“Infrastructure” is a broad term for the physical and institutional systems that support human communities at all scales, from physical systems that have been necessary since the beginning of human urbanization, such as Rome’s road and water infrastructure, to newer forms that reflect cutting edge technology, such as 5G wireless systems. All infrastructure combines two functions. One is explicit: a road carries traffic, for instance, and water pipes carry water. The more subtle one is an enabling function, as all infrastructure also supports other technologies and more infrastructure. Electric infrastructure, for example, performs the explicit function of generating and distributing power; more importantly, it enables electrified housing, industrial production, and information and communications technology.
In the past, most people recognized a new railroad, canal, or university as infrastructure. They knew why it was important. This has changed. A new form of infrastructure, unlike any we’ve seen before, is rapidly developing around us. A host of internet connected devices, communications networks, servers, artificial intelligence programs, and other digital technologies, as well as the humans and communities who use them, are combining to create an emergent cognitive infrastructure that stands in stark contrast to older systems: We don’t recognize it as being infrastructure, and we don’t know exactly what its implications will be.
Our failure to recognize this new infrastructure sprouting up around us is a potentially fatal blindness. Cognitive infrastructure constitutes the medium in which fundamental human activities such as war, politics, psychological manipulation, and soft authoritarianism increasingly take place. Russia, as its agents pushed disinformation on social media, relied on this new cognitive infrastructure during the 2016 US presidential campaign. The ransomware attackers who’ve hijacked government computer networks in Atlanta, New Orleans, and Baltimore in attempts to steal money are another example of new challenges that arise from the integrated cognitive infrastructure, as are the bot armies which now roam Twitter, Facebook, and other social media.
Without first recognizing the emergence of this new infrastructure and understanding its characteristics and behaviors, regulating it or developing policies around it will be difficult. For example, the Defense Innovation Board, a Pentagon advisory arm, laid out five ethical requirements for US military artificial intelligence, including the principle that technical experts be able to understand what the AI system is doing and how it is doing it. This reflects the more fundamental US Defense Department policy that humans always be able to understand what led an artificial intelligence program to its conclusion. However, given that AI is proving most useful precisely in those areas that are too complex or fast-moving for human cognition; that it is only a part of an integrated cognitive infrastructure; and that the landscape of cognitive infrastructure is not yet even adequately perceived, such a human-centered principle may well be doomed from the start.
What is cognitive infrastructure? Historically, when new infrastructure appeared, it was usually explicitly and commercially important—think of the canal systems, and then the railroads, that knit together the United States, or the birth of electric infrastructure in the early 1900s. These examples may be quite complicated, but they are single, intelligible systems. Cognitive infrastructure, however, is a meta-infrastructure; it involves accelerating capability and capacity across a number of seemingly unrelated systems and technologies, including 5G communications networks, artificial intelligence and big-data analytics programs, social media, internet-connected appliances and devices, media creation and manipulation tools, cloud storage, and more. Moreover, it is institutionally complex; communities and institutions ranging from activist groups to private firms to militaries develop and use elements of cognitive infrastructure.
The Oxford Dictionary defines cognition as “the mental action or process of acquiring knowledge and understanding through thought, experience, and the senses.” As with most definitions of intelligence and cognition, this is a very anthropocentric definition—“mental” referring to the human mind, and “thought,” “experience,” and “the senses” clearly anchoring cognition in the human sphere. This reflects the general term humans give to so-called thinking technologies—they are called artificial intelligence, as if the only intelligence format possible were ours. But humans are only a part of cognitive infrastructure, and there is no necessary reason that the cognitive processes of, and outputs from, cognitive infrastructure will mirror human thinking and decision-making, with all of the human frailties, heuristics, and emotional shortcuts that behavioral economics increasingly demonstrates underlie our cognitive processes.
Better to adopt a functional definition: Cognition is often taken to include, among many processes, perception, learning, reasoning, memory, and communication with other cognitive systems. Once we adopt a functional definition of cognition, we’ll recognize that the integration of many independent technologies into this new meta-infrastructure increasingly mirrors the capabilities we associate with cognition.
In short, we are building a global cognitive infrastructure without recognizing it.
By 2020, some 425 million servers will be deployed globally, and the number of objects attached to the Internet will run between 25 and 50 billion, including everything from cars and refrigerators to microwaves and, of course, mobile phones and computers. These devices are not just communicating with each other; they all include sophisticated sensor systems and data generation capabilities as well. In fact, by 2020, the National Science Foundation estimates that 1 trillion sensors will be deployed. Integration of these cognitive elements will be provided by artificial intelligence programs networking and powering ever more technologies. Memory? A report by market analysis firm IDC estimates global stored data will increase to 175 zettabytes by 2025, up from 33 zettabytes in 2018. (A zettabyte is a unit of information roughly equal to 1021 bytes.)
Managing the emerging cognitive infrastructure. Cognitive infrastructure involves a level of complexity and information flow that humans can neither understand nor perceive. People are low-bandwidth cognitive mechanisms in a world where cognitive infrastructure even today operates at far higher bandwidths, much faster speeds, and higher levels of complexity than individuals can hope to access.
Knowing, however, that the emerging cognitive infrastructure involves a conceptually new, more complex system provides a bracing reality check to many of the policy formulations floated by governments and academics. Just as the US government could not regulate Rockefeller’s Standard Oil by using the rules applicable to a local coal supplier, attempts to manage the evolving cognitive infrastructure in the same way governments regulated, say, telecoms, are based on mistaken assumptions and doomed to failure. Among many other issues, for example, the cognitive infrastructure is a complex adaptive global system, so the likelihood that national or local regulation or legislation will have the desired effect is very low.
It may be premature to consider tantalizing questions such as how humans should respond as the technology to physically interface our minds with elements of cognitive infrastructure such as devices or networks is developed—a very real, but very challenging, question. But it is not premature to recognize that this new infrastructure is already emerging, and to try and understand some of the implications.
Without that first step, ethical, rational, and appropriate policy and institutional adaptation will remain beyond reach.