Policy Significance Statement
We propose two maxims to aid policy making and the preparation of guidelines aimed at creating a more resilient and healthier Department of Computer Science UCL—dictums that, in conjunction with our roadmap, are of potential utility to analogous departments at other universities domestically, regionally, and internationally. The first general principle is that resource use needs to be both minimised and minimal: reduced in relative as well as absolute terms. The second is that responsible research and innovation entails not simply decreasing the resource footprint of a research facility, organisation, institution, or project but also considering nontechnological solutions to complex real-world problems in conjunction with the opportunity costs when the ICT/EEE ecosystem is the first port of call for answers.
Introduction
The Fourth Industrial Revolution (4IR) encompasses a range of new technologies that are fusing the physical, digital, and biological worlds (Schwab, Reference Schwab2016).Footnote 1 Discourse on this economic, industrial, and social paradigmFootnote 2 is replete with phrases such as artificial intelligence (AI), machine learning (ML), the platform economy, automated/algorithmic decision-making, distributed ledger technology (DLT), advanced (bio)manufacturing, genetic codes, and biodiversity data. This is because at its heart lies the knowledge economy: a complicated interlaced mixture of digitisation, datafication, information, analytics, software, hardware, operations, and networks (World Economic Forum, 2017). Since digital technologies reshape the individuals and communities manufacturing and using them (Winner, Reference Winner1980; Qiu, Reference Qiu2016; Eastwood et al., Reference Eastwood, Klerkx and Ayre2017; Hankey and Tuszynski, Reference Hankey and Tuszynski2017), the multitudinous impacts of the ICT/EEE ecosystem, component wise and collectively, are subjects of increasing concern and scrutiny (Chatfield et al., Reference Chatfield, Borsella, Mantovani, Porcari and Stahl2017; Morozov, Reference Morozov2017; Stahl et al., Reference Stahl, Timmermans and Flick2017; Bandara and Carpenter, Reference Bandara and Carpenter2018; Crawford and Joler, Reference Crawford and Joler2018; Levine, Reference Levine2018; McQuillan, Reference McQuillan2018; Nardi et al., Reference Nardi, Tomlinson, Patterson, Chen, Pargman, Raghavan and Penzenstadler2018; Bassey, Reference Bassey2019; Dãno and Prato, Reference Dãno and Prato2019; Donovan and Park, Reference Donovan and Park2019; Howell, Reference Howell2019; Kolinjivadi, Reference Kolinjivadi2019; Lohmann, Reference Lohmann2019; Lohmann and Hildyard, Reference Lohmann and Hildyard2019; Sadowski, Reference Sadowski2019; Thomas, Reference Thomas2019; United Nations Conference on Trade and Development (UNCTAD), 2019). Issues include the fitness for purpose of competition/antitrust law in areas such as algorithmic pricing, collusion, compliance, climate change, and sustainability (Schwalbe, Reference Schwalbe2018; Autorité de la concurrence and German Bundeskartellamt, 2019; Calvano et al., Reference Calvano, Calzolari, Denicoló and Pastorello2019; Competition Authorities Working Group on Digital Economy, 2019; Conner, Reference Conner2019; Deng, Reference Deng2019; Lianos, Reference Lianos2019; Trade and Development Board, Trade and Development Commission, and Intergovernmental Group of Experts on Competition Law and Policy, 2019; Holmes, Reference Holmes2020), the transformation of data from objects to assets (Leonelli, Reference Leonelli2019), frameworks for community data ownership (Singh and Vipra, Reference Singh and Vipra2019), intra- and inter-generational inequalities (Guttieres, Reference Guttieres2019), tax-related provisions in rules on digital trade (James, Reference James2019), developmental justice (Gurumurthy and Chami, Reference Gurumurthy and Chami2019), participatory technology assessment (Ribeiro, Reference Ribeiro2019), and the elusiveness of the benefits promised by dematerialisationFootnote 3 (the reduction in the quantity of materials required to deliver the same level of functionality).
Application of raw and interpreted data to societal issues such as economic efficiency, industrial/agricultural production, healthcare, and education has brought to the fore the concept of “ethics engineering.” When broadly understood as guiding principles that engineers are to keep in mind when going about their activity (non-corruption, legality, public good considerations, and so on), one example is Royal Academy of Engineering (n.d.). When interpreted as the weaving of ethical functionality into technology-based products (ethical-by-design), one example is Engin (Reference Engin2018); “compliance engineering” can be interpreted similarly. During times of crisis such as natural disasters and global pandemics, efforts to apply the techniques and tools of data science, AI, and ML to clinical, epidemiological, social media, whole genomic sequence, telephony, and other types of data may relegate ethical, legal, regulatory, and related concerns to an afterthought rather than regard them as integral to the endeavor as statistical methods. The development of models for prognostication in intensive care units requires not just technical refinements but careful implementations according to principles such as autonomy, justice, beneficence, non-maleficence, and explicability (Beil et al., Reference Beil, Proft, van Heerden, Sviri and van Heerden2019).
Greater awareness of the knowledge economy’s materiality and appreciation that the virtual realm is neither ephemeral nor intangible (Bhardwaj, Reference Bhardwaj2018) has significant consequences for the evolution of the 4IR. Notwithstanding system challenges (Stoica et al., Reference Stoica, Song, Popa, Patterson, Mahoney, Katz, Joseph, Jordan, Hellerstein, Gonzalez, Goldberg, Ghodsi, Culler and Abbeel2017), consider AI, a pillar of the business and industrial strategy of the United Kingdom (Department for Business, Energy & Industrial Strategy, 2017) and many other countries (Dutton, Reference Dutton2018; Castro et al., Reference Castro, McLaughlin and Chivot2019). Typically, efforts to understand the effects of AI on and interaction with society encompass tasks such as the following. First, developing principles and tools for assessing and estimating transparency, accountability, fairness, (in)justice, robustness, explainability, and provenance with respect to data, algorithms, and robots. Second, addressing risks to the privacy and autonomy of individuals. Third, ascertaining liability when algorithms make mistakes. Finally, articulating the legal standing of groups whose composition is dynamic because new members join and old ones leave on an ad hoc basis (for instance, the categories induced by Big Data analytics, demographic groups, people on a social network site, and online communities formed in response to crime, transport, health, citizen science, or other social concern)—see, for example, Association for Computing Machinery (ACM) U.S. Public Policy Council and ACM Europe Policy Committee (2017), O’Neil (Reference O’Neil2017), Torresen (Reference Torresen2018), Benjamin (Reference Benjamin2019), Birhane (Reference Birhane2019), Birhane and Cummins (Reference Birhane and Cummins2019), Birhane and van Dijk (Reference Birhane and van Dijk2020), Floridi and Cowls (Reference Floridi and Cowls2019), Jordan et al. (Reference Jordan, Fazelpour, Koshiyama, Kueper, DeChant, Leong, Marchant and Shank2019), Koshiyama and Engin (Reference Koshiyama and Engin2019), Marcus (Reference Marcus2019), Mietchen et al. (Reference Mietchen, Schwaiger and Beyan2019), Mittelstadt (Reference Mittelstadt2019), Ochigame (Reference Ochigame2019), Phillips and Mian (Reference Phillips and Mian2019), Mateos-Garcia (Reference Mateos-Garcia2020), and Raff (Reference Raff2020).
Here, our focus is not the ethical, economic, sociopolitical, historical, and related aspects of applications such as national and international public and private AI for (social) good-related projects (Tekisalp, Reference Tekisalp2020; AI for Good Foundation (AI4Good), n.d.; ITU, XPRIZE Foundation, 37 United Nations agencies, and ACM, n.d.; Sharma, Reference Sharman.d.; Villani, Reference Villanin.d.). Rather, while concrete machines, networks, and networked systems implement computational abstractions, mediate organism-machine “mergers,” support dynamic groups, and perform other functions which may be perceived as “phantasmagorical” by the general public, the resource-related aspects of AI/ML itself has received less attention. That is, the material aspects of this technology per se is also a societal issue, not least because practical topics and externalities such as scalability and sustainability (Moore, Reference Moore2019; Schwartz et al., Reference Schwartz, Dodge, Smith and Etzioni2019) raise concerns about, for instance, the physical resources, human labour, and data required to run a large-scale AI system (Crawford and Joler, Reference Crawford and Joler2018). The social, built, and natural environments in which AI is embedded and the matter with a real and independent existence embodied in AI/ML—metals, minerals, energy, water, land, and so on—are equally noteworthy.
The ICT/EEE ecosystem is a downstream outcome of research and development performed at universities, companies, the military, and other organisations. Given their fundamental role in shaping the present and charting the future, such institutions can make important contributions to militating against the knowledge economy’s deleterious resource-related impacts. Although not a formal case study, this work is one step towards realising this potential: an empirical investigation of the resource footprint of an academic computer science department. We take measure of, and interrogate the energy, (raw) materials including water, space, and time footprint of the Department of Computer Science UCL. Our focus is the complex, indeterminate, and ever-changing nature of the challenges we face with respect to the resources consumed by the buildings (place), their occupants (people), and their activities (pedagogy).
In the “Background” section, we provide the context for this up-close and in-depth self-examination (including a bird’s eye view of the Department). Then in the “Pedagogy” section, we discuss the present state of affairs and suggest ideas for the future. For reasons of space, similar discussions of Place and People are available as Supplementary Material. In the “A more resilient and healthier Department” section, we sketch three interlinked paths in our roadmap for the Department. Finally in the “Concluding remarks” section, we discuss the limitations of this study and highlight the need for national and international bodies such as United Kingdom (UK) Research and Innovation (UKRI) to develop policies and guidelines on the resource footprint of research projects, organisations, and facilities supported by public funds. Since computer science departments and nonuniversity computing centres elsewhere have similar missions, our approach, observations, roadmap, and high-level principles provide starting points they can tailor, adapt, and generalise to suit their own needs. With this introspection, we hope to identify and highlight the direct responsibility of computer scientists/engineers with respect to issues of resilience and sustainability and to place them at the forefront of our community.
In March 2020 as a response to the coronavirus outbreak, UCL shifted from day-to-day operations out of its campus to remote teaching, learning, and services. This ongoing event means that the Department’s current pattern of resource consumption is radically different from that described here: from the buildings being largely unoccupied because people are working from home to enormous stresses being placed on our communication networks, personal computers, and data centres as well as the larger ICT/EEE ecosystem in which they are embedded; all activities are occurring online, for instance, videoconferencing has replaced in-person meetings. Beyond impacts on the usage and resiliency of energy and telecommunications infrastructure (McWilliams and Zachmann, Reference McWilliams and Zachmann2020; Stolton, Reference Stolton2020), the COVID-19 pandemic is raising the profile of the precautionary principle (Eyres, Reference Eyres2020), showing the value of caution as a guiding influence on our future understanding of the technical and nontechnical aspects of how to create living laboratories for teaching and learning about resource-constrained computing, computation, and communication (Cordes, Reference Cordes2020). We hope this self-reflection initiates broader conversations among and between students, staff, researchers, policy makers, funding agencies, the general public(s), and other stakeholders on what it means to be a resilient and healthy computer science department. We propose that the food, fibre and dye ecosystemFootnote 4 provides lessons for the ICT/EEE (and biological) ecosystem of which the Department is an integral part. In turn, our place–people–pedagogy framework can inform analogous efforts by stakeholders in, for instance, the health ecosystem (Abbasi, Reference Abbasi2020)—“studies from the UK, US, Australia, and Japan indicate that healthcare’s climate footprint is equivalent to 4.4% of global net emissions” (Pencheon and Wight, Reference Pencheon and Wight2020).
Despite the aforementioned caveat, our enquiry can inform the development of more holistic approaches to the governance and assessment of converging disruptive technologies and future spaces at the physical–digital–biological interface. We envisage new “ways of seeing” (Berger, Reference Berger1972) society and the environment as well the genesis of these interacting realms (Dürrenmatt Reference Dürrenmatt2006; Matlack, Reference Matlack2012). For instance, stakeholders exploring the implications of data-driven technologies, automation, molecular technologies, or earth-system engineering approaches could widen the spectrum of factors they take into account by inspecting their subject using the lenses of the recently proposed 
$ FL{E}^5 SH $ framework (
$ F $ = Financial, 
$ L $ = Legal, 
$ {E}^5 $ = Economic, Ethical, Equity, Environmental, and Ecosystem, 
$ S $ = Sociopolitical, 
$ H $ = Historical) (Phillips and Mian, Reference Phillips and Mian2019).Footnote 5
Background
Energy consumption: the rebound effect or Jevons paradox
In the academic and public spheres, there is increasing acceptance of the vast quantities of power utilised by the ICT/EEE ecosystem and awareness that the anticipated efficiency savings of new technologies or other measures are often partly or completely clawed back by behavioural changes (Walsh, Reference Walsh2013; De Decker, Reference De Decker2014; LaMonica, Reference LaMonica2014; Gurzu, Reference Gurzu2017; Santarius, Reference Santarius2017; Turpin, Reference Turpin2017). The latter phenomenon—the rebound effect or Jevons paradox (Ruzzenenti et al., Reference Ruzzenenti, Wagner, Sorrell, Galvin, Vivanco and Walnum2018)—occurs when innovations in production or consumption induce an increase in energy consumption that offsets the technology-derived saving (Stern, Reference Stern2011). Building on the concept of energy efficiency rebound, the circular economy rebound is said to occur when “circular economy activities, which have lower per-unit-production impacts, also cause increased levels of production, reducing their benefit” (Zink and Geyer, Reference Zink and Geyer2017). Resource-related aspects of elements of the ICT/EEE ecosystem such as data centresFootnote 6 networks,Footnote 7 DLT and blockchain,Footnote 8 cryptocurrencies,Footnote 9 ML,Footnote 10 and computer numerical controlled machine tools (3D printers)Footnote 11 continue to be of interest.
Life cycle analysis: cradle-to-grave
ICT/EEE ecosystem components utilise resources across their entire life history with their environmental and ecological footprint being visible in multiple places: from mining (exploration, extraction, and processing) of nonrenewable materials and manufacturing, through production and transportation, to utilisation and disposal (De Decker, Reference De Decker2009; Piccirillo, Reference Piccirillo2011; Dunlap, Reference Dunlap2019; Electronics Takeback Coalition, n.d.; Singer, Reference Singern.d.).Footnote 12 Concerns about day-to-day operational use of digital technology include the increasing demand for power, the growing need for raw materials, the generation of ever larger amounts of plastic waste (Geyer et al., Reference Geyer, Jambeck and Law2017; Liboiron, Reference Liboiron2018) (including by 3D printers, Brice, Reference Brice2017), and the emission of greenhouse gases (Parks and Starosielski, Reference Parks and Starosielski2015). In 2011, the number of digital electronic and radio-frequency identification–chipped devices connected wirelessly to the internet was projected to reach 50 billion by 2020, or 
$ \sim $7 per person (Evans, Reference Evans2011). According to the Solving the E-waste Problem (StEP) international initiative, “e-waste” covers all types of EEE and their parts that have been discarded by the owner as waste without the intention of reuse (Kuehr, Reference Kuehr2014).Footnote 13 E-waste is a product of the largest and fastest growing manufacturing industries: 
$ \sim $41.8 million metric tonnes (Mt) was generated in 2014 and could reach 50 Mt by 2018 (Baldé et al., Reference Baldé, Wang, Kuehr and Huisman2015); the total may escalate to 100 Mt by 2020, probably more given current research and development in areas such as the Internet of Things and wearable technology (Zhang et al., Reference Zhang, Schnoor and Zeng2012; Agyepong, Reference Agyepong2014).Footnote 14 Across a wide range of geospatial and temporal scales, the ICT/EEE ecosystem affects not just human and environmental health (Harkinson, Reference Harkinson2015; Rucevska et al., Reference Rucevska, Nellemann, Isarin, Yang, Liu, Yu, Sandnaes, Olley, McCann, Devia, Bisschop, Soesilo, Schoolmeester, Henriksen and Nilsen2015; Ahmed, Reference Ahmed2016) but also (agricultural) biodiversity.Footnote 15
UCL Computer Science: a bird’s eye view
In order to quantify the Department’s resource footprint, we would require full cost accounting models for every activity, product, process, service, and infrastructure: systematic approaches that identified, summed, and reported the costs involved in the complete life cycle—from direct private costs, through indirect private costs, to social, environmental, and other costs (Dutta and Hasan, Reference Dutta, Hasan, Altmann, Vanmechelen and Rana2013). Clearly, enunciating what we need models for, pinpointing appropriate models, collecting relevant data, and applying models in the real world are nontrivial tasks. To comprehend the magnitude of this task, a short overview of the Department and a brief explication of the activities that take place in pursuit of its goals is useful. Every activity takes place in a specific building and/or utilises various defined ICT/EEE elements. Each location in and of itself consumes resources such as electrical energy, water, and space. Every service consumes resources and requires other hardware and software for its operation. The manufacture, transport, installation, and decommissioning of physical hardware generates greenhouse gas emissions and waste. For each activity, the agents or actors involved include staff, students, and external providers such as the higher organisational and academic structures in which the Department is embedded. Some factors can be quantified fairly precisely (e.g., the electrical power consumed by a server and the number of bits through a network port), others can be approximated only because they pass outside the control of the Department (e.g., hardware manufacturers), and many—if not most—cannot be measured at all (but it may be possible to determine their contribution to the total).
Consider the Department’s undergraduate module on ML. Designed to enhance its teaching provision, this course attracted an extra 20 students in the 2017—2018 academic year, a number that will increase in subsequent years given its popularity. The students need physical space in the form of, for example, classrooms that require heating, ventilation, cooling, and electrical power. The ICT/EEE services they need will consume resources. The two new graphics processing unit (GPU) servers provided to facilitate teaching require 6U of rack space and add an extra 3 kW to the power drawn by the Department; this in turn requires an equivalent amount of cooling. The manufacturer does not disclose the resources consumed during their production and transport, but UCL disposes of their packaging—
$ \sim $25 kg of cardboard, expanded polythene foam, and the pallet on which they were transported—through an external recycling company. These servers utilise a portion of the existing network and storage infrastructure. The course uses a cloud software package, provided free of charge by the vendor, but this adds additional network load throughout the path, consuming resources wherever the vendor is hosting the cloud application. The bandwidth consumed by the application can be measured on the departmental router, allowing an estimate of Watts per bit to be established.
After 3–4 years of use, the servers will be replaced by the next generation of hardware. The Department will continue to operate the original machines as part of a general compute/GPU cluster for another 3–4 years. As this period will exceed the hardware’s “normal” lifespan, each component will be out of warranty and so not replaced as it fails. Any useful parts that are compatible with other devices (such as power supplies) will be kept—but these need to be stored and as is often the case, may never see usage again. On final removal from the Department, UCL Estates and Facilities disposes of the hardware through an external e-waste company. In the absence of information about the company’s practices, it is unknown how much material is ultimately extracted, whether this is recycled into the manufacturing process, and the energy costs of either or both. ML-related teaching and research activities are part and parcel of academic computer science department worldwide, so although the exact settings and precise circumstances will vary from place to place, many of the problems and difficulties we have described will be familiar.
The heterogenous nature of the Department’s research and funding sources is reflected in the bespoke nature of its research infrastructure and data centre. Thus, in order to calculate a metric such as data throughput, CPU cycles, power, and temperature when ascertaining consumption of a particular resource, a substantial amount of time and effort would be required to gather consistent and relevant information from a plethora of agents and actors. We expect other academic computer science departments wishing to determine energy efficiency metrics for their own data center (Newcombe, Reference Newcombe2008) will encounter similar hurdles.
Pedagogy—Teaching, Learning, Investigating, and Other Activities
Present state of affairs: classroom
Green computing
Green IT is the “study and practice of designing, manufacturing, using and disposing of computers, servers and associated subsystems—such as monitors, printers, storage devices and networking and communications systems—efficiently and effectively with minimal or no impact on the environment. Green IT also strives to achieve economic viability and improved system performance and use, while abiding by our social and ethical responsibilities” (Murugesan, Reference Murugesan2008). Environmental sustainability, the economics of energy efficiency, and the total cost of ownership (including that of disposal and recycling) fall under the rubric of “green” computing. Ideas such as green networking and energy-aware security (Merlo et al., Reference Merlo, Migliardi and Caviglione2015) are of practical and theoretical relevance: from materials to devices to circuits to complete systems, fundamental limits to computation exist in areas such as manufacturing, energy, physical space, design and verification, and algorithms (Markov, Reference Markov2014). Currently, the Department has few undergraduate/postgraduate courses, reading groups, or other vehicles that could be categorised as addressing green computing-related problems and solutions.
Ideas for the future: resource-constrained computing, computation, and communication
Resource-efficient hardware, software, and security
The Department could chart a course toward resource use being both minimised and minimal by incorporating the concept of resource-constrained computing, computation, and communication into the fabric of instruction, research, and development—facilitating and encouraging exploration of topics such as quantifying the resources required to achieve a given level of efficiency in hardware (computing), software (algorithms), and security (information transmission in the presence of adversaries and eavesdroppers). With respect to developing, training, testing, and running models, “green AI” advocates evaluating accuracy as a function of computational and financial costs (Schwartz et al., Reference Schwartz, Dodge, Smith and Etzioni2019). The financial and environmental costs of training a variety of popular off-the-shelf neural network models for natural language processing have been assessed by estimating the kilowatts of energy consumed and converting them into approximate carbon emissions and electricity costs (Strubell et al., Reference Strubell, Ganesh and McCallum2019); model inference is also of concern (Biewald, Reference Biewald2019). Software systems developers have considered the technical and economic requirements as well as the social and environmental dimensions of their craft (Lago et al., Reference Lago, Akinli Koçak, Crnkovic and Penzenstadler2015). Realising smaller, lighter, faster, cheaper, and cooler ICT/EEE hardware and software will require advances such as reducing power and raw material consumption, lowering the financial costs of computation and digital preservation, decreasing carbon emissions, lessening environmental impact, improving systems performance and use, and saving physical space (Muelhlhauser, Reference Muelhlhauser2014; Rosenthal, Reference Rosenthal2014). Since using less energy produces less heat waste yielding higher clock speed, reversible computing is one potential solution (Stauffer, Reference Stauffer2013; Lynch and Demaine, Reference Lynch and Demaine2014).
A living laboratory for experimental computer science
The Department’s Technical Support Group (TSG) could explore the feasibility of creating a fully functional machine room that simultaneously enables and facilitates staff and students to observe, monitor, and investigate the operation and behaviour of a complex real-life computing facility. The resultant information could be used to define, refine, and implement solutions for reducing the Department’s resource footprint. Unfortunately, calculating efficiency is difficult because the latest processors will switch processing speed depending on workloads but will cap these turbo speeds if particular instructions are used (Intel Developer Zone, 2017). The TSG is performing experiments to ascertain which processor model is best suited to which particular job type(s).
A back to the future interest group
Informal groupings of staff and students from computer science and other disciplines could re-examine historical technologies and approaches with a view to informing the present and future, culturally and practically. The rebound effect or Jevons paradox refers to expected savings not being (fully) realised because of induced demand (De Decker, Reference De Decker2018; Ruzzenenti et al., Reference Ruzzenenti, Wagner, Sorrell, Galvin, Vivanco and Walnum2018). This counterintuitive notion dates back to the industrial revolution: the more efficient use of coal made possible by technology caused the extraction and consumption of more coal rather than the preservation of existing reserves (Jevons, Reference Jevons1865). Work on systemic drivers of this phenomenon concluded that “sustainability cannot be achieved by technological innovations alone, but requires a continuous process of institutional and behavioural adjustment” (Giampietro and Mayumi, Reference Giampietro and Mayumi2018). In the late 19th century, the nature and availability of materials such as rubber, gutta-percha, copper, and hessian shaped development of the telegraph and transatlantic communication (Burns, Reference Burnsn.d.). Virtually every technology invented in the last 30 years—smartphones, wind turbines, hybrid/electric cars, MRI scanners, and so on—uses rare earth metals and rare gases (Simpson, Reference Simpson2011; Chao, Reference Chao2012). Rising demand in the green/clean/alternative technology sectors is depleting rapidly the world’s entire supply of strategic materials (MIT Solving Complex Problems, 2012; Bardi, Reference Bardi2014). What more can be learnt from the past’s understanding of technology’s materiality?
Modern subjects such as latency, bandwidth, and delay (disruption) tolerant networks could draw lessons from the 18th-century optical telegraph, a communications network for forwarding coded messages over long distances without the need for wires, electricity, horses, or postmen and an e-mail system that could achieve transmission speeds of 
$ \sim $1,400 kilometres per hour (De Decker, Reference De Decker2007). From early antiquity, private persons, governments, the military, press agencies, stockbrokers, and others have used carrier pigeons to convey messages.Footnote 16 Sneakernets—the physical transport of classical information stored in removable media—are used today (De Decker, Reference De Decker2015; Chong, Reference Chong2017; Moss, Reference Moss2017; Wall, Reference Wall2017) and have been proposed as a low-latency high-fidelity network architecture for quantum computing across global distances: ships carry error-corrected quantum memories installed in cargo containers (Devitt et al., Reference Devitt, Greentree, Stephens and Van Meter2016). What more can the evolutionarily ancient method of communication using molecules such as hormones, pheromones, and metabolites (Nakano, Reference Nakano2017) teach us about the virtues of reliably and securely transmitting information (Rose and Wright, Reference Rose and Wright2004; Rose et al., Reference Rose, Mian and Ozmen2019)?
The challenges of digital preservation are technical (rendering accurately authenticated content over time), financial (using limited resources to maximise the value delivered to future users), and legal (the public, private, and criminal law covering the initial conservation and subsequent reuse of and access to data, metadata, documents, and software) (Rosenthal, Reference Rosenthal2017). Whereas the environmental and economic costs of digital preservation are known (Rosenthal, Reference Rosenthal2014), less well appreciated are practical consequences of technical properties such as the “fragility of academic communication in the Web era as opposed to its robustness in the paper era” (Rosenthal, Reference Rosenthal2015).Footnote 17 What is the resource footprint of the policies, strategies, and actions needed to ensure that digital information of continuing value remains accessible and usable?
Horizon scanning interest group
Informal groupings of staff and students could identify emerging issues in science, technology, engineering, mathematics, and medicine, employ the 
$ FL{E}^5 SH $ framework to analyse their implications for society, and sketch out advice on possible future changes, threats, and options for policy makers and Parliament. Consider the following. With predictions of 44 zettabytes of data stored by 2020 (a 10-fold increase from 2013), there is a need to adopt “a more aggressive policy of data archiving on long-term, low-energy, ‘cold’ storage” (Hormann and Campell, Reference Hormann and Campell2014). Given DNA’s remarkable longevity and enormous information density in the natural world (Allentoft et al., Reference Allentoft, Collins, Harker, Haile, Oskam, Hale, Campos, Samaniego, Gilbert, Willerslev, Zhang, Scofield, Holdaway and Bunce2012), this molecule—on its own or inserted into the genome of a living organism—is seen as an attractive medium for archival storage of digital information (Neiman, Reference Neiman1964; Baum, Reference Baum1995; Bancroft et al., Reference Bancroft, Bowler, Bloom and Clelland2001; Yachie et al., Reference Yachie, Ohashi and Tomita2008; Extance, Reference Extance2016; Bornholt et al., Reference Bornholt, Lopez, Carmean, Ceze, Seelig and Strauss2017; Heckel et al., Reference Heckel, Shomorony, Ramachandran and Tse2017; de Groot, Reference Groot2018; Ho, Reference Ho2018; Lenz et al., Reference Lenz, Siegel, Wachter-Zeh and Yaakobi2020; Lima, Reference Lima2018; Stefano et al., Reference Stefano, Wang and Kream2018; Tavella et al., Reference Tavella, Giaretta, Dooley-Cullinane, Conti, Coffey and Balasubramaniam2018).Footnote 18 A February 2018 analysis suggested that 
$ \sim $10 tons of DNA could store all the world’s data, an amount that could fit in a semitrailer (Cornish, Reference Cornish2018).
Technical challenges include lowering costs ($102 for storing 1 megabyte in DNA but $0.0001 per year using tape), increasing throughput (synthesis and sequencing are inherently slow, whereas access times of hard drives are milliseconds), and reducing errors in writing, reading, storing, and handling nucleic acids (mismatches between the information conveyed by the physical material and the digital data theoretically associated with it—for example, complete loss of DNA strands and aggregate insertion, deletion, and substitution rates of 
$ \sim $0.01 errors/base) (Heckel, Reference Heckel2018; Heckel et al., Reference Heckel, Mikutis and Grass2019; Organick et al., Reference Organick, Ang, Chen, Lopez, Yekhanin, Makarychev, Racz, Kamath, Gopalan, Nguyen, Takahashi, Newman, Parker, Rashtchian, Stewart, Gupta, Carlson, Mulligan, Carmean, Seelig, Ceze and Strauss2018). Predictable information security challenges include the embedding of malware in synthetic DNA (Ney et al., Reference Ney, Koscher, Organick, Ceze and Kohno2017) and cyberbiosecurity threats arising from the underlying bio-automation and biotechnology (Wintle et al., Reference Wintle, Boehm, Rhodes, Molloy, Millett, Adam, Breitling, Carlson, Casagrande, Dando, Doubleday, Drexler, Edwards, Ellis, Evans, Hammond, Haseloff, Kahl, Kuiken, Lichman, Matthewman, Napier, ÓhÉigeartaigh, Patron, Perello, Shapira, Tait, Takano and Sutherland2017; Peccoud et al., Reference Peccoud, Gallegos, Murch, Buchholz and Raman2018). What extant and new hazards would the convolution of e-waste and biomedical/biological waste (b-waste) pose? What new types of data centres would be required?
Still in their infancy, nucleic acid–based systems for data storage and other purposes (Mueller et al., Reference Mueller, Jafari and Roth2016; Song and Zeng, Reference Song and Zeng2018) necessitate the design and orchestration of diverse cyber-physical systemsFootnote 19 for many aspects of the encoding, synthesis, storage, management, retrieval, decoding, and other steps. Unappreciated, unrecognised, and unanticipated ways in which this nascent component of the data economy could traverse the physical, digital, and biological spheres raise issues such as scalability, sustainability, (bio)safety (consequences for human, environmental, and ecosystem health), (bio)security (dual use), and (bio)privacy. A full cost accounting model of nucleic acid–based archival storage of digital information will require cradle-to-grave studies of known and novel products, processes, services, and infrastructure across their entire life (cf. the practical, financial, ethical, and other aspects of bioresources (Gonzalez-Sanchez et al., Reference Gonzalez-Sanchez, Lopez-Valeiras and García-Montero2014)). In toto, what existing techniques and tools as well as new strategies and methods are needed to identify, determine, and assess the short-, medium- and long-term impacts of a nucleic acid–based knowledge economy on the lithosphere, biosphere (including humans and their societies), atmosphere, and hydrosphere?
A More Resilient and Healthier Department: A Roadmap
Successful development and deployment of a roadmap for the Department requires (a) understanding the basic attitudes, values, and patterns of behaviour that are common to staff and students, including patterns of consumption or nonconsumption; (b) rethinking discarded materials as resources; (c) reducing waste so that it is diverted from landfills, incinerators, and the environment (no burial, burning, or emission into air, water, and land); (d) promoting the interconnected nature of human and environmental health; and (e) scrutinising concepts such as progress and modernity (Nandy, Reference Nandy1988; Merchant, Reference Merchant2006; Mignolo, Reference Mignolo2011; Giannella, Reference Giannella2015; Horton, Reference Horton2017). More broadly, it is important to avoid solutionism, “an unhealthy preoccupation with sexy, monumental, and narrow-minded solutions—the kind of stuff that wows audiences at TED Conferences—to problems that are extremely complex, fluid, and contentious. …solutionism presumes rather than investigates the problems that it is trying to solve, reaching ‘for the answer before the questions have been fully asked.’ How problems are composed matters every bit as much as how problems are” (Morozov, Reference Morozov2013) (for examples of solutionism in the context of the COVID-19 pandemic, see (Ada Lovelace Institute, 2020; Joseph, Reference Joseph2020; Morozov, Reference Morozov2020)).
Formulating policies and developing guidelines that create a living laboratory for teaching and learning about resource-constrained computing, computation, and communication will require a multi-, trans-, and inter-disciplinary approach. We will need data pertaining to the technical aspects of resource consumption as well as information and ideas relevant to the architectural, human, and philosophical dimensions of the task. Undergirding the making, taking, or advocating of a particular course of action is critical discourse on the viewpoints, uncertainties, contexts, decisions, and/or possible outcomes of proposed directions on the widest possible spectrum of stakeholders. One potential strategy is the establishment of a resource-aware problem-solving laboratory spanning the Department, the Slade School of Fine Art, the Bartlett School of Architecture, the Sarah Parker Remond Centre for the Study of Racism and Racialisation, and members of the general public. By virtue of its ability to examine problems from multiple angles, the resultant UCL Department of (Re)search would be well placed to facilitate dialogue in which participants could probe topics such as energy sufficiency (reducing the growth in energy services as well as the floors and ceilings of energy use) (De Decker, Reference De Decker2018), environmental indicators within which the Department should operate (cf. EAT-Lancet Commission, Willett et al., Reference Willett, Rockström, Loken, Springmann, Lang, Vermeulen, Garnett, Tilman, DeClerck, Wood, Jonell, Clark, Gordon, Fanzo, Hawkes, Zurayk, Rivera, De Vries, Majele Sibanda, Afshin, Chaudhary, Herrero, Agustina, Branca, Lartey, Fan, Crona, Fox, Bignet, Troell, Lindahl, Singh, Cornell, Srinath Reddy, Narain, Nishtar and Murray2019), and the 
$ FL{E}^5 SH $ framework (Phillips and Mian, Reference Phillips and Mian2019). Its initial remit might be to explore how the three aims outlined below could be achieved.
Cap annual power consumption and greenhouse gas emissions
In the 1990s, researchers proposed a pragmatic step toward a sustainable Western lifestyle whereby each person in the developed world—primarily the USA, Canada, Western Europe, and Australia—would consume no more than 2,000 W and emit no more than 1 ton of CO2 per year. The idea gained acceptance in the city of Basel and then other regions in Switzerland as well as in Germany. Assessment of the environmental behaviour of 
$ \sim $4,000 Swiss inhabitants plus a life cycle assessment indicated that whereas restraining energy demand to 2,000 W is possible, limiting CO2 production to under 1 ton per person per year is difficult (Notter et al., Reference Notter, Meyer and Althaus2013). Defining per person per year caps applicable to each and every member of the Department would be difficult. An alternative might be setting activity- and group size–based bounds, for instance, lower limits for formal methods and postgraduate courses compared to ML and undergraduate classes. However, the Department is not a hermetically sealed system so even if atomised accounting procedures and mechanisms could be developed, bidirectional movement of people and goods across its porous borders poses challenges such as what fraction of an allowable resource budget should be allocated to an individual with multiple affiliations. Thus, articulating and establishing precise goals for all possible scenarios is likely to be extremely time-consuming and impossible to implement.
A more productive option might be to identify and characterise the most resource hungry practices by soliciting input from stakeholders irrespective of whether they have, do, or will contribute to said usage. Rather than formulating, incentivising attainment, and penalising noncompliance of specific targets, it might be preferable to encourage a sense of individual and collective responsibility, enable co-creation of qualitative as well as quantitative mitigation and/or elimination strategies over a range of spatiotemporal scales. It would be better to foster a shared culture that takes action to reduce resource consumption within our Department but is cognizant of the ripple effect such changes would have on the wider world: our focus should be preparing and promoting a community guide to resilient and healthy computer science departments (see, for example, Conant and Fadem, Reference Conant and Fadem2008).
Knowledge of energy efficiency does not necessarily translate into energy savings (technologies designed originally to reduce energy use can give rise to new applications that eventually raise energy consumption as well as technological obsolescence), energy consumption does not equal electricity consumption (an ICT/EEE with a given kilowatt-hours of electricity rating requires the production of a larger amount of energy because the conversion of one form of energy into another is accompanied by loss of energy), and life cycle analyses may be out of date, incomplete, or not exist (the lifespan of a technology bears witness to myriad parts, materials, and processing techniques, each with its own resource requirements) (De Decker, Reference De Decker2009; Zehner, Reference Zehner2012; De Decker, Reference De Decker2018). Barriers to estimating the Department’s energy consumption include its complex infrastructure and the fast-changing nature plus rapid evolution of the networks, methods, assumptions, and models researchers employ. Given that the 2012 global communications network (end-use devices, networks, data centres, and manufacturing) is postulated to have consumed 8% of that year’s global energy production and the ever-increasing energy consumption per internet user, a “speed limit for the Internet” has been proposed (De Decker, Reference De Decker2015). Furthermore, reductions in the energy intensity of the Internet (energy utilised per unit of information sent) are more than offset by ever higher total energy use arising from shifting consumption patterns (system-level factors) (De Decker, Reference De Decker2015). Self-imposed limits on the demand side of digital communication is one mechanism for ensuring that resource use is not just minimised but also minimal.
Become a zero waste institution
Resource life cycles can be redesigned so that all products can be repurposed to serve the same or similar function so that eventually, nothing is sent to landfills and incinerators (Zero Waste International Alliance, n.d.). Higher than the Pollution Prevention Hierarchy, the Zero Waste Hierarchy of Highest and Best Use considers not just the entire carbon life cycle of materials but also the embodied energy used to extract virgin resources, manufacture a product, and transport a product to market. In essence, if a product cannot be “reused, repaired, rebuilt, refurbished, refinished, resold, recycled or composted, then it should be restricted, redesigned, or removed from production.” Since sustainable resource management is the joint responsibility of producers, communities, and politicians, a UCL Department of (Re)search could make contributions in all three areas: industrial production and design at the front end, a governmental and regulatory landscape in the middle, and consumption, discard use, and disposal at the back end. A UCL Department of (Re)search could be tasked also with articulating what “zero emission” and “zero energy” buildings (Torcellini et al., Reference Torcellini, Pless, Deru and Crawley2006; MacKay, Reference MacKay2009; Gmach et al., Reference Gmach, Chen, Shah, Rolia, Bash, Christian and Sharma2010; Wikipedia contributors, n.d.) mean in the context of an academic computer science department.
Rejuvenate and (re)integrate the natural and built environments
Exposure to and contact with the biome’s micro- and macro-organisms may be important for immune development and might reduce several types of diseases and conditions associated with the modern era (Parker and Ollerton, Reference Parker and Ollerton2013; Rook et al., Reference Rook, Raison and Lowry2014). Indoor plants provide beneficial bacteria, positively influencing human health (Berg et al., Reference Berg, Mahnert and Moissl-Eichinger2014). The Department’s indoor and outdoor natural environment is vital to providing a healthy workplace. An agroecological approach to the building’s landscape could enhance the well-being of students, staff, and visitors by, for instance, facilitating the flow of beneficial soil- and plant-associated micro- and macro-organisms indoors. Some microbes can induce deterioration of building materials and artifacts such as compact discs (Cappitelli and Sorlini, Reference Cappitelli and Sorlini2005; Sterflinger and Piñar, Reference Sterflinger and Piñar2013; Wei et al., Reference Wei, Jiang, Liu, Zhou and Sanchez-Silva2014). How might the microbiomes of the Department’s enclosed private and public spaces affect the daily to long-term operation of its ICT/EEE and vice versa? A “seed-to-skin-to-soil” approach has been applied to a hoodie grown, designed, and crafted using materials from a 150 mile supply chain where, at the end of its life, the nutrients in the composted garment (apart from the metal zip) could be returned to pasture or farmland used to produce fibres and dyes and hence raw materials for subsequent hoodies (Anderson, Reference Anderson2014). Is an analogous “soil-to-soil” philosophy (Fibershed, n.d.) for ICT/EEE feasible?
Concluding Remarks
This self-deliberation focused on characterising the resource footprint of the Department of Computer Science UCL. Beyond the factors discussed here, a full accounting will require identifying and enumerating all manner of externalised costs such as off-site data centres, not least their energy, land, raw material, and water requirements. Despite such limitations, practical steps toward a minimal power consuming, minimal greenhouse gas emitting, and zero waste Department where the natural and built environments are (re)integrated do exist. Design philosophies rather than specific technologies are key: for example, passivhaus, ecological sanitation, rainwater harvesting, and agroecology are place-based approaches that are shaped by local landscapes, communities, building materials, and climate (such issues are discussed in the Supplementary Material). Buildings are not merely mechanical entities that can be deconstructed into parts with certain dimensions—windows and doors of a certain height, area, and volume. Rather, they are socio-technical-ecological networks whose organisation and dynamics are governed both by the physical structure and interactions among and between their (a)biotic components across wide spatiotemporal scales.
Major infrastructure can last for 30–100 years, and even academic curricula have a life span of many years. Thus, the earlier and faster the Department understands the many facets of resource-constrained computing, computation, and communication and mitigates or eliminates its consumption of resources, the less adaptation will be required in the future. We propose two maxims to aid policy making and guideline preparation. First, resource use needs to be both minimised and minimal: reduced in relative as well as in absolute terms. Second, responsible research and innovation (RRI) entails not just decreasing the resource footprint of a research facility, organisation, institution, or project but also considering nontechnological solutions to complex real-world problems and the opportunity costs when the ICT/EEE ecosystem is the first port of call for answers.
Other academic computer science departments can exploit our findings to develop their own roadmaps. For example, the technology company Nvidia produces GPUs for the gaming, cryptocurrency, and professional markets, but its DGX-1 supercomputer is aimed at ML tasks in high-performance data centres, for example, accelerating the use of deep learning by combining GPUs with integrated deep learning software. Our current experiments with Nvidia’s DGX-1 machines show that they each consume 
$ \sim $3 kW, and if we are to include the amount of energy required to efficiently cool these systems, then we are consuming 
$ \sim $3.5 kW. After many hours (and kWh) of training, a student may identify an image such as that of a cat with close to 100% reliability. How users respond to energy-related information about their computations is a topic for further investigation. Beyond this, one of us (D.A.T.) has direct experience of managing a national computing facility funded by the UK’s Engineering and Physical Sciences Research Council (EPSRC) and where one of the first considerations was the ability to power such a system (Giles et al., Reference Giles, Hetherington, Montana, Armour, Bush, Sandler, Gray, Richmond, Trefethen, Cox, McIntosh-Smith and Timm2016). Free to all academic users, this Joint Academic Data Science Endeavour high-performance computing resource is designed to support the needs of ML and related data science applications at six university partners.
According to the EPSRC, “responsible innovation is a process that seeks to promote creativity and opportunities for science and innovation that are socially desirable and undertaken in the public interest” (EPSRC, n.d.). Its AREA (Anticipate, Reflect, Engage, Act) framework (EPSRC, n.d.) seeks to (a) describe and analyse “the impacts, intended or otherwise, (for example economic, social, environmental) that might arise. This does not seek to predict but rather to support an exploration of possible impacts and implications that may otherwise remain uncovered and little discussed”; (b) reflect on “the purposes of, motivations for and potential implications of the research, and the associated uncertainties, areas of ignorance, assumptions, framings, questions, dilemmas and social transformations these may bring”; (c) open up “such visions, impacts and questioning to broader deliberation, dialogue, engagement and debate in an inclusive way”; and (d) use “these processes to influence the direction and trajectory of the research and innovation process itself.” Similarly, “ProGReSS” is a European Commission–funded project whose mission is to promote a European approach to RRI: “research and innovation which is ethically acceptable, sustainable by avoiding significant adverse effects and drives towards the common good, i.e., societal desirability” (European Union, n.d.).
Operating across the whole of the UK with a combined budget of more than £7 billion, UKRI brings together the EPSRC, Arts and Humanities Research Council, Biotechnology and Biological Sciences Research Council, Economic and Social Research Council, Medical Research Council, Natural Environment Research Council, Science and Technology Facilities Council, Innovate UK, and Research England (United Kingdom Research and Innovation (UKRI), n.d.). As part of its newly launched environmental sustainability strategy, UKRI “aspires to be ‘net-zero’ for its entire research undertaking, including reducing and mitigating all carbon emissions from owned operations, and looking beyond carbon to ensuring its wider environmental contribution is a positive one” (United Kingdom Research and Innovation (UKRI), 2020). This study highlights the need for new policies and guidelines on the resource footprints of academic departments, research projects, and e-infrastructure. These could be developed using the findings of a report prepared by a national working group convened by UKRI.
Acknowledgments
We thank E. Kazim, T. Stevens, F. Murtagh, C. Wardle, and P. Stenetorp for comments on the manuscript.
Funding Statement
This work received no specific grant from any funding agency, commercial or not-for-profit sectors.
Competing Interests
The authors declare no competing interests exist
Authorship Contributions
Conceptualization, I.S.M. and D.A.T.; Methodology, I.S.M., D.T., and D.A.T.; Investigation, I.S.M., D.T., and D.A.T.; Writing-original draft, I.S.M.; Writing-review & editing, I.S.M., D.T., and D.A.T. All authors approved the final submitted draft.
Data Availability Statement
Data availability is not applicable to this article as no new data were created or analysed in this study.
Ethical Standards
The research meets all ethical guidelines, including adherence to the legal requirements of the study country.
Supplementary Material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/dap.2020.12.
Open access
Comments
No Comments have been published for this article.