task. By increasing the layers in the neural network, the system can usually improve accuracy with minimal internal handling.
4. Key technologies driving AI
Hardly had anybody heard of artificial intelligence just a few years back! Until a couple of years ago, the concept of artificially producing intelligence was confined to sci-fi movies only. Then came humanoid robots making lifelike appearances. Even those thought it to be yet another publicity stunt by a bunch of wealthy men. But before people could forget those tall claims, came a piece of unbelievable news. And that is, the world of artificial intelligence has announced its arrival.
Is it the other side of the moon being discovered? The thought was almost equally bewildering. Pretty soon, another application of artificial intelligence hit the headlines. And that is, a computer program has started writing. So, artificial intelligence has finally paid off. The thinking machine is here to stay, and perhaps, to reign supreme someday soon. A question then arises: what are the key technologies spearheading this development? The journey of AI started with a question: can machines do what we can do? Can machines carry out tasks as efficiently as we label those as requiring intellectual capabilities? The initial answer to all such questions is given by logic-based systems. Also regarded as ‘symbolic’ systems, these are based on manipulation of symbols representing knowledge in a specific domain, mostly requiring expert intervention.
Conclusion
As we have said earlier, companies typically perceive technologies as a means, not as an end. Although AI technologies will most likely play only a handful roles in terms of their business operations, the payoff is potentially larger than that of any business technology since the mainframe computer or the PC. But how to use this technology is less clear. It is essential, we believe, to approach the task of making sense of AI from a perspective that encompasses all business functions and all aspects of corporate leadership. For both line executives and strategic planners, the former must become familiar with the evolving capabilities of sophisticated AI technologies and the pace of their advancement, and combine this knowledge with solid business fundamentals and proven leadership mechanisms. Planning on the potential disruption from AI, the course for the future corporate strategy, and other critical technology choices, implies not only knowledge of AI, but a synthetic understanding of what other technologies are capable of and what they are currentlyย promising
Leave a Reply