A complex system approach to address world challenges in food and agriculture

Accepted Manuscript A complex system approach to address world challenges in Food and Agriculture H.G.J. van Mil, E.A. Foegeding, E.J. Windhab, N. Perrot, E. van der Linden

PII:

S0924-2244(14)00157-5

DOI:

10.1016/j.tifs.2014.07.005

Reference:

TIFS 1565

To appear in:

Trends in Food Science & Technology

Received Date: 29 August 2013 Revised Date:

11 June 2014

Accepted Date: 6 July 2014

Please cite this article as: van Mil, H.G.J., Foegeding, E.A., Windhab, E.J., Perrot, N., van der Linden, E., A complex system approach to address world challenges in Food and Agriculture, Trends in Food Science & Technology (2014), doi: 10.1016/j.tifs.2014.07.005. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Title:

A complex system approach to address world challenges in Food and Agriculture. Author names and affiliations: H.G.J. van Mila, E.A. Foegedingb, E.J. Windhabc, N. Perrotd, E. van der Lindena,e a

M AN U

SC

RI PT

TI Food and Nutrition, Nieuwe Kanaal 9A, 6709 PA Wageningen, The Netherlands, [email protected], [email protected] b Department of Food, Bioprocessing, and Nutrition Sciences, North Carolina State University, Box 7624, Raleigh, NC 27695-7624, United States. e-mail: [email protected] c Laboratory of Food Process Engineering, Institute of Food, Nutrition and Health, Department of Health Science and Technology, ETH Zurich, Schmelzbergstrasse 9, 8092 Zurich, Switzerland. e-mail: [email protected] d UMR782 Génie et Microbiologie des Procédés Alimentaires, AgroParisTech, INRA, 78850 Thiverval-Grignon, France. e-mail: [email protected] e Laboratory of Physics and Physical Chemistry of Foods, Wageningen University and Research Center, Bornse Weilanden 9 (building 118) 6708 WG Wageningen, The Netherlands. e-mail: [email protected]

AC C

EP

TE D

Corresponding author. E. van der Linden. e-mail: [email protected] Phone: +31 317 485515

1

ACCEPTED MANUSCRIPT

Abstract

M AN U

SC

RI PT

The quality and amount of the world food supply is crucial to the well-being of every human on the planet in the basic sense that we need food to live. It also has a profound impact on world economy, international trade, and global political stability. The choice of land used for agriculture, and the livestock and plants raised on the land, will impact the sustainable use of global resources. On a global scale, there are communities where insufficient food causes nutritional deficiencies, and at the same time, there are other communities eating too much food leading to obesity. Both conditions have accompanying diseases with associated financial consequences. The above issues relate to various scales, from local to global, and to a range of scientific disciplines. Moreover, their various elements are part of an interdependent, continuously changing, and adaptive system. This implies that the response of a combination of individual elements cannot usually be inferred from the response of each individual element or subsystem. This makes the identification of an appropriate intervention to change one or more elements a complex problem. We propose that a complex system approach should be used to address the global challenges of the agriculture and food system. The complex system approach accounts for the needs of stakeholders and policymakers in the agriculture and food system, and helps to determine which research programs will enable stakeholders to better anticipate and respond to emerging developments in the world. Moreover, the approach will enable them to determine effective intervention strategies to simultaneously optimise and safeguard their interests and the interests of the environment. The approach is formulated in terms of a roadmap of procedures. It encompasses an array of methods utilised in an integrative multi-scale and inter-disciplinary way.

1. A qualitative description of the problem and a solution Our global society faces several grand challenges such as the eradication of poverty and diseases and providing adequate food production. Expanding on the latter, the combined agriculture and food (Agri&Food) enterprises have the overarching goal of continuously providing safe, tasty, healthy, affordable, adequate, and sustainable food production.

EP

TE D

Food production, by the growing of plants and animals and by the conversion of raw materials into food products, is part of an Agri&Food system that interacts with various other systems in the world. For example, the Agri&Food system is linked with the economic, health-care, and socialcultural systems. All of these systems are continuously changing, self-organising, interdependent, and adaptive, or, in other words: they are all ―complex systems‖. This has been illustrated more precisely by professor Allen (Allen, 2012)(Allen & McGlade, 1987)(Allen & McGlade, 1987)(Allen & McGlade, 1987) ―(Complex) Systems evolve qualitatively over time and interact with each other. For example, the human body has adapted to food and in turn the food has been adapted to it. The food we consume is governed by culture and lifestyle, but also by climate, soil and energy issues. All these aspects are in a continuous state of flux‖.

AC C

It turns out that the identification of an appropriate intervention to change one or more factors in a complex system, such as the food production system, is a problem with many complications. Firstly, because of the continuously changing, self-organising, interdependent, and adaptive character of all the systems involved, a response in one system usually cannot be inferred from responses of individual factors that are present at more detailed levels in that system. Secondly, an intervention in one system may cause changes in other systems, which reversely may induce changes in the original system (i.e. cause non-linear responses). These two characteristics illustrate the fact that (Allen, 2012) ―one needs to be aware that our predictions do not necessarily come true‖. In fact, the identification of an appropriate intervention is an evolutionary and adaptive problem that usually will not have a single and unique solution. The best achievable goal is to articulate strategies towards solutions instead of formulating the solution. Indeed, we have to take into account the inherent uncertainties in complex systems, and we need to address how to obtain optimally accurate predictions on the basis of limited and widely dispersed information (Jaynes, 2003). We need to abandon reductive approaches that imply one solution, and instead acknowledge the necessity of a non-reductionistic approach that is integrative over different levels of detail. All these aspects are part of a ―complex systems approach‖. 2

ACCEPTED MANUSCRIPT We propose a systematic methodology that takes into account the different levels of detail and multiple disciplines in an integrative fashion. We start with describing in more detail a typical food process to illustrate the concept of complexity. We then describe the complexity involved in agri&food challenges in general, and how to address them in an integrative (complex systems) approach.

2. Bread making as an illustration of complicated and complex

TE D

M AN U

SC

RI PT

Let us consider the age-old practice of bread making. Producing bread with acceptable properties is complicated because there are many steps that can be sub-optimal or simply turn out wrong (kneading time, temperature control, time in the oven, relative ingredient composition, forgetting emulsifiers, forgetting yeast, and others). Describing the bread making process is complicated because of 2 main aspects involved in bread making. One is that the critical processes are in part interdependent and occur at different length scales (gluten formation, starch dissolving partially, yeast yielding carbon dioxide, bubble expansion, bubble interface being covered by a mix of different surface active molecules, fibres interfering with bubble expansion, extension of gluten network, crust formation, water vapour release, drying of outer part, browning reactions, flour composition, and more). The other is that the processes that must occur to produce bread are founded in different scientific disciplines (physics, chemistry, and biology). In short, the description of the bread making process is complicated because it involves forming specific molecular interactions to produce gluten (chemistry), gas generated from yeast (biology), and heat and mass transfer (physics) during baking. The complicatedness lies in the different length scales and disciplines involved. Although there is uncertainty regarding, for example, the position of a given molecule, such uncertainty is not relevant for the properties at a larger scale. We can say that the bread making process remains tractable when one has high enough accuracy to describe each step. The tractability lies in the control of the environmental circumstances like time, temperature, humidity in the oven, and other factors. As science has progressed over the centuries, bread making has been increasingly automated due to 1) understanding critical properties that need to be controlled and 2) advances in engineering that replace human work and judgement with machines.

AC C

EP

Let us now consider the bread making in the context of its environment. If an environmental condition during the process, for instance the composition of flour, is intractable, the bread making process itself becomes intractable. This intractability of flour composition can be due to, for instance, uncontrollable weather conditions during the growth of the wheat. Or it may be due to the fact that the choice of type of flour is controlled by the baker responding to the price of the flour, which can be determined by unpredictable factors like a bad harvesting year. Or by adaptation of the baker to a diminishing bread market, for instance caused by a suddenly emerging popular diet that preaches to avoid eating carbohydrates. The baker may restructure the baking process to provide gluten free bread for people with celiac disease or to lower the salt level for health reasons. For economic reasons, the process could be scaled up to a larger daily output to increase profits. This scaling up in turn may affect bread quality and thereby acceptability to customers, jeopardising their economy of scale strategy and profits. All these examples of intractable situations are complex. The considerations regarding the influence of the environment on the system that is set up around the bread making process illustrates some important concepts of complexity (see also box). These concepts are generally present in systems whose behaviour is intractable (i.e. a complex system). First of all, the environment rules much of the bread making process. The environment is composed of many systems (like that related to the weather, economy, or health), which have an interdependency with the existing bread making process. All these other systems show a continuous dynamics and are interdependent with yet other systems that are also in a continuous state of dynamics. The bread making process adapts and re-organises to these changing circumstances by means of an actor, i.e. the baker. The overall system around the bread making process, including the baker, constantly self-organises to adapt to changing circumstances. The bread making process that exists at a given point in time is a structured sequence of steps that has 3

ACCEPTED MANUSCRIPT

RI PT

evolved over time by accommodating the requirements of the consumer. This process continuously adapts to the changing environment. The adapting process affects the environment (for instance a cheaper product is bought more by the consumer and therefore the baker will order more ingredients from his supplier, which will change the buying behaviour of the ingredient supplier). The possible ways of the baker to intervene are many, and there is not one best intervention. The baker can at best develop a strategy and monitor the responses of their surroundings. The bread making process has evolved over centuries. It has co-evolved with changes in the existing circumstances. This co-evolution makes the process intricately connected with the surrounding systems. This explains why trying to produce a specific type of bread with vastly different ingredients (due to the requirement of having to be able to be flexible in ingredient usage) while keeping the taste of the bread constant, is like trying to produce a car of the 21st century with only plastic. Describing a complex system is possible to a certain extent. However, designing it from scratch with all the desired properties, in this case the cost of bread, is not possible due to connectivity of un-controllable variables. The price of wheat will remain constant only for a finite time.

M AN U

SC

We would like to note an aspect regarding knowledge across various scales. The evolution of the bread making process during the 16th century in an old city, say Leiden in the Netherlands, did, at that time, not require detailed knowledge of molecular properties. Therefore, variations in the bread making process and bread quality associated molecular properties were not understood and the adaptations were phenomenological and based on astute observations. With the growing knowledge at the molecular level regarding ingredients in general, one also experiences more incorporation of knowledge on the molecular scale into the bread making process. This helps in controlling the process and nutritional properties of molecules. The improvement of a once established best strategy will involve more refined knowledge, including more knowledge at the smaller length scale.

TE D

The environmental circumstances of the bread making process are part of the Agri&Food system. Also within this system many interconnections exist, both with scales (spatial-temporal) and disciplines (conceptual domains). In addition, one has many interconnections with other systems such as the ecological, economical, transport, and healthcare systems. We take a pragmatic pluralistic approach to the problem taking into account the interdependencies within the Agri&Food system.

3. Interdependencies within the agri&food system: different disciplines and scales

AC C

EP

The description of the Agri&Food system may be started by the two main sources of food, i.e. of animal and plant origin. Both categories can deliver ready to use products, like milk, eggs, and fruits (see Figure 1). These ready to eat products may also be used as ingredients for other products (milk for cheese, egg for a cake, apple for apple juice). The more ingredients the more complicated the recipes become. Quite complicated products can sometimes have dozens of ingredients and tightly controlled preparation steps. One can construct a network of ingredients connected to recipes, yielding interesting information on how some ingredients are connected to many recipes, while others are only connected to a few. Such a network provides a way of structuring the information regarding interdependencies between products, ingredients, and flavours. See (Ahnert, 2013) and also (Ahn, Ahnert, Bagrow, & Barabasi 2011) for an analysis of how certain ingredients and flavours often occur in the same combinations. As we have seen in section 2 for the bread making process, the steps of the process are connected to many aspects outside that process. This also holds for the final product. The product should have desirable taste, smell, texture, and should have nutritional value by which it contributes to our wellbeing. In this way, the product properties connect to the processes during eating and digestion. During digestion, the product is broken down to molecules seen on the scale of nanometers. The connections between product, its perception, and nutritional value require 4

ACCEPTED MANUSCRIPT

connections between macroscopic and molecular scale. They also require connections between disciplines such as material science, life science, and behaviour science. The important aspect of food acceptability depends on decisions made by consumers. These aspects relate to the social sciences and relate to phenomena on large scales, such as those of communities of people. The following sub-sections provide general information on scientific disciplines and associated scales in relation to our approach. It should be noted that these are broad classifications. Some of the characteristics of each of the disciplines were also briefly addressed by Wolfram (Wolfram, 2002).

3.1 Material sciences

AC C

EP

TE D

M AN U

SC

RI PT

Starting at the material science level, in the lower left part of Figure 2, food can be seen as the result of a process where biological molecules assemble into food structures over a period of time as a function of molecular composition and concentration, externally applied stresses, and internal stresses/molecular mobility. Stress in this case can be of different natures, e.g. mechanical or electromagnetic. This 4 dimensional representation is somewhat similar to the idea of principal axes in a jamming state diagram as introduced by Liu and Nagel (Liu & Nagel, 1998). The two axes, on external stresses and internal stresses, reflect a distinction between the food system and its environment. We note that from an engineering point of view, it would be more natural to consider applied power per volume instead of applied energy per volume (i.e. stress), but in view of symmetry with respect to the axis for internal stress/mobility, we have chosen for stress. The resulting structure can be described at different levels. Starting with individual molecules at the nano-scale, they are assembled into a range of structures that are characterized at the meso-scale. The meso-scale structures similarly contribute to the macrostructure. All or only specific structural levels are important for attributes describing food products. For example, aroma intensity profiles are molecule based, while graininess (a textural attribute) is based on the mesostructure and macrostructure levels (one only perceives grains larger than around 20-50 micrometer in diameter). Following Crutchfield (Crutchfield & Machta, 2011; Crutchfield, 2011), this mesostructure embodies the historic information that has been stored within the specific food as it has co-evolved over its specific trajectory throughout the diagram until time t. As such, food structure is a reflection of the historic information acquired, i.e. encompassing the specific route that has led to the food. Recall that in this context, historical refers to the time course of growing or directly assembling molecules into the food (the latter also often referred to as ―processing‖ the food). The specific route is defined by the type of ingredient(s), their concentration, time, externally applied stresses), and internal stress/mobility. This is the case for processed food such as bread, as well as for foods such as fruits and vegetables where the structures are formed by biological processes during growth and ripening. Thus, we arrive at a triangle of interrelationships between structure, properties, and processing. The importance of such triangular interrelationships has been put forward and extensively addressed by one of us (Windhab 2008). The effect of ingredient type, and the inability to have them interchanged without affecting food properties have also been the subject of many papers (see e.g. a paper on the replacement of egg white protein with milk proteins in making angel food cake (Berry, Yang, & Foegeding, 2009; Pernell, Luck, Foegeding, & Daubert, 2002; Yang & Foegeding, 2011). Structure is important for many different fields, such as in designing complex chemical systems using nature inspired approaches (Coppens, 2012). For foods, structural changes are also important while the food it is being stored, transported, or displayed in shops. Structural changes over that history (time) can affect sensory properties as well as change nutrient availability. The data and models at this level are quantitative. Models aim to explain the most probable state given a set of parameters and variables. These models are exemplifications of a theory of physics or chemistry. Although more competing models may exist, there is a widespread belief that one model is the best representation of the mechanism underlying the phenomena. Data to test the models can generally be obtained under controlled conditions, which makes it possible to strictly speak of parameters and control. The perspective on complexity in this field is closely related with the nonlinearity of the phenomena under investigation leading to critical behaviour and emergent properties (Oono, 2013). Here, if the nonlinear relations are strong, individual (chaotic) behaviour of molecules or other objects becomes intractable and is therefore not suited to quantitative analysis over longer time or spatial scales. The probabilistic behaviour of homogenous populations 5

ACCEPTED MANUSCRIPT

of these molecules or objects, however, may be analysed in a quantitative way; critical behaviour and phase transitions are such examples (Sornette, 2006). In our perspective, these problems are called complicated and not complex per se (see boxes). There are however, instances in physics where we can speak of complex phenomena. One relevant example is complex fluids that deals with heterogeneous materials with interactions of different types. With some hand waving, Econophysics can also be considered as a form of complex physics (Helbing, 2008). Within the domain of physics (and complex physics), stability and invariance are still unambiguously defined concepts with precise mathematical representations (Oono, 2013; Sornette, 2006).

RI PT

3.2 Life Sciences

AC C

EP

TE D

M AN U

SC

The life sciences come into play in the selection, evolution, and annual growth of plants and animals determining the composition and distribution of food, and the formation of plant and animal materials into food structures. This also fits within the process-structure-properties triangle (Figure 2). Once structure is formed, it becomes connected with a person through and the breakdown during mastication and assimilation of nutrients in the digestive system; also a Life Science process. The breakdown and assimilation involve the mixture of desirability-related traits (appearance, flavour, taste, and texture) that initiate consumption; and the processes occurring in the gastrointestinal track that determine delivery of nutrients and bioactive compounds. These are related to the macro- and meso-structure dynamics (during mastication, swallowing, and digestion). As such, material properties are related to sensory attributes and nutrient delivery. For sensory attributes, we distinguish texture (mechanics), fracture and breakdown during mastication (mechanics and sound), colour (optics), taste (tastant release and transport to sensors), and smell (volatile release and transport to sensors). As a complicating factor in sensory perception, the senses are known to mutually influence one another due to the rules that the brain uses to integrate and interpret information. For instance, certain sensory attributes of a product perceived via one or more modalities (such as vision and touch) can bias a consumer’s perception of other attributes of that product derived from other sensory modalities into alignment, and consequently, modulate a person’s overall (multisensory) consumption experience (Spence, 2012; Piqueras-Fiszman & Spence, 2012). Once the food enters the gastrointestinal tract, the food starts to co-evolve with the bio-system present there: the food is broken down by its interaction with the digestive system, and this breakdown in turn regulates the digestive system. The dynamics, adaptability, and selforganisation occurring in this environment is staggering. It is remarkable that one still manages to deduce guiding principles that describe the essentials of the system behaviour (Barbara M. Bakker, Karen van Eunen, Jeroen A.L. Jeneson, Natal A.W. van Riel, & Frank J. Bruggeman, 2010; Teusink, Westerhoff, & Bruggeman, 2010). Data and models in the life sciences are both of the qualitative and quantitative types. The state concept, as defined by physico-chemical parameters, is still used to describe system properties like effectiveness, stability, adaptively, and robustness (Csete & Doyle, 2002; Kitano, 2004; Wagner, 2012) and are used in both qualitative or quantitative ways. The mathematical representation of these concepts may differ between authors or are only heuristic. Different models of a particular phenomenon may exist that can exemplify strategies to realize a particular goal; there is not one best model or strategy, as a strategy is conditional to environmental states and/or correlated to the history of the internal states of the organism or population. The life sciences are possibly the best exemplification of a discipline dealing with complex systems per se, so much so that a subfield in the science of complexity has emerged called bio-complexity. The application of science of complexity in the life sciences appears on many scales and conceptual frameworks. The development is less based on theory and more on phenomenological models. Complex relations among large and complex datasets on the one hand, and model building on the other hand shows that the life sciences, in comparison with the material sciences, are more phenomenological. Relations among genetic and metabolic networks to cell or tissue function are based on empirical correlations using advanced statistical techniques. To get an idea of the vast amount of methodologies involved in, for instance the ―-omics‖ domain (transcriptome, proteome, metabolome, and physiolome), the website of the bioconductor project (Gentleman, R., Carey, V., Huber, W., Irizarry, & R., Dudoit, 2005) is a nice illustration (www. bioconductor.org). The concept of complexity can also arise in many other forms: complex interactions between and within gene6

ACCEPTED MANUSCRIPT

and metabolic networks, along with complex interactions between and within the metabolism and resulting physiology, in relation to the environment (ecology). Due to the complex interactions involved, the field of network theory is very popular to visualize and analyze the complex interactions and to discover underlying structures in the information (Alon, 2007; Barabási & Oltvai, 2004).

3.3 Behavioural Sciences

3.4 Social Sciences

TE D

M AN U

SC

RI PT

A person’s response to the properties of food is influenced by higher ordered process-structureproperties relationships in society, such as peer pressure influencing food choice (seen in upper section of Figure 2). The physical and chemical properties of food determine what is sensed by physiological processes, but the sensing will often be modified by previously stored information derived from experience. An interesting coupling exists between structure and behaviour during mastication. The structure may be broken down during mastication in a specific way, leading to a specific perception of texture. Reversely, this perception determines how we masticate (behaviour). Molecular absorption during consumption and digestion can also influence eating behaviour, both on a short and longer-term basis. Food liking and resulting buying behaviour are normative reflections based on a complex mix of personal preference within a particular demographic and cultural background; these are partially genetically determined (Bartoshuk, 2000) and by the social interactions within a group and between groups of people. This gets us to the society level. The empirical models in behavioural science like psychology and cognitive science become more and more qualitative and a quantitative description of the state of an individual, in terms of physicochemical parameters, has yet to develop. The ordering of model quality is based on individual preferences and is therefore value laden. Interesting methodologies and theories dealing with the complexity in behavioural sciences have been developed where its borders the social sciences; axiomatic construction of models based on game and decision theory that includes an experimental approach, (French, 1986) and allows for prediction. Some hold the position (e.g. T.Sambrook, A, Whiten, 1997 and Pauli Brattico, 2008), that a computational approach to the behavioural sciences runs into limits if simple reductionist strategies are implemented, and that different sources of complexity will emerge if deep reductionism to the level of physics is attempted (Brattico, 2008; Sambrook & Whiten, 1997). We do not want to enter into the controversies surrounding the hypothesis that behaviour can be reduced to physics but instead will propose a pragmatic and pluralistic attempt to cross these barriers in terms of scales and concepts. But it should be clear that behavioural sciences, although strongly empirical, do have methods to achieve quantitative results and prediction, dealing with the complexity within its discipline.

AC C

EP

At a social level, the preference and liking of groups of persons will depend on their personal preferences, their health, as well as their history and interactions with their peers in that group (culture). Groups (regional, social-economic, language, generation, and more), in turn, can show interactions with each other. All these interactions have social dimensions, conditions alluding to food security, and economic and sustainability aspects, to name a few. For the purpose of modelling, human beings are often considered as agents, analogous to molecules in a material. Their individual properties and interactions with other agents can be characterized as would be done for wheat proteins and starch forming a bread structure. This leads to an interesting analogy between the social science level and the material science level. For the social science level, one dimension is formed by the typical properties of each individual agent, as well as the composition and density within that group of agents (analogous to type and concentration of molecules). Another dimension at the social level is formed by culture, food scarcity, peer pressure, social pressure, and so-called PAN variables (Preference, Acceptance, Needs) (Windhab 2008)). These factors are the analogue of externally applied stresses at the material level. Yet another dimension on the social level is formed by internal relations and personal mobility within the group (the analogue of internal stresses/mobility at the material level). The analogy extends itself to an emerging social structure that embodies the historic information that is stored within the social system as it has co-evolved along the trajectory in the diagram (Figure 2). Also, on this level, the amount of historic information that a system stores may be considered a measure of the (structural) complexity of that system (Crutchfield, 2011). 7

ACCEPTED MANUSCRIPT

Furthermore, analogous to the material point of view, the emerging social structures on this level are determining how the social systems process and store information, i.e. how they are determining the system properties. One of the interesting features at this level is that the actions of the agent at this higher level (i.e. of the human being) have cognitive and psychophysical determinants that are different in risky and non-risky contexts (Tversky & Kahneman, 1983), and the agents have judgements that are subject to heuristics and biases (Tvenky & Khaneman, 1974).

SC

RI PT

Other aspects at the social level are of an economic nature, including product development processes and innovation processes (Bonabeau & Meyer, 2001; Frenken, 2006). Some recently developed tools to address and illustrate emerging structures at that level have been described (Chavalarias & Cointet, 2008). An interesting example is the optimisation of cheese ripening, which includes aspects on quantitative material science and qualitative knowledge of process operators (Barrière et al., 2013; Baudrit, Sicard, Wuillemin, & Perrot, 2010; Perrot, Trelea, Baudrit, Trystram, & Bourgine, 2011). The role of computer science and the description of digital systems for business conduction is nicely addressed (Razavi, Moschoyiannis, & Krause, 2009). In regards to issues of economy and food scarcity, some important aspects have been addressed recently (Lagi, Bertrand, & Bar-yam, 2011). In regards to a living system example, the ecology of fish populations serves as an example (Allen and McGlade, 1987).

AC C

EP

TE D

M AN U

At this level, models and data form a complicated mix of quantitative and qualitative properties. The behaviour of populations can be abstracted, since the probabilistic parameters are amenable to mechanistic modelling. The models were however, qualitative in nature; at short time scales the models might achieve quantitative results, but at longer time scales the models are used more qualitatively. Recent trends in quantitative socio-dynamics however, makes it possible to achieve a higher degree of quantitative precision (Helbing, 2010). It's interesting to note that formalisms developed at the smallest scale, in physics, have been mapped to the largest scale of interest, the social sciences, by a field called econo-physics. Again, we do not want to enter into the discourse if social sciences can be reduced to physics, but effective use of a common mathematical formalism and statistical theories of interaction is interesting, to say the least. The question is open to what type of invariance would be of interest. We should be careful not to naively map the result of one discipline onto another. The conceptual relations and generated meanings within physics and within the social sciences are just too different from one another. As a result, much of the observed complexity is attributed to non-linearity, and analogies are concluded with concepts like critical behaviour, phase separation, and stability. Testing these models is much more difficult and expensive (Helbing, 2008) as compared to most experiments in physics (apart from the large colliders or other research infrastructures used in physics). The data can be very complex, finding their representation in large databases, which is not always straightforwardly linked to a model. Again, complexity arrives here in different guises.

8

ACCEPTED MANUSCRIPT Complicated-complex For a complicated system its behaviour is tractable in terms of principle parameters. For a complex system its behaviour is not tractable.

TE D

M AN U

SC

RI PT

What is a complex system? A complex system can be defined as a system that consists of parts that are interrelated and from which one cannot infer the behaviour of that system. Partially this is caused by the fact that the system interacts with other such systems in its environment, forming an open system that is sensitive to changes of its environment. A measure of complexity can be defined as “the amount of information necessary to describe the system” (Bar-Yam, 1997). Many typical characteristics of a complex system can be mentioned (Bar-Yam, 1997). One is the existence of different spatial scales where the overall system behaviour needs a multiscale description. For a human being for example, the smallest relevant scale may be the molecular scale, at which molecules assemble into structures that together form the larger scale, i.e. that of the cell. The cells in turn are assembled into organs, muscles and bones, which together are assembled into parts of the human body. The behaviour at the scale of a cell influences behaviour on the larger scale of an organ, and reversely, stresses at the scale of the animal affect processes on the smaller scale of the cell. The properties of the cell require a different terminology than the properties of the entire human. Many different (meta-) stable states are possible. Interdependencies within one system occur because of interactions among the parts that exist at the same or different spatial scales within the system. A popularised version on the early history of complex systems research can be found in Waldrop (Waldrop, 1993).

AC C

EP

Characteristics of complex systems - the systems are part of the continuously changing, self-organising, interdependent and adaptive systems in our world. - the relevant factors determining their behavior are in a continuous state of flux - the response of a combination of factors cannot be inferred from the response of each individual factor - the factors exist on different time and length scales and refer to various disciplines - a reductionist analysis is incomplete - ways to intervene for changing the systems should preferably articulate strategies towards the desired change, instead of trying to formulate the best intervention

9

ACCEPTED MANUSCRIPT 4. From reductive approaches for complicated systems to nonreductionist approaches for complex systems

TE D

M AN U

SC

RI PT

A multi-level and multi-functional approach has been used in many existing investigations. For example, in formulating the relation between the fine structure of a surfactant molecule and the phase behaviour of a surfactant/water/oil system. Another example is the description on how a protein in water forms a gel system as a function of temperature, pH ,or salt concentration. Different levels, such as protein unfolding and aggregation at the nano-scale, and overall gel rigidity at the macro-scale, are taken into account. Also, different time scales, like protein unfolding and aggregation kinetics, are considered. In these spatial and temporal scales approaches, the environment is constant or changing slowly relative to the other relevant parameters, at all scales. If the problem is complicated and not complex, a ―one model fits all‖ may well apply. One can deduce a simple solution after careful analysis of a problem that is complicated at first sight. For example, the problem of a high number of people having goiter in the early 20th century was eventually solved by adding iodine to table salt (practiced in the United States of America from 1924 onwards). The problem for anybody who had goiter was deficiency of iodine, and the simple solution, for everybody, was increasing the amount of iodine in the diet by adding it to a food ingredient that everybody consumes (table salt). Similarly, the lack of vitamin C causes scurvy, which can be easily cured by administering vitamin C and prevented by regularly eating food that contains vitamin C. In comparison, a problem like obesity, which is caused by many different factors that are part of different environmental systems, becomes complex (see also section 2). For an overview of factors of obesity see (Murphy, 1960), for ingredient effects (Meydani & Hasan, 2010), and for oral processing effects (Rolls, 2011). Other topics that are amenable to the ―complex systems‖ approach are, for example: achieving robust food-supply systems, meeting sustainability criteria, balancing preferences between energy/water/food and food/fuel/feed, anticipating urbanisation, balancing food production (locally and globally), ensuring ingredient formulation flexibility in relation to sustainability and health considerations. On the level of a person, an example is understanding how oral processing contributes to sensory perception and liking in terms of brain activity, healthy aging, and achieving a healthy diet. This would also have cultural and other influences.

AC C

EP

A solution to problems like obesity requires multi-level interventions and policies that are based on different disciplines. One cannot infer an intervention from considering factors separately. The key to success is the completeness of the research network in terms of knowledge and the efficiency of the flow of information through the network (Allen & McGlade, 1987; Frenken, 2006). The problem of completeness can be solved by powerful methods based on semantic analysis (Chavalarias & Cointet, 2008). The resulting research network, as defined by these algorithms, can be mapped using the procedure described in section 6. Once the network is realized, a model is developed that is based on multidisciplinary cooperation. The multi-level nature of complex problems, involving everything from molecular to social interactions, inherently requires cross-disciplinarity approaches. Researchers must extend their working knowledge beyond their own discipline to achieve the required knowledge and semantic overlap among the disciplines; the dictates that the research network needs to learn. In this context, some models and conditions for innovation have been discussed in the context of Swarm intelligence (Bonabeau & Meyer, 2001) and in terms of agent communication languages (Razavi et al., 2009). There are many methods used in complex system investigations and a few are presented here as examples: 1. Machine learning methodologies (Chavalarias & Cointet, 2008; Lutton et al., 2011) could be an integral part of a complex systems approach to analysing problems in the Agric&Food area. 2. Models for communications and information distribution can follow existing lines (Allen & McGlade, 1987; Bonabeau & Meyer, 2001; Frenken, 2006). 3. Methods developed in systems biology (Teusink et al., 2010) and behavioural research (Piqueras-Fiszman, Velasco, & Spence, 2012; Spence, Harrar, & Piqueras-Fiszman, 2012) are key due to their complex and integrated designs. 10

ACCEPTED MANUSCRIPT

M AN U

SC

RI PT

4. Computational methods for analyzing the resulting complex datasets are akin to general algorithms from machine learning and graphical models (Chavalarias & Cointet, 2008). New developments in artificial intelligence and evolutionary algorithm and interactive learning are also interesting approaches that allow capturing implicit expert knowledge (Lutton et al., 2011). Some applications of such methodologies have already been applied to food science (Perrot 2011). 5. Scaling relations need to be given for a reinterpretation at the different levels. Here input is needed from the relevant disciplines. An example is the Structure, Process, and Properties (SPRO2) method (Windhab 2008) constructing a modular model for direct implementation. 6. A theoretical approach can be used in deriving relations from different information sources. This method can help with the interpretation within the different levels. 7. Formal methods with logical or mathematical structure extended by directed data-searches. The actually mappings are the domain of mathematics. 8. Gamification methods developed in the area of recreational games can be implemented in nongame contexts. These methods can be implemented in a cooperation research model where different actors bring their own expertise into the projects akin to the so-called Foldit game (Hescher, 2012). The inferential power of this gamification method of humans is attenuated by inferential power of computers by giving more attention to the human - computer cooperation (Sankar, 2012). Data visualization, akin to the methods developed in bio-informatics and economics (H Rosling, 2011), comes naturally with the machine learning approaches and will be important within the game context.

TE D

We envision that a combination of methodologies is most suitable for the best non-reductionist approach to Agri&Food complex systems. A nice example is work by Perrot et al. (Perrot et al. 2011), in which quantitative knowledge on process- and quality control of cheese making (obtained by electronic sensors) is linked to the tacit and often qualitative knowledge and experience of process operators. Cheeses of different kinds have their own particularities that make them unique. Cheese quality is often laden with cultural and historical pride in the product. By applying fuzzy logic and evolutionary algorithms on operator knowledge and sensory data, one obtained a better process and quality control while maintaining the cultural uniqueness of the product.

5. Systematic approach to complex problems There are three main aspects that are important to the complex system approach that should be considered.

AC C

EP

a) The first is how one describes the specific problem or topic in terms of its conceptual level and scale of observation (spatial and temporal). For example: The level of obesity increases in country x. What intervention strategies can be employed to decrease the rate of increase or cause an overall decrease? b) The second is the harvesting of the information within each level. For this there are many different methods available and used (MacKay, 2003). This amount of information is dependent on the scale (detail) of observation of the system for which the problem is described. For example, describing the random motion of gas molecules requires a lot of information. This information can be greatly reduced when describing phenomena at a macroscopic scale, say pressure, by means of the use of only a few thermodynamic parameters. Increasing the scale of the observer simplifies the description, i.e. it decreases the amount of information required to describe the system at that scale. This is referred to as the so-called complexity profile (Bar-Yam 1997). This profile helps to simplify the description of a complex system. Refinements to this approach have been put forward recently (Harmon, & Bar-Yam, 2012). c) The third important aspect is to estimate the effectiveness of interventions. Effective intervention might call for orchestrated actions between different spatial scales to obtain a desired and robust result. In order to define a robust strategy for action, one needs to identify those spatial levels that effectively (as in the sustainability issue) affect the desired out11

ACCEPTED MANUSCRIPT

M AN U

SC

RI PT

come. To be effective, those levels need to allow for some level of controlled intervention. To allow for such complex orchestrated interventions, a variety of disciplines that are based on fundamentally different conceptual frameworks need to work together; this calls for an interdisciplinary input into the complex problem. The effectiveness of an intervention is not only related to the level of control. It is also related to the conceptual strength of the relevant academic discipline and how it can incorporate results from other, sometimes conceptually disjoint, disciplines of the other spatial scales. We can ascertain this incorporation from conceptually disjoint disciplines by carefully organizing the information flow, and explicitly mapping relevant bits of information to disciplines higher in the organizational level. There is a method in physics of how to connect different time or length scales, referred to as the scaling method (Barenblatt, 2003). There are also examples using scaling approaches in the other levels such as life and behavioural sciences (Bettencourt, Lobo, Strumsky, & West, 2010; Kello et al., 2010). A good measure for conceptual strength can be defined in terms of information theory; the least complex and most relative informative model is to be preferred (Y Bar-Yam, 2005). Developing models at the different spatial scales and combining them into one larger model to allow for the orchestrated action is the best strategy. The control at different relevant spatial scales and integrated interpretations of the relevant conceptual levels are key to an effective approach. The impact of the possible interventions/policies should be studied, or simulated, and according refinements regarding the relevant spatial scale and complexity should be made in subsequent cycles. Regarding the effectiveness of interventions in complex systems, a ―one solution fits all‖ interventions/policies will not be effective, but, instead, multi-level interventions that are congruent with the complexity of the problem are required (Bar-Yam 2004). Our proposed methodology is based on the following roadmap of procedures (see Figure 3):

AC C

EP

TE D

1. Formulating a well-posed research question, at a particular level, sets the context that defines the conditional structure. Its operational form is as a study design where lower hierarchal levels are nested in the higher levels. In Figure 3 we give an example for four different levels (represented by different sciences). 2. Modern methods in statistics (MacKay, 2003) allow for the analysis of these complex and compound nested data-sets. Here the field of machine learning and Bayesian algorithms provide key methods. These methods can do the "bookkeeping" of the project as they particularly focus on the conditional structure. Power laws and dimensions of the leading variables will serve as output for these algorithms. These power laws are derived inductively. 3. The interpretation of the power laws at the relevant aggregation levels is expressed in scaling laws. These laws identify relations that can span a number of levels and are applied in almost every relevant discipline; they are expressed in the form of powers of dimensions relevant to that aggregation level. Scaling laws however, are not always easy to find or interpretable in the different levels. These scaling laws are derived deductively. 4. Scaling laws are indications for special types of underlying structures. They are well known in physics (Barenblatt, 2003) and engineering. They can be linked to underlying principles of control, robustness and self-organisation (Csete & Doyle, 2002), and conservation laws. 5. When these underlying principles have been identified, the resulting structures can be investigated in relation to each other; this can be done conceptually or by using mathematics at a more abstract level. These relations can be seen as mapping of one structure into the other. It is the inverse problem of the nested design in procedural step 1 above. This step leads to a generic model linking all hierarchal levels to each other. 6. This last result in 5 can be used in formulating a new research question, which brings you again to procedural step 1. For application and product development, steps 1 to 3 are sufficient. In order to obtain generalizations, steps 4 and 5 are necessary. 12

ACCEPTED MANUSCRIPT

To summarize our approach, an inductive empirical description sets the conditional and correlation structure of the problem. This is followed by a deductive theoretical description of the underlying structure.

7. Conclusions

M AN U

SC

RI PT

Real-life global and local challenges often require a multi-level and open-system approach. In linking different hierarchical levels and their respective disciplines, one is often hampered by the fact that these levels are conceptually disjoint, leading to conceptual dichotomies. The combined effort of the proposed approaches tries to rectify this problem. Usually, at the highest hierarchical level, we expect more than one class of correct answers. For instance, individual persons make their decisions based on a weighing of complex factors. This implies that one should not aim for identifying ―the most probable state‖, but, instead, accept the existence of different distributions in different populations, and the existence of individuals with different ―life-histories‖. Our approach consists of a roadmap of procedures and encompasses an array of methods utilised in an integrative multi-scale and inter-disciplinary way. This approach will allow researchers to apply a systmatic approach to address Agri&Food challenges. Policy makers that aim at addressing complex problems should realise that a system approach is required, and that there does not necessarily exist one single model; they should adopt an adaptive policy strategy with respect to the flow of information produced by the evolutionary system with multiple realisations. Our systematic approach will enable all stakeholders (practitioners, policy makers, and researchers) to effectively address future challenges in Agri&Food, while maintaining the capacity to endure.

Acknowledgements.

AC C

References.

EP

TE D

We acknowledge the Top Institute of Food and Nutrition for financial support and organizing the Food Innovation Summit 2012 and for financial support later on (Harald van Mil, Erik van der Linden). The current paper originates from the outcome of a symposium held in Oosterbeek, The Netherlands, and on subsequent work. The symposium entitled ―Agri-Food and Science of Complexity‖ was held in June 2012, under sponsorship of the Top Institute Food and Nutrition in The Netherlands. We acknowledge discussions with all contributors during the summit and/or comments on the manuscript by Peter Allen, Yaneer Bar-Yam, Eric Bonabeau, David Chavalarias, Brian Guthrie, Marc-Olivier Coppens, Koen Frenken, Bruce German, Peter Krause, Theo Odijk, Marcel Paques, Betina Piqueras-Fiszman, Peter Schall, Charles Spence and Bas Teusink. EvdL gratefully acknowledges the inspiring discussions on the topic of complexity during the last 5 years with Theo Odijk.

Web resources:

Peter Allen, 2012, http://ezines.tifn.nl/intouch/july2012/06-peter-allen.html Rebecca Hersher, 2012, Nature blog: FoldIt game’s next play: crowdsourcing better drug design. http://blogs.nature.com/spoonful/2012/04/foldit-games-next-play-crowdsourcing-better-drugdesign.html?WT.mc_id=TWT_NatureBlogs (date: 31-07-2013) M. Lagi, Yavni Bar-Yam, K.Z. Bertrand, Yaneer Bar-Yam, UPDATE February 2012 — The Food Crises: Predictive validation of a quantitative model of food prices including speculators and ethanol conversion. arXiv:1203.1313, (date: 31-07-2013) Shyam Sankar, 2012, TED lecture: The rise of human computer cooperation .http://www.ted.com/talks/shyam_sankar_the_rise_of_human_computer_cooperation.html (date: 31-07-2013) 13

ACCEPTED MANUSCRIPT Hans Rosling, 2011, TED lecture: New insights on poverty. http://www.ted.com/talks/hans_rosling_reveals_new_insights_on_poverty.html (date: 31-07-2013) Erich J. Windhab, Sep 30, 2008 – A Process Engineering Approach. Dialogue on Food, Health and Society. Sept. 29 - 30, 2008. Swiss Research Centre http://www.zhaw.ch/fileadmin/user_upload/life_sciences/Dateien/News_Veranstaltungen/Tagungen /Dialogue_Food_Health_Society/Presentation_Erich_Windhab.pdf . (date: 31-07-2013)

RI PT

Literature

Ahn, Y.-Y., Ahnert, S. E., Bagrow, J. P., & Barabasi, A.L. (2011). Flavor network and the principles of food pairing, Scientific reports, 1(196).

SC

Ahnert, S. E. (2013, January 9). Network analysis and data mining in food science: the emergence of computational gastronomy. Flavour. 2(1), 2-4.

M AN U

Allen, P. M., & McGlade, J. M. (1987). Modelling complex human systems: A fisheries example. European Journal of Operational Research, 30(2), 147–167. Alon, U. (2007). Network motifs: theory and experimental approaches. Nature Reviews. Genetics, 8(6), 450–61. Barabási, A.-L., & Oltvai, Z. N. (2004). Network biology: understanding the cell’s functional organization. Nature Reviews. Genetics, 5(2), 101–113.

TE D

Barbara M. Bakker, Karen van Eunen, Jeroen A.L. Jeneson, Natal A.W. van Riel, Frank J. Bruggeman, B. T. (2010). System biology from microorganisms to human metabolic diseases - The role of detailed kinetic. Biochem Soc Trans., 38(5), 1294–1301. Barenblatt, G. I. (2003). Scaling (1st ed). Cambridge University Press.

AC C

EP

Barrière, O., Lutton, E., Wuillemin, P.-H., Baudrit, C., Sicard, M., & Perrot, N. (2013). Cooperative Coevolution for Agrifood Process Modeling. In E. Tantar, A.-A. Tantar, P. Bouvry, P. Del Moral, P. Legrand, C. A. Coello Coello, & O. Schütze (Eds.), EVOLVE- A Bridge between Probability, Set Oriented Numerics and Evolutionary Computation, Studies in Computational Intelligence (Vol. 447, pp. 247–287). Springer Berlin Heidelberg. Bartoshuk, L. M. (2000). Comparing sensory experiences across individuals: Recent psychophysical advances illuminate genetic variation in taste perception. Chemical senses, 25(4), 447–460. Bar-Yam, Y. (1997). Dynamics of Complex Systems Studies in Nonlinearity. (S. Editor & R. L. Devaney, Eds.). Reading, Massachusets: Addison-Wesley. Bar-Yam, Y. (2005). About engineering complex systems: Multiscale analysis and evolutionary engineering. In Brueckner, SA and Serugendo, GDM and Karageorgos, A and Nagpal, R (Ed.), Engineering self-organising systems: Methodologies and applications (Vol. 3464, pp. 16–31). Springer Berlin Heidelberg. Bar-Yam, Y., Harmon, D., & Bar-Yam, Y. (2013.). Computationally Tractable Pairwise Complexity Profile, Complexity,18(5), 20–27.

14

ACCEPTED MANUSCRIPT

Baudrit, C., Sicard, M., Wuillemin, P. H., & Perrot, N. (2010). Towards a global modelling of the Camembert-type cheese ripening process by coupling heterogeneous knowledge with dynamic Bayesian networks.Journal of fooed engineering, 98(3), 283–293. Berry, T. K., Yang, X., & Foegeding, E. A. (2009). Foams Prepared from Whey Protein Isolate and Egg White Protein: 2. Changes Associated with Angel Food Cake Functionality. Journal of food science, 74(5), E269–E277.

RI PT

Bettencourt, L. M. a, Lobo, J., Strumsky, D., & West, G. B. (2010). Urban scaling and its deviations: revealing the structure of wealth, innovation and crime across cities. PloS One, 5(11), e13541. Bonabeau, E., & Meyer, C. (2001). Swarm intelligence - A whole new way to think about business. Harvard business review, 79(5), 106-114.

SC

Brattico, P. (2008). Shallow Reductionism and the Problem of Complexity in Psychology. Theory & Psychology, 18(4), 483–504.

M AN U

Chavalarias, D., & Cointet, J.-P. (2008). Bottom-up scientific field detection for dynamical and hierarchical science mapping, methodology and case study. Scientometrics, 75(1), 37–50. Coppens, M.-O. (2012). A nature-inspired approach to reactor and catalysis engineering. Current Opinion in Chemical Engineering, 1(3), 281–289. Crutchfield, J. P. (2011). Between order and chaos. Nature Physics, 8(1), 17–24. Crutchfield, J. P., & Machta, J. (2011). Introduction to focus issue on ―randomness, structure, and causality: measures of complexity from theory to applications‖. Chaos, 21(3), 037101.

TE D

Csete, M. E., & Doyle, J. C. (2002). Reverse engineering of biological complexity. Science (New York, N.Y.), 295(5560), 1664–1669. French, S. (1986). Decision Theory: An Introduction to the Mathematics of Rationality (p. 448). Ellis Horwood, Ltd.

EP

Frenken, K. (2006). Techological innovation and complexity theory. Econ. Innov. New Techn., 2006, Vol. 15(2), 137–155

AC C

Gentleman, R., Carey, V., Huber, W., Irizarry, R., Dudoit, S. (Eds. . (2005). Bioinformatics and Computational Biology Solutions Using R and Bioconductor. Springer Berlin Heidelberg. Helbing, D. (2010). Pluralistic Modeling of Complex Systems, 1–20. Retrieved from http://web.sg.ethz.ch/wps/pdf/CCSS-10-009.pdf (3-2-2013) Helbing, D. (2010). Quantitative Sociodynamics(1st ed). Berlin, Heidelberg: Springer Berlin Heidelberg. Jaynes, E. T. (2003). Probability Theory The Logic of Science (1st ed). Cambridge University Press. Kello, C. T., Brown, G. D. a, Ferrer-I-Cancho, R., Holden, J. G., Linkenkaer-Hansen, K., Rhodes, T., & Van Orden, G. C. (2010). Scaling laws in cognitive sciences. Trends in Cognitive Sciences, 14(5), 223–32. Kitano, H. (2004). Biological robustness. Nature Reviews. Genetics, 5(11), 826–37. 15

ACCEPTED MANUSCRIPT

Lagi, M., Bertrand, K. Z., & Bar-yam, Y. (2011). The Food Crises and Political Instability in North Africa and the Middle East, arXiv:1108.2455 1–15. Liu, A. J., & Nagel, S. R. (1998). Jamming is not just cool any more. Nature, 396, 21–22. Lutton, E., Paris-sud, U., Cedex, F. O., Team, A., Île-de-france, I. S., & Graphdice, U. S. (2011). Visual Analytics of EA Data. In N Krasnogor (Ed), GECCO’11 Proceedings of the 13th annual conference companion on Genetic and evolutionary computation (145-146)

RI PT

MacKay, D. J. C. (2003). Information Theory, Inference, and Learning Algorithms. Cambridge University Press. Meydani, M., & Hasan, S. T. (2010). Dietary Polyphenols and Obesity. NUTRIENTS, 2(7), 737– 751.

SC

Murphy, R. (1960). The complexity of the problem of obesity. Medical clinics of north America, 44(2), 439–445.

M AN U

Oono, Y. (2013). The Nonlinear World - Conceptual Analysis and Phenomenology . Springer Berlin Heidelberg. Pernell, C. W., Luck, P. J., Foegeding, E. A., & Daubert, C. R. (2002). Heat-induced changes in angel food cakes containing egg-white protein or whey protein isolate. Journal of food science, 67(8), 2945–2951. Perrot, N., Trelea, I. C., Baudrit, C., Trystram, G., & Bourgine, P. (2011). Modelling and analysis of complex food systems: State of the art and new trends. Trends in food science & technology, 22(6), 304–314.

TE D

Piqueras-Fiszman, B., & Spence, C. (2012). The influence of the feel of product packaging on the perception of the oral-somatosensory texture of food. Food quality and preference, 26(1), 67– 73.

EP

Piqueras-Fiszman, B., Velasco, C., & Spence, C. (2012). Exploring implicit and explicit crossmodal colour-flavour correspondences in product packaging. Food quality and preference, 25(2), 148–155.

AC C

Razavi, A., Moschoyiannis, S., & Krause, P. (2009). An open digital environment to support business ecosystems. Peer-to-peer networking and applications, 2(4), 367–397. Rolls, E. T. (2011). Taste, olfactory and food texture reward processing in the brain and obesity. Int J Obes, 35(4), 550–561. Sambrook, T., & Whiten, A. (1997). On the Nature of Complexity in Cognitive and Behavioural Science. Theory & Psychology, 7(2), 191–213. Sornette, D. (2006). Critical Phenomena in Natural Sciences. Berlin/Heidelberg: Springer-Verlag. Spence, C., Harrar, V., & Piqueras-Fiszman, B. (2012). Assessing the impact of the tableware and other contextual variables on multisensory flavour perception. Flavour, 1(1), 7. Teusink, B., Westerhoff, H. V, & Bruggeman, F. J. (2010). Comparative systems biology: from bacteria to man. Wiley Interdisciplinary reviews-systems biology and medicine, 2(5), 518–532.

16

ACCEPTED MANUSCRIPT

Tvenky, A., & Kabneman D. (1974). Judgment under uncertainty: heuristics and biases. Science (New York, N.Y.), 185, 1124–31. Tversky, A., & Kahneman, D. (1983). Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment. Psychological Review, 90, 293–315. Wagner, A. (2012). The role of robustness in phenotypic adaptation and innovation. Proceedings. Biological Sciences / The Royal Society, 279(1732), 1249–58. doi:10.1098/rspb.2011.2293

RI PT

Waldrop, M. M. (1993). Complexity: The emerging science at the edge of order and chaos (1st ed). Simon & Schuster. Wolfram, S. (2002). A New Kind of Science (1st ed), Wolfram Media.

AC C

EP

TE D

M AN U

SC

Yang, X., & Foegeding, E. A. (2011). The stability and physical properties of egg white and whey protein foams explained based on microstructure and interfacial properties. Food hydrocolloids, 25(7), 1687–1701.

17

ACCEPTED MANUSCRIPT

Figure Captions

Figure 1. Animal and plant based products that can be used as ready to eat and as ingredients for other products.

Figure 2. Processes leading to structures and properties at various scales within an Agri-Food context.

AC C

EP

TE D

M AN U

SC

RI PT

Figure 3. Inductive and deductive methods to address specific research questions requiring a complex system approach

18

*Submission Checklist

ACCEPTED MANUSCRIPT Submission’s checklist

1. Manuscript including figure captions at end of article 2. Three figures 3. Highlights

RI PT

4. Letter to the editor

AC C

EP

TE D

M AN U

SC

5. Response to reviewers

Figure 1

Agriculture

ACCEPTED MANUSCRIPT

Livestock

Crops

Harvested Plant Material

Added ingredients Orange Juice

Orange Solids

Pectin

Food Science

Processed Meat

AC C

Soufflé

EP

Processing Technologies

TE D

M AN U

Finished Animal

SC

RI PT

Land

Figure 2

Properties

ACCEPTED MANUSCRIPT

Innovativity Economy Productivity

Process Internal stresses /mobility (in social media, Internet, ...)

Structure Geographical distribution of continents on the planet

RI PT

time

Internet structure

External stresses (peer, social, food scarcity, culture, Preference acceptance, Needs (PAN), ...)

...

M AN U

SC

Groups of persons Compostion/density

Structure of information flow within a company

Social sciences

Person brain mouth

Process Internal stresses /mobility

AC C

EP

TE D

GI-tract

Properties (Mechanics/ Mass transport/ Optics, ...)

Functions (Texture/ Taste/Smell/ Color/ Nutrition, ...)

Structure mm

Person brain

Bubble Droplet

time

mm

Interface

breakdown and dynamics

GI-tract

Network External Stresses Groups of molecules Composition/density

Material sciences

nm

mouth

Tastant/volatile nutrient transport

Figure 3

2. Inductive method Nested empirical route

1. Research question

ACCEPTED MANUSCRIPT

Procedural roadmap

Societal, environmental, ... Sciences

5. mapping

X1~b1,1X1,1+b1,2X1,2

4. interpretation

Overlap up to dichotomous barrier

EP

Study design

AC C

X1,1~b1,1,1X1,1,1+b1,1,2X1,1,2

Material sciences

5. mapping

TE D

Life sciences

Study design

Overlap up to dichotomous barrier

M AN U

Study design

RI PT

Behavioural Sciences

SC

Y~b1X1+b1X1

4. interpretation

3. Scaling relations

5. mapping

Study design

4. interpretation

Overlap up to dichotomous barrier

4. interpretation

3. Deductive method

X1,1,1~b1,1,1,1X1,1,1,1+ ...

2. Inductive power laws

Y ~ X21,3,1 × X2/35,1,2

*Highlights (for review)

ACCEPTED MANUSCRIPT Highlights on “A complex system approach to address world challenges in Food and Agriculture”

- Challenges within Food and Agriculture encompass a wide range of different spatial and temporal scales. - Challenges within Food and Agriculture are researched within a variety of distinct scientific disciplines.

RI PT

- Challenges that involve multi-scale and multi-disciplinary interdependencies are classified as complex.

AC C

EP

TE D

M AN U

SC

- In this paper a systematic pluralistic pragmatic approach is proposed to design intervention strategies for addressing these complex challenges