Top 5 of AI specialist literature
Reading between the lines: Artificial intelligence in specialist literature
Everyone in our team all started out at the bottom. We have learned, listened, gained our experiences – that’s what it takes to become an AI crack. Of course, we’ve also read a lot. Here are some of our favourite books - our very personal top five specialist books on the subject of artificial intelligence.
Martin Bäumler recommends Nick Bostrom: Superintelligence
Nick Bostrom is a Swedish professor at the University of Oxford and a co-founder of the World Transhumanist Association, and his book Superintelligence presents a speculative warning. If people don’t think about how intelligent robots and computers could be programmed with moral thresholds, we run the risk of being superseded by these machines. This is subject to the current developments in artificial intelligence resulting in it being possible for computers to develop independent activities in the future. It is also subject to this independence being contrary to human interests.
However, Bostrom suggests that this would result in an “intelligence explosion”. This happens as soon as humans are no longer able to control computers’ independence. At the moment, this is moving out of the realm of science fiction, as companies such as Google&Co. currently take approaches to develop artificial intelligence akin to kids playing with dynamite. Experts believe that at the latest in 2075 machines will have been developed that can do just that: Optimise and further develop themselves without humans being able to influence this.
In order to prevent this risk, Bostrom has developed several hypotheses on how this could be combated in advance. He believes that “indirect normativity” could be one possibility: Computers develop their own ethics, which cannot contravene human needs. Bostrom observes the scenarios for artificial intelligence with a long-term horizon: From brain slicing to theoretical super intelligence. An exciting read.
Christian Beckmann recommends Sebastian Raschka: Python Machine Learning
Python is a much-loved programming language for machine learning. Python Machine Learning written by Sebastian Raschka (2015) is an introduction to programming machine learning models: An excellent book for a practical introduction to the subject of machine learning with Python.
The author’s initial objective is for programmers to recognise which of the different models for machine learning are best for the respective data. There are instructions in this regard, showing how neural networks can be built from the Python Keras and Theano libraries. Raschka also shows how clean code should be written in order to optimise algorithms, and how the programmed model is to be implemented into the Internet application.
The machine learning models can predict target results for a Web site using regressive analyses, discover behavioural patterns using cluster analyses and effectively evaluate data via pre-processing – a pre-selection of relevant information. Sentiment analyses can also be applied in order to take a deeper look at social media data. In doing so, the book uses a large number of open-source libraries that have been created in the past few years, and which programmers can use to gain rapid entry to machine learning.
Christian Beckmann recommends Ian Goodfellow, Yoshua Bengio and Aaron Courville: Deep Learning
Deep Learning from Ian Goodfellow, Yoshua Bengio and Aaron Courville was published in 2016 and gives beginners an initial introduction to this subject. Deep learning refers to methods which optimise artificial neural networks. This process uses a cascade of many new layers of processed information which are passed on to the next respective layer. These are increasingly abstracted these until they reach a visible output – the final layer. No human intervention is required in these processes, as the computer learns complex concepts from the hierarchical interaction of simple concepts.
The three authors have broken the book down into three sections: They initially introduce algebra, probability, and information theories as well as machine learning, and then in a second section they present the deep learning techniques used in industry. These include feedforward (non-circulatory) networks, optimisation algorithms, convolutional neural networks, and sequence modelling. They go on to explain practical areas of application: natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and games.
In the last section, the authors explain advanced areas of research. The seven chapters show various theoretical models from auto-encoders through to Monte Carlo simulations and the Deep Generative Model. Readers can find both practical information as well as theoretical insights into this issue.
Erik Meijer recommends Peter Norvig and Stuart J Russell: Artificial Intelligence: a modern approach
Artificial Intelligence: a modern approach is now in its third edition and is a compendium on artificial intelligence written by Peter Norvig and Stuart J Russell that every ML/AI expert should know. This book is almost 1200 pages long and offers an introduction to theory and practical work in all areas of computer development that have to do with the learning behaviour of and for machines. The first edition was published in 1995, the third in 2009.
Seven sections explain the topic. The first, “Artificial Intelligence”, describes interfaces which can take intelligent decisions and thus offers an introduction to the subject. The second section is entitled “Problem-solving” and describes methods that allow pre-emptive actions to be developed, for example for chess computers. Section three deals with “Knowledge, reasoning and planning” and presents how knowledge can be presented in an AI environment, and how this knowledge can be logically further developed. It is followed by the fourth section, “Uncertain knowledge and reasoning”, with a continuation: This deals with uncertain conditions for logical methods. The fifth section discusses machine learning: How can findings be generated in decision-making modules? This section also presents the artificial neural network. The sixth section investigates communicative elements: How do machines perceive their environment as a result of touch or sight? The final section deals with AI’s past and future and rounds off the observations with ethical and philosophical questions.
Erik Meijer recommends Toby Segaran: Programming Collective Intelligence
Toby Segaran’s book Programming Collective Intelligence dates back to 2007 and deals with so-called data mining and machine learning algorithms for online behaviour and user-generated content in the programming language Python. It deals with evaluating larger quantities of data to evaluate trends and interconnections. This book offers instructions for Web programmers to obtain data information from Internet users. I can warmly recommend it to anyone working in the fields of ML and AI. Search machines, rankings, and social bookmarking such as the Web site del.icio.us – what information can be used and how can it be evaluated? How are filters built that allow online retailers to present the right adverts or purchase proposals to their Website visitors? How can a program further extrapolate on knowledge it has obtained? How can this be optimised, filtered, or classified? Segaran offers concrete codes for these forms of machine learning.
Segaran finally discusses the issue of Genetic Programming in chapter 11. This is a form of machine learning based on biological theories of evolution. Programs are set for problems with pre-defined parameters. The best algorithms are selected from the resulting competition and modified: Either as a mutation or a hybrid using another algorithm, allowing new, ever better programs to be created. In order to test the genetic programs, these have to prove their worth time and time again. For example, in the case of games, they could be improved if they initially face an equal opponent and only face real persons in later replays, when they can already play well.
Olaf Erichsen recommends Pedro Domingos: The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World
In his book The Master Algorithm which appeared in 2015, Pedro Domingos describes his vision of a single algorithm which subsumes all of the other learning algorithms for intelligent systems. At present, we have learning systems in various areas – credit cards, search engine runs and other methods, and in the future there will be only one.
Domingos, a professor at the University of Washington, describes the history of learning machines and how the concepts have developed using various channels through to the present day. In doing so, he uses plain language – this book is also geared towards interested newcomers to the subject.
The author uses geographic metaphors in order to explain what is still missing in developing this artificial intelligence. The Master Algorithm is the head office in a city of symbolists, connectionists, evolutionaries, Bayesians and analogizers. Symbolists believe that they can learn from manipulating symbols. Connectionists want to recreate the mind. Evolutionaries write Darwinistic programs. Bayesians thinking is based on Thomas Bayes, that everything we have learned is uncertain, while analogizers make predictions using comparative codes. Domingos uses the examples to present what these respective regions have contributed to technical developments, developing his hypothesis of a master algorithm.
Private life is also mentioned, briefly, at the end of the book. Domingos proposes data unions that monitor personal data. This should form a counterweight to the all-powerful governments or companies.
Jane Trümner recommends Ethem Alpaydin: Introduction to Machine Learning
Introduction to Machine Learning by Ethem Alpaydin, professor in IT at Istanbul’s Boğaziçi University, is an introduction to this subject and was published for the first time in 2004. I was given a copy as a gift when I joined the eLIZA project.
Machine learning means computer programs capable of using sample data or past experience to solve problems. Many applications are already being successfully used: Facial or speech recognition for example, or optimised robot behaviour which only uses few resources.
This introduction to machine learning, explicitly mentioned in the title, approaches the issue taking a wide view. It discusses statistics, sample recognition, neural networks, artificial intelligence, signal processing, control, and data mining. All learning algorithms are explained so that the student can easily move from the equations in the book to a computer program.
Alpaydin deals in detail with Bayes decision theory, parametric methods, multivariate methods, dimension reduction, clusters, non-parametrised methods, decision trees, discriminatory function, multilayer perceptrons, and hidden Markov models. Finally, classified algorithms, multiple learning, and reinforcement learning are not neglected. This book’s strengths are in its portrayal of the statistical aspects.
The original version was published in English and a third, extended version was published in 2015 which also discusses recent developments. This book is regarded as being an easy to understand standard reference work around the world, and a German translation was published in 2008.