The following questions arise: (1) How much more scientific advancement would be possible if more people were involved? (2) Can we integrate what people and computers, respectively, do well? We would like to maximize the effectiveness of this human-computer symbiosis, to find places were computational power is most useful and where human ability can best be applied. People and computers are often good at solving different types of problems; for example, a person would likely translate a passage of text more naturally, and a computer would likely be able to numerically optimize a function faster. We would like to be able to combine the best abilities of each in order to solve challenging problems that neither could alone.
The goal of this book is to determine if it is possible to design the coevolution of human-computer symbiosis to solve currently open problems in science. Two particular areas where humans can excel are spatial reasoning and creativity. People are able to reason spatially by forming mental models of objects, their environment, and the spatial relationships between them [Byrne and Johnson-Laird 1989, Tversky 1993 ].
People enjoy expressing their creativity, and many successful video games require players to think about objects in space and their spatial relationships to each other. Tetris1 is a popular example of this. There are many physical puzzles that rely on spatial reasoning as well, such as Rubik’s Cube2 and many sliding block puzzles. Recently, there has been a rise in popularity of video games whose explicit purpose is to help people train their cognitive skills, such as spatial reasoning. Brain Age3 and Big Brain Academy4 are examples of this.
A scientific field that naturally requires spatial reasoning and creativity for problem solving is biochemistry. Many problems in biochemistry are fundamentally spatial structural problems, particularly when dealing with protein structures. Proteins are important to biochemistry and our understanding of life itself, because they are indispensable to living systems and perform many important tasks in the cell, including structural, transport, defensive, and catalytic roles. The way proteins achieve their function is due to their shapes and how they interact with other molecules. They involve the relationships between physical objects in three-dimensional space; a protein’s structure determines its function [Zhang and Kim 2003 ].
To explore the potential of this human-computer framework for solving scientific problems, we have developed Fold it,5 an online videogamethat casts protein structure manipulation as a puzzle solving competition. The game tries to predict naturally occurring protein structures and to design novel proteins not previously seen in nature. In order to achieve this goal, the game gives players the ability to manipulate and optimize protein structures while competing and collaborating with other players to discover the best structures. Foldit’s YouTube channel can be found at http://www.youtube.com/user/uwfoldit; http://www.youtube.com/watch?v=lGYJyur4FUA gives a good introduction to the game.
1.2 Problem Statement
1.2.1 Game Design Problem
Designing a game for scientific discovery presents many distinct challenges. Oneof the primary purposes of using a game is to maximize the engagement and retention of the players. However, it is not enough to simply make the game as fun as possible; this goal must also be balanced with the need for relevant scientific outcomes. For most games, the designer is free to make decisions based only on what will make the game fun. In a scientific discovery game, the tension between the freedom to design for engagement and the realism of the scientific constraints is a key challenge. Thus, an important question is, how can we design a game that is both engaging and produces useful results? We would also like to know what kinds of problems would lend themselves to such contributions by non-scientists, and how can we identify these problems and map them onto a game.
We presume that the game players begin without any knowledge of the scientific field the game is based in. Given this, we would like for the players to gain the domain knowledge necessary to make a contribution to a challenging scientific problem quickly. This is not necessarily general expertise in the subject area or formal scientific expertise. Players may develop their own specialized form of expertise unique to the problem presentation within the game. How can the game best support the training of players to the point where they can make a contribution, and integrate players into the scientific process? We would like to use structures from games to teach players, and keep players interested and involved long-term. We would also like to spread the expertise gained by experienced players and to help new players learn from it.
The game’s architecture should support the coevolution of both the players and the game itself. In this way, as the players adapt to the game by gaining experience in how the game works and solving the problems presented, the game can also adapt to how the players best use it to become a better tool. How can we allow for this coevolution of the game and the player base?
1.2.2 Biochemistry Discovery Problem
Predicting protein structures computationally is a central goal for computational biochemists because so much can be understood about a protein’s function once its structure is known, and because it is so challenging to observe a protein’s structure directly. Proteins—chains of smaller molecules called amino acids—are central to biochemistry because they are the primary chemical for almost all cellular processes. Understanding a protein’s structure is necessary to understand its functions, because a protein’s shape determines how it will interact with other molecules. Thus, an important problem in biochemistry is the protein structure prediction problem: given the sequence of amimo acids that make up a protein, what is its structure? It is possible to experimentally determine a protein’s structure through methods such as X-Ray Crystallography and Nuclear Magnetic Resonance spectroscopy. Experimental methods, however, can be costly, time consuming, and difficult. This makes computational methods that can accurately predict a protein’s structure an attractive solution. However, computational methods are often intractable; the vast number of possible shapes a protein can take make it difficult to find the correct structure. The spatial nature of this problem makes it a good candidate for the application of human spatial reasoning.
A related problem is that of protein design: given a desired function for a protein, what is the amino acid sequence that, when folded, will carry it out? In this case, computational methods are even more attractive. Synthesizing proteins to test every design would be prohibitively expensive, while computational methods can allow us to filter out designs that are not likely to work. Protein design has implications for drug design, in inhibitors and vaccines, for biofuel design, in enzymes, and for other areas. Human creativity can be applied to help create novel proteins that did not exist before.
1.3 Outline
In this book we will show the effectiveness of the game-based framework as an approach to scientific discovery. Chapter 2 discusses the literature related to this book. Chapter 3 gives an overview of the game-based framework used in this research, describing the dual goals of engagement and scientific relevance, and the coevolution approach we take. A discussion of using this framework for problem solving as applied to protein structure prediction is given in Chapter 4, and we show that players can predict the unknown structures of naturally occurring proteins, even where all previous methods have failed. Further discussion of applying this framework to leverage player creativity for protein design is given in Chapter 5, and we show that players can become an integral part of the design of novel and effective proteins. Chapter 6 describes an approach to allowing players to codify and automate their strategies, and we show that players can socially develop highly effective algorithms. Finally, Chapter 7 provides a summary and discusses possible future directions for research.