Seminar on Programmable Infrastructures
On March 2nd, 2020, we held an afternoon event at the Faculty of Technology, Policy and Management discussing the emergence of "Programmable Infrastructures”. The event welcomed Meredith Whittaker (AI Now), Seeta Peña Gangadharan (LSE), Stefania Milan (UvA), Eleanor Saitta (Structures Systems), as well as Frances Brazier, Aimee van Wynsbergh, Gerd Kortuem, and Marijn Janssen of TU Delft as speakers. Programmable Infrastructures also marked the first of a series of events supported by the new TPM AI Lab, and came about through collaborations across the departments of the faculty. To frame the day, Seda Gürses and Roel Dobbe, in collaboration with Martha Poon, composed an introductory text that describes the main features of programmable infrastructures, which we provide below. We hope it will invite further conversations on Programmable Infrastructures across the different TPM departments and beyond.
This seminar is supported by the TU Delft TPM AI-Lab.
Introduction to the Programmable Infrastructures Event
This event is an invitation to consider the story of two different kinds of infrastructure. And we’d like us to consider that perhaps one of them is intent on capturing the other one for its own purposes.
The first kind of infrastructure is the one we live with, the one we have already come to know and govern. We can define them as physical, institutional and human structures that organize everyday social life. For example – water systems, sewage systems, electrical grids, roadways, railways and waterways. They also include cityscapes and their administration, systems that deliver healthcare, and education infrastructures; they co-exist with us and rely on political administrators, public authorities, commercial ecosystems, labor organizations and other social groups. In this first category we also include the information or data that play an increasingly operational role in how we organize institutions and everyday activities.
By virtue of our intimate acquaintance with this first category of infrastructure, societies around the world have a deeply historical sense of the values that these infrastructures are intended to serve. Over time, we have developed governance structures and societal norms based on which we come to manage these infrastructures. The public interest tends to matter. So do a variety of goals like universal access and justice. Although, we should acknowledge that in some cases infrastructures are built and sustained by unjust practices like colonization or labor exploitation, in others, they are used to control or violate populations, which further underlines their significance for the fabric of our societies.
During this event let's refer to this first category as "common infrastructure".
Today, we are seeing the rise of categorically distinct type of infrastructure, something that is replacing what we used to think of as ICTs, information and communication technologies and traditional software development. Consisting of a vast global network of data centers, network infrastructure and devices, as well as software platforms, this second category of infrastructure can be deployed onto common infrastructures to automate and augment workflows, optimize supply chains and disrupt markets.
It is this second category that we think we should call “computational infrastructure”. We’d like to give you a brief description from our viewpoint of this newer infrastructure based on our interdisciplinary backgrounds anchored in computer science and engineering.
Let’s think about some of the dramatic changes in the software industry over the last 20 years that have delivered to us computational infrastructures. Traditionally, there were companies around the world that developed software and provided it to different actors through markets for products and services. From the 1980s to the 2000s, software was predominantly a product that was sold to run ‘standalone’ on computers. Fast forward to today, and we see software being increasingly offered and developed through ‘platforms and infrastructures-as-a-service’. These ‘service-oriented architectures’ promise customers the ability to respond more quickly and cost-effectively to changing market conditions. They enable rapid data feedback about people’s behavior, physical environments and digital environments. In response, software developers have moved towards a more data intensive form of developing digital services using iterative engineering methods.
These changes in software architectures, development methods, and business models have given the software industry a new form. We no longer have many software companies that develop monolithic software products from scratch to fit their customers’ requirements. Rather, today’s software companies exist and survive in a so-called "ecosystem" dominated by four mega corporations, namely, Microsoft, Amazon, Google and Apple.
What follows is a high level description of how the ecosystem works – The big four provide computational infrastructure to everyone else and support the ecosystem with a suite of tools and development environments, either developed in-house or by partner companies. Developers use these environments to launch their startups, or to "lift and shift" existing enterprises into the cloud. This data centric software development, mediated by application programming interfaces (APIs) and enabled by a hoard of specialized chips, further intensifies asymmetrical concentration of infrastructural resources in the hands of these mega-companies which can collect large scale data and muster the computational power to process them at scale. These companies have pivoted their business models to align with their increasing customer base who are building on top of their ‘clouds’ or (increasingly) sitting on their ‘edges’.
Our concern is that the computational infrastructures are far more than a technological ecosystem alone. Like all infrastructure, they incentivize us to embed their values, and therewith much of their politics in the lower layers of the technology stack. Comparable to the cables and control equipment in electrical networks that determine what can and cannot be connected to it, computational infrastructures embed constraints on what can and cannot be built on top of it, as well as what is accessible to those needing to audit or validate its functionality. Furthermore, the rapidly growing environmental/carbon implications (already on par with the global aviation industry) as well as the detrimental reliance on labor remain largely non-transparent.
Beyond these concerns, there is an issue of sheer political and economic power. This new infrastructure comes not only with its own tools, values and environmental implications, but also with an expansive political economy, propelled by immense amount of global capital investment. We are talking about computational infrastructures that are run by companies currently valued at above 3 trillion dollars, and that despite the Corona Virus correction going on right now in financial markets.
So what then are programmable infrastructures? And, what does this have to do with common infrastructures?
The term “programmable infrastructures” refers to the political, economic and technological vision that advocates for the introduction of computational infrastructure onto our common infrastructures. If common infrastructures come with extensive planning and expensive updates, the promise of programmability is that by adding a digital layer, the plans and policies of common infrastructures can be abstracted from their underlying physical constraints. This, it is claimed, will make them easy to reconfigure just like digital systems. In other words, legacy physical infrastructures can be further freed from their physical constraints and can ostensibly be made as programmable as native computational systems.
Going forward, we would like to raise five features of programmable infrastructures –
1. Programmability of Human Behavior: One of the novelties in programmability is that it offers to manage human behavior. Among other things, AI refers to systems that can manage not only the performance of fixed infrastructures, but also the people circulating in them. With mobile technologies, be they phones or autonomous vehicles, everything and everyone can be nudged and optimized through continuous feedback and control. How feasible and desirable is the proposal to extend control and management of infrastructures to human behavior?
2. Dominance of Values-in-Infrastructure: The use of clouds and services comes with a standardization of software functions that can challenge local parties’ ability to develop, test and validate engineering specifications in an end-to-end fashion. This can also pose a challenge to addressing context specific privacy, security and safety measures. How will we ensure that situated values are well represented and systems are accountable when we use these computational infrastructures?
3. Economics of Programmability: Programmability means accommodating ideological, managerial, political and economic consequences of adopting it. With a commitment to programmability, we will enter another dependency relationship with technology providers. We’ll have to pay them and all the startups that run their services on their computational infrastructure. Further, AI frameworks promote reformulating social welfare functions as a problem that can be optimized computationally rather than solved through the complex consideration and negotiation of human experience and expertise, shielding the management of infrastructures from democratic forms of control. How can public interest be assured when it is submitted to these economic terms?
4. Power Asymmetries of Cloud Providers: We won’t deny the potential benefits of programmability for both companies, public institutions and the citizens they serve. But just as the social media giants, like Facebook, find themselves struggling to respect increasing requests to democratize the operation of their networks, safeguarding an unprecedented revenue machine enabled by online advertising, how will computational infrastructure’s relationship to the demands of global capital impact upon our ability to address environmental and other social challenges? Can democratic values and imperatives truly prevail over the bottom line?
5. Avoidance and Reshaping of Democratic Governance: We’d like to raise the question of whether the introduction of programmability to common infrastructures will always be a choice. When startups or larger software companies like AirBnB, Google Maps or Uber, hook programmability onto common infrastructure they do so through their user base, thereby either avoiding or proactively trying to reshape the forms of public governance that oversee related domains. This suggests the usual mechanisms like contracts will not be sufficient to protect the public welfare function, giving a new dimension to private and public ordering of conflicts around common infrastructures. How will we ensure that we step into programmability in a deliberate and democratic manner?
With these five features, we have laid out a framework for understanding programmability. Now we ask you: What are some ways of understanding its promises as well as its perils? And, how can we at TU Delft who have been working on common and computational infrastructures come to conceptualize, engage and change the course of these developments so that we can continue to have democratically managed infrastructures that serve the public interest?
We have invited our guests today to speak to some of the concerns that the wedding of these two types of infrastructures might pose. We thank you for joining us in kicking of this conversation, which hopefully will provide a fertile ground for engaging the challenges of programmable infrastructures across the departments of TPM and beyond.
TU Delft TPM AI-Lab
The TU Delft TPM AI-Lab is an initiative of the TPM Faculty at TU Delft University. This cross-department collaboration aims to strengthen TPM partnerships among researchers within all TPM departments. It also serves as a platform for stimulating new research possibilities among TPM researchers who are working on aspects of artificial intelligence (AI) and its socio-political implications. More details about the initiative's constitution will follow.