The Creativity Code - Marcus Du Sautoy

A
Literary Saloon
&
Site of Review.

Trying to meet all your book preview and review needs.

Contents:
Main

the Best

the Rest

Review Index

Links

to e-mail us:

support the site

the complete review - mathematics / artificial intelligence

The Creativity Code

by
Marcus Du Sautoy

general information | review summaries | our review | links | about the author

Title:	The Creativity Code
Author:	Marcus Du Sautoy
Genre:	Non-fiction
Written:	2019
Length:	287 pages
Availability:	The Creativity Code - US
	The Creativity Code - UK
	The Creativity Code - Canada

UK subtitle: How AI is learning to write, paint and think
US subtitle: Art and Innovation in the Age of AI

- Return to top of the page -

Our Assessment:

B : interesting examples; fine state-of-AI overview

See our review for fuller assessment.

Review Summaries
Source	Rating	Date	Reviewer
Financial Times	.	1/3/2019	John Thornhill
The Guardian	.	12/3/2019	Jonnie Wolf
Nature	.	19/2/2019	Barbara Kiser
The NY Times Book Rev.	.	5/5/2019	Rachel Riederer
The Spectator	.	20/4/2019	Steven Poole
Sunday Times	.	17/2/2019	James McConnachie

From the Reviews:

"Du Sautoy suggests that machine learning techniques are becoming so good that, in some fields, it might be possible to attain transformational creativity. (...) Du Sautoy largely debunks the myth of the lone genius. (...) The increasing pervasiveness of algorithms needs to be carefully handled, Du Sautoy concludes. But if we can get that right then our powerful new tools can significantly enhance the human code. " - John Thornhill, Financial Times

"Du Sautoy’s discussion of computer creativity in the arts is fascinating but the computer art itself is underwhelming. (...) While most of the chapters are given to a discussion of computer art, The Creativity Code is at its best when Du Sautoy, a mathematics professor at Oxford University, discusses his own subject." - Jonnie Wolf, The Guardian

"Whatever one makes of du Sautoy’s final verdict, the journey to it is eloquent and illuminating." - Barbara Kiser, Nature

"Du Sautoy, a British mathematician, wants to answer the question: "Can computers be creative ?" He parses the actions involved in creativity -- exploring, combining and transforming -- and reveals the history of A.I. through the turning points in which machine learning has progressed toward these milestones." - Rachel Riederer, The New York Times Book Review

"Spoiler: there is actually no creativity code, and AI can’t yet perform artistic feats that are plausibly human. The interest of this book is really the journey, as our author travels around various labs to be shown the state-of-the-art in machine learning. (...) By the end of this elegantly conceived book, du Sautoy has subtly but fatally pricked the giant PR bubble of tech ‘AI’, while at the same time composing an inspirational hymn to the power of man and machine working in harmony." - Steven Poole, The Spectator

Please note that these ratings solely represent the complete review's biased interpretation and subjective opinion of the actual reviews and do not claim to accurately reflect or represent the views of the reviewers. Similarly the illustrative quotes chosen here are merely those the complete review subjectively believes represent the tenor and judgment of the review as a whole. We acknowledge (and remind and warn you) that they may, in fact, be entirely unrepresentative of the actual reviews by any other measure.

- Return to top of the page -

The complete review's Review:

       The Creativity Code offers an overview of the state of Artificial Intelligence (AI) today, specifically its applications to activity we consider 'creative'. Mathematician Marcus Du Sautoy's focus here is on the question of whether or not computers can, or will be able to, produce what humans can that goes beyond the familiar and formulaic -- art, for example, or not just calculating mathematical problems but devising proofs of theorems. Does it come down to code -- essentially, an algorithm --, even in humans (the 'creativity code' of the title), or is something more -- a uniquely human attribute, or at least consciousness -- required to produce actual art ?
       Du Sautoy looks at a variety of examples of what AI can already do, and while many of these are familiar from recent media reports it's still a useful and interesting tour, often with helpful insights. He begins with games: with their strict and limited rules and clear objectives, they're ideal proving ground for computer programs. It's been repeatedly demonstrated that, for many games, vast computing power (and the right algorithm ...) is sufficient to play at and beyond human capabilities. Tic-tac-toe (noughts and crosses), with its limited possible moves, was easy to crack (in particular because, as is easy to figure out, optimal play invariably leads to a draw); chess was already a much more complicated problem, but was also cracked years ago.
       Because of the far, far greater number of possible moves in the Chinese game of Go -- "the complexity of Go makes it impossible to analyze the tree of possibilities in any reasonable timeframe" -- it was long thought to be beyond computer reach. The key to cracking it came with the approach: not pure number-crunching, considering all the possible moves -- the long prevalent way of employing computers, since that seems to be where they have such a great edge over humans, able to crunch numbers so much more quickly --, but rather programming the computer to figure it out for itself. As Du Sautoy writes of AlphaGo developer Demis Hassabis:

His idea was that, rather than try to write the program himself that could play Go, he would write the meta-program that could write the program to play Go.

       Instead of taking a purely top-down approach -- a program that is essentially a decision tree, with fixed rules set by the programmer -- the approach was essentially bottom up: "allowing the algorithm to create its own decision tree based on training data". Unleashed on the game of Go, AlphaGo essentially began with nothing but the rules and objective, and then played itself, again and again, learning -- by trial and error -- along the way. Since modern computers are incredibly powerful, it was able to play -- and learn -- a lot; stunningly, it fairly quickly became better than any human player.
       Du Sautoy describes this very well, as well as its implications. While Go is based on the simplest of rules, it is a game of great complexity; among the interesting observations Du Sautoy makes is that the game seemed to have been in a bit of a rut when AlphaGo came on the scene: human play was at a high level -- but perhaps at just a local, rather than absolute peak. AlphaGo's self-learned approach to the game actually suggests new vistas for the game -- though that perhaps can't quite compensate for the fact that the computer is clearly better than humans, who will never catch up .....
       The sheer power of modern computers, and the amount of data available to them, have radically changed AI. Machine learning has long been viable, but obviously benefits tremendously from these -- essentially, it now has much, much more to digest, and the ability to do so much faster. ("Humankind now produces in two days the same amount of data it took us from the dawn of civilization until 2003 to generate", he notes (alas, without any attribution or accounting of/for the numbers); it is indeed a whole new world -- or many, daily.) For purely rule-based tasks -- like playing a game like Go -- this leads to quick success, but other tasks still prove challenging. Visual identification, for example, is something that the human mind is adept at but that computers have a very hard time with.
       Du Sautoy is particularly interested here in computer programs' creative potential: can algorithms be written that create art ? Recognition of what art is -- what we look for in a painting, or want from a piece of music -- is important, and machine-learning has made great strides in this area. Feed in enough information about a painter's work or about chart-topping pop tunes, and programs can already imitate and recreate these reasonably well. Music -- the most straightforwardly rule-based of the arts -- seems to be the one where the greatest strides have been made, with Du Sautoy citing numerous examples of compositions that fooled/convinced audiences. Painting is already more of a challenge, but here too imitation (in the broadest sense) can lead to striking results, such as The Next Rembrandt project he discusses.
       Writing -- texts -- are a greater hurdle, though as Du Sautoy points out, data-driven articles such as corporate earnings summaries are already often churned out by algorithm. Indeed, fact-based texts lend themselves to some automation -- with Du Sautoy admitting he handed over a small sliver of his duties in this very book to such a program:

A 350-word section of the book was written by an algorithm that specializes in producing short-form essays based on a number of keywords that you supply. Did it pass the literary Turing test ? Did you notice ?

It's quite a leap from that to successful automated story-telling -- but mathematician Du Sautoy also states: "I think storytelling is actually the closest creative act to proving theorems", and considers how successful computers have been in his own creative field; unsurprisingly, he's particularly good in these parts about his own area of expertise.
For all the incredible things that computers can do, Du Sautoy does reminds us at the end that:

At the moment, all the creativity in machines is being initiated and driven by the human code. We are not seeing machines compelled to express themselves. They don't seem to have anything to say beyond what we are getting them to do. They are ventriloquist dummies and mouthpieces serving our urge to express ourselves.

       He argues -- and this seems correct -- that it will take a machine becoming conscious for computers to act -- or rather create -- on their own (something beyond the human experience). Which is of course where it will get really interesting, as the question become what AI consciousness will look/be like .....
       The Creativity Code is a book of the here and now. Remarkable strides have been made, especially in very recent years, and Du Sautoy offers a very good survey-overview of the present-day state-of-the-field. He does speculate some about what might be possible, but overall remains almost surprisingly grounded in what is currently possible (and what can be extrapolated from the currently possible). Much of this is quite interesting -- and much is also quite familiar, from media reports (including, for example, IBM Watson's Jeopardy ! appearance); Du Sautoy packages it all together and offers an interesting perspective (and experiences) in part, but doesn't really seem to dig that much further.
       Somewhat disappointingly, there are few illustrations with the text (at least in the US edition; I haven't seen the UK one): some diagrams and a few helpful musical samples but no reproductions of any of the many paintings (and painting-approaches) he describes, which certainly would have been welcome. They're easy enough to find online, but it certainly would be easier to have them right with the text; in-text pointers -- well, there only appears to be one: "You can view the images yourself at https://arxiv.org/abs/1706.07068." -- aren't the most reader-/user-friendly of approaches. (One of the interesting questions regarding these advances Du Sautoy mentions is that of copyright -- who holds it for these algorithm-created works of would-be art ? Maybe that was an issue regarding including them in the book ?)
       Perhaps because it is meant to be a more text for a general reader (though published by a university press in the US), there is practically no supporting apparatus -- no foot- or endnotes, or attributions for citations or data. A 'Selected Bibliography' is useful for those looking for additional information, but notes would certainly have been helpful throughout.
       The interesting examples and Du Sautoy's admirably clear discussion make The Creativity Code a good and interesting read -- though throughout one wishes he were more willing to speculate as to what the future might hold.

- M.A.Orthofer, 23 April 2019

- Return to top of the page -

Links:

The Creativity Code:

Fourth Estate publicity page
Harvard University Press publicity page
Q & A at The Verge
Excerpt

Reviews:

Marcus Du Sautoy:

Faculty page

Other books of interest under review:

See Index of books dealing with Mathematics
See Index of books about the Internet and Computers

- Return to top of the page -

About the Author:

Marcus Du Sautoy teaches maths at Oxford. He was born in 1965.

- Return to top of the page -