Artificial Intelligence and The Rediscovery of Space and Time in Business

BCG GAMMA editor
Aug 14, 2019 · 10 min read

by Philipp Gerbert

Image for post
Image for post

Digital technology has shrunk the planet and enabled effective globalization — at least from a communication and control point of view. But has it? As intelligent machines proliferate, their much faster decisions and actions across the planet hit the physical limits of synchronization throughout space and time. The coordination challenge is reminiscent of the initial global stretch in the age of Christopher Columbus and Vasco da Gama, when the capitals of the Iberian empires could receive information and impose their will only with a significant delay. It creates novel challenges for business and for society at large: machine intelligence will be distributed around the world with only loose connections and a whole new architecture of decentralized autonomy and centralized control is required. In nature, the decentralized intelligence of an Octopus might provide inspiration for future business systems.

Today, we enjoy a connected planet. Synchronous communication has become possible across the globe and we can have high quality transcontinental conversations without experiencing a time delay. With increasing digitalization, this instantaneous decision and action capability has penetrated business processes, both within and across companies, as well as between companies and consumers.

We might not realize it, but from a physical point of view this capability is due both to an important achievement and a fortuitous coincidence. There is a basic constraint in physics: the speed of light, commonly denoted by ‘c’¹, is the absolute limit for signal transmission. Its value is 299.8 thousand km/s, rounded to c=3*10 8 m/s. No information can flow faster than c.

  • First, the achievement: We actually have learnt to communicate at the speed limit c. Optical fiber and microwaves through air transmit signals very close to c, and this is even true for electronic signals in copper (although the electrons themselves are much slower), and we have continuously reduced the switching delays. By contrast, sound travels at 300m/s, i.e. a million times slower. We all know that thunder trails lightning, and if in races we only transmitted the sound of the start pistol, times would be distorted. The sound-light speed relation is actually quite fundamental to our physical world and we shall return to it later.
  • Second, the coincidence: The very shortest human reaction time, e.g. the subconscious blinking reflex of an eyelid, is about 0.1 seconds. The limit for a signal travelling to the other end of the world (roughly 20 thousand km) is about 0.07s. So if we manage routing and latency well, we will not notice any time delay and communication appears instantaneous.²

As we are switching to machine action — in the future increasingly guided by artificial intelligence — commands at a distance will cease to feel instantaneous across our planet, with progressively important implications.

For the business community, the first demonstration of this re-emergence of time and space has probably occurred with the rise of high-frequency trading, popularized by Michael Lewis in his best-selling book Flash Boys³:

The Chicago Mercantile Exchange and the NASDAQ are about 1.160 km apart as the bird flies. It takes light about 3.9ms to travel the distance directly, more time than to execute electronic trades, so markets can no longer be truly synchronized. As popularized by Lewis, in 2007 Dan Spivey realized that this opened an arbitrage opportunity. Dan later co-founded Spread Network, that built a 1.331 km rather direct and low-latency line of optical fiber, cutting the then fastest communication of 17ms down to about 13ms. This allowed the traders paying for the fast connection to make an arbitrage gain on a trade — a practice known as ‘front-running’⁴. Following a similar logic, high-frequency traders started to minimize the physical distance of their computers to the NASDAQ by collocating their servers at the exchange. All other delays being equal, physical distance made the difference and was at the root of huge financial gains (and losses). Space and time re-emerged as critical in a digital world acting at machine speed.

For a more general discussion, let us quantitively assess the emerging speed of both decision-making and action as pertinent to business in a machine age:

Decision Making in a Machine Age:

Human and machine intelligence have very different processing times:

  • Human nerve signals are transmitted at 50–100m/s, while electronic or light signals can travel close to the speed of light at 3*10 8 /s, i.e. about 5 million times faster.
  • Also, while human brain neurons fire around 200 times/second, clock speeds in computers are currently at 3 GHz, i.e. 3*10 9 cycles per second, i.e. about 15 million times faster.

As we progressively introduce artificial intelligence into some areas of decision-making, we thus need to adapt to these decisions being made about 10 million times faster.

Action in a Machine Age:

For macroscopic physical action, the speed scale is normally between 1m/s and 300m/s. The upper limit is the speed of sound — for a variety of reasons the practical boundary for most large mechanical motions, including human action. So, if the ultimate action is both physical and macroscopic, the signal delay might matter less. But for:

  • Electric systems — as in power grids
  • Digital systems (electronic or optical) — as in financial markets
  • Micro-/nanoscopic systems — as in an increasing number of modern devices

all ultimately electromagnetic in nature, the above mentioned difference between the speed of light and the speed of sound means the action itself might happen more than a million times faster than ‘normal’ macroscopic action.

A consequence of this increase in potential decision making and action speed is that the world becomes a much more loosely connected place of local action zones. The time it takes for data to travel to the other end of the globe (even ignoring latencies) feels to a machine a bit like a month-long sailing in Columbus’ times. While digital communication seems instantaneous to us, it is impracticably slow for synchronizing machines. Spatial distance matters a lot and implies locally autonomous systems, whose coordination — possibly with a central intelligence — becomes of paramount importance.

Image for post
Image for post

Now let us discuss some of the affected areas in more detail:

For power grids, engineers have of course always been aware of the synchronization challenges introduced by signal delays. Maintaining a stable frequency of the electric current is highly critical and occurs at three levels of locality and response times: Primary response triggers stabilization within seconds — but without exchange of information between generators. It stabilizes the frequency, but not necessarily at the correct target value. Secondary response, involving central control, restores the nominal level at a regional level within minutes. Finally, tertiary response reoptimizes the reserve power, so the system is again prepared for future disruptions. All these actions today are still limited by the macroscopic rotating masses, and they follow a rigid, pre-defined hierarchical control routine, impracticable for most business applications. When super-fast power electronics as well as artificial intelligence are introduced, distance delays will play a comparatively larger role.

Digital action is already very common today and its universal prevalence is rising fast. This will certainly constitute the primary realm of the space-time coordination challenge. In financial services, for instance, the above mentioned high-frequency trading was the precursor. It did, however, not yet involve intelligence at both ends, but consisted of a mere frontrunning competition in execution. Going forward, the emerging introduction of artificial intelligence will quickly raise the coordination challenge in financial markets, with similar issues in any type of electronic trading, including commodities. Ad serving is another interesting application area, where simple ‘intelligent’ decisions have already led to some decentral autonomy. Cyber security is arguably the most ubiquitous field, where the quickly increasing use of AI in attack and defense, as well as in the very processes targeted by these actions, will introduce the synchronization challenge for concerted action. Many more digital areas are bound to follow.

But our physical world will not escape the challenge. Let us first discuss autonomous mobility as an example, although this still involves macroscopic action. For artificial intelligence, learning can and should be centralized as much as possible, in order to leverage the maximum amount of training data. Action, however, for reliability and safety reasons still needs to be controlled decentrally. Thus, for self-driving vehicles, cars can drive autonomously, but all their data are collected and the learning occurs centrally. The decentralized software is then periodically updated with an improved (and extensively tested) version. For robo-cabs, however, in a typically local environment, fleet managers and cities might interfere with the driving autonomy. And for autonomous trains, control normally shifts even more away from the vehicle and towards the (often centralized) operator intelligence. The overall final architecture of autonomous mobility will be a balanced system of interacting decentral and central intelligence.

When combining AI with IoT more broadly, we continue to want to centralize learning and control, but with increasing action speed we face more severe latency (and possibly data volume/bandwidth) constraints. This has led to the introduction of the ‘edge’ computing architecture, where a local intelligence makes the most time-critical decisions and performs some pre-processing and compression. For less time-critical optimization and control issues, the central unit takes responsibility.

The exhibit below summarizes the emerging challenges of synchronizing intelligence across space and time in the Machine Age:

Image for post
Image for post

As we can see, in an increasing number of operational and business settings powered by AI, we start facing the dilemma of distributed intelligence that requires some kind of coordination or synchronization of actions. This starts happening when the distance, over which we want to coordinate, multiplied with the action speed⁵ is above the speed of light, so that no information signal can be exchanged for full synchronization. Often such situations can still be solved by pre-defined rules-based hierarchies. However, such hierarchies may not be flexible enough, and in complex environments they tend to perform poorly — a lesson learnt from the rules-based expert systems of the 1980s. As the speed of intelligent decision making and action increases, decentral intelligent systems need to have partial autonomy that can only be coordinated with a time delay. Of course, ‘humans-in-the- loop’ are far too slow and play no role in the process. We have precious little experience with such systems in business: How to devise the optimal guidelines from a business and liability perspective? And what if ‘rules’ prove too rigid for the purpose? Sophisticated financial market participants might be the first to reach that frontier over the next years.

Interestingly, there are biological systems that face similar issues of coordinating distributed intelligence. While we are used from humans and other vertebrates to a central nervous system directing the periphery, the octopus represents at least one different intelligent design. The octopus and related species (i.e. cephalopods) decoupled from the human evolutionary arm in pre-Cumbrian times around 600 million years ago. They have evolved an entirely different nervous system, with more than half of their nerve cells in their eight arms (leading to the famous — if misleading — term ‘the nine brains of an octopus’). The arms have their own smell and touch sensors, can process this information locally and act autonomously — even when severed from the main body⁶. However, coordination remains possible: for instance, the central eyes can guide an arm even to food outside of water where local sensory input to the arms is not helpful⁷. The head and the arms have learned to co-operate. Perhaps, a deeper study of how this is structured and achieved can provide some guidance for designing robust machine-based business systems.

Artificial Intelligence and the speed of decision making and action by machines will introduce many challenges. One will be that — due to a fundamental speed limit for information flow — a global AI system can no longer be synchronized, and needs to be structured as a distributed intelligence with a complex coordination challenge⁸. The intelligence of the agents involved renders the task much more complex than in traditional machine control. The phenomenon arises in many different guises depending on the nature and speed of action of the system: virtual vs physical, macro- vs microscopic. It will re-introduce a renewed relevance of space and time in managing decision making and action across distributed, intelligent agents in our ‘globalized’ world. The distributed intelligence of an octopus that has learnt to cooperate could be a great inspiration — perhaps we can learn something from 600 million years of evolutionary wisdom.

Acknowledgements: The author would like to thank Dragan Obradovic, Martin Reeves, Michael Spira, Sebastian Steinhäuser and Thibaut Willeman for comments on the manuscript, as well as Chris Somerville for raising my interest in cephalopods.

About the author: Philipp Gerbert is Fellow of the BCG Henderson Institute focusing on AI in Business, and a Senior Partner at BCG. By background he is a physicist and holds a PhD from the Massachusetts Institute of Technology (MIT) in Cambridge, MA, USA.

¹Yes, this is the c of Einstein’s famous energy-mass equivalence E=mc2

²The moon, by contrast, is on average about 385 thousand km away, so a signal travelling back and forth takes more than 2.5s. A moon-rover thus cannot effectively be remote-controlled from Houston. Communication with Mars is even more constrained. Since the planet is 56–400 million km away — depending on the current location in the orbit — astronauts and devices on Mars will always be on their own for 6–45 min.

³Michael Lewis, Flash Boys, WW Norton&Company, 2014.

⁴Spread Networks was later beaten by a micro-wave connection that cut another 4.5ms of transmission time.

⁵In technical terms this is not a speed, but a frequency, i.e. the inverse of a time an action takes.

⁶This goes far beyond the trivial ‘reflexes’ vertebrates are capable of.

⁷For a more in-depth discussion see Peter Godfrey-Smith Other Minds: The Octopus, the Sea and the Deep Origins of Consciousness, William Collins 2017.

⁸For science fiction fans: No authoritarian central super-intelligence will ever rule the planet, as observed by Max Tegmark in Life3.0, Penguin Books 2017.


Medium is an open platform where 170 million readers come to find insightful and dynamic thinking. Here, expert and undiscovered voices alike dive into the heart of any topic and bring new ideas to the surface. Learn more

Follow the writers, publications, and topics that matter to you, and you’ll see them on your homepage and in your inbox. Explore

If you have a story to tell, knowledge to share, or a perspective to offer — welcome home. It’s easy and free to post your thinking on any topic. Write on Medium

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store