The field of art has been attracted to high-end technologies like artificial intelligence (AI) for a long time. There are AI-driven drawing programs, for example. However, on a broader scale, the adoption of technology in the art world has not been high. On the other hand, AI technologies such as deep learning influence almost every area known to us, from shopping and education to healthcare and agriculture. What is the outlook for extending this influence to the arts? Which use cases are promising and which are not? This book explores this topic, looking at both the positive and negative aspects.

A key point presented is that art is not just an image--there is a lot more than meets the eye. The book also questions whether AI can do justice to these elements. Consider the notion of style. The art world talks of different styles of work based on period, geography, or even a specific person. Beyond types of brushstrokes or materials used, the notion of style is much deeper. The art world considers style to be a subjective property; whereas when you bring it within the purview of a computer program, it tends to become an objective factor. Another area of concern is the comparison of artworks, which cannot be just feature based.

Classification is generally child’s play for AI algorithms. How well does that kind of approach work for art? Often, art classification happens based on style--which is itself another loose concept, as mentioned earlier. The commercial value of art has a lot to do with its classification; hence the stakes are high. It appears as though art classification will be too complex to be left to AI-type approaches.

One aspect seriously discussed in the book is art ownership. The usually practiced manual approach is discussed, and then the challenges of using technologies like deep learning are explored. The variability in this regard across human experts is a concern. Looking at detectable features focuses the attention on type of stroke, type of ink, and the like; this may vary across time, even for a single artist. At the same time, there is a lot of interest in an automated, objective solution for this, since art is a high-value market and both fakes and fake claims are not unusual.

These topics are the focus of the book. Overall, the tone of the book is negative, that is, technologies like AI need to be cautiously used in this domain. The book quotes liberally from various authors to support the arguments. However, I found the quantitative analysis and specifics lacking. The discussion goes on, quoting source after source, and tends to wander. A more focused discussion listing the major concerns and approaches, including a critique of these, would have been easier to read and follow.

There are only four chapters in the book. The first chapter, “Introduction: Return to Form,” looks at common challenges in the art domain and relates them to the AI-like technologies in vogue today. It starts by looking at the face image comparison used by Google. Topics like digital humanities, deep learning technologies to redress the challenges of traditional machine learning, and so on, are touched upon. None of these are dealt with in a focused manner, however, leaving the chapter a bit too open. A brief overview of the book at the end would have also been helpful.

Chapter 2, “The Shape of Data,” looks at the art data, characteristics, and so on. Data, undoubtedly, is the king of today’s approach to AI. Chapter 3, “Deep Connoisseurship,” looks at authenticating the ownership of art and the role technology can play there. Chapter 4 concludes the book, summarizing the key observations so far. There are appendices on art styles, notes, and so on.

In summary, this is certainly an interesting topic and there is much to learn. The most relevant audience for this book is those in the art field with an interest in computational applications. It should also be useful for AI/information technology (IT) people looking at applications in the art realm. The book adopts a moderately technical tone throughout, avoiding very formal discussions and mathematical details. That being said, the narrative tends to be longwinded and the flow could be better organized, with practical examples, to make the book more readable.