Although history frames the Second World War as the dawn of the Nuclear Age, it might someday be remembered more as the opening of the age of computing.
Alan Turing – the godfather of computer science – used the computers at Bletchley Park to crack the German ‘Enigma’ encryption, a feat that arguably won the war.
John Von Neumann, mathematician extraordinaire, divided his time between performing complex physics calculations for the Manhattan Project and automating mathematics on the Harvard Mark I — an immediate forerunner of the electronic computer.
Norbert Wiener studied human and machine behaviour, then built a system to model both – creating an anti-aircraft gun that could track an aircraft as its pilot tried to avoid incoming fire from the weapon.
All of these efforts unexpectedly supplied the postwar world with a new apparatus for automating thought. Wiener penned a manifesto – Cybernetics: Or, Control and Communication in the Animal and the Machine – making explicit the informational relationship between ourselves and these new ‘thinking’ machines.
Although very much original to Wiener, cybernetics was simultaneously of its moment, as an emerging intellectual class grasped the significance of the harnessing of information.
A month before the war ended, Vannevar Bush – head of the wartime Office of Scientific Research and Development, and forerunner of The Defence Advanced Research Agency (DARPA) – wrote the highly influential essay ‘As We May Think’. In it, Bush described a machine – the ‘memex’ – using microfiche, the high technology of the time, to provide a personal library with extraordinary resources, all of which could be searched and linked together, making the memex a powerful research tool: effectively, a mechanical implementation of the Web.
The idea that information could change us just as we change it (a core observation of Cybernetics) found its physical expression in the advent of systems – computers – that could manipulate vast amounts of information. While memex remained a thought experiment, the mechanical Harvard Mark I gave way to fully electronic ENIAC, EDSAC, CSIRAC, and, in fairly short order, to a generation of commercial computers, such as UNIVAC, Ferranti Mark I – and an endless series of models from IBM.
Although these machines had all of the same capabilities of today’s computers (while being thousands of times slower and millions of times more power hungry), they were first put to work automating very routine tasks, such as data entry and tabulation of records, accounts and so forth. That they could be used for far more (and soon would) became one of the themes within a landmark series of events that framed the concerns of a community of those who, nearly eighty years later, we can identify as cyberneticists – people who wanted to better understand our relation to information, and our relations to these new machines.
The Macy’s Conferences on Cybernetics kicked off in March of 1946 – just as the world began to assume its new post-War dimensions. Wiener and Von Neuman both had prominent roles; Turing was invited, but could not attend. But rather than a festival for the mathematicians, chair Walter McCulloch purposely championed an interdisciplinary approach: biologists and neurologists would be there to talk about the information flows in the animal, while at the same time, legendary anthropologists Margaret Mead and Gregory Bateson could connect the microcosm of feedbacks within an organism or machine to the organising patterns of culture and society.
Instead of dedicating itself to resolving the answer to any of the questions raised by cybernetics, McCulloch and his conferees instead sought resonances, similarities and lessons common to very differently framed models of information flow – whether in a neuron, a servomechanism, or a tribe. This freedom to disrespect the bounds of any particular discipline propelled participants through 10 events across six years, and arguably had a principal role in shaping our contemporary culture. We have nuanced discussions about the benefits of automation – or what qualifies as intelligence – in part because these topics found free expression in the Macy’s Conferences on Cybernetics.
It didn’t take long to lose some of that nuance. In 1956, only three years after the final conference, a month-long workshop at Dartmouth University introduced the topic of artificial intelligence – embodying the principles of feedback and learning in computational models, yet remaining separate from the world, and above it.
That approach failed to give us intelligent machines. Instead, those first-generation researchers had their greatest success not with artificial intelligence, but in mapping out exactly how poorly we understood the nature of our own intelligence – a revelation that they might have come to more quickly (and less expensively) if they’d listened to the psychologists and anthropologists involved in the Macy’s Conference.
Cut off from its moorings in cybernetics, artificial intelligence achieved a few early gains in the 1960s and 1970s, then stagnated for nearly two decades (a period known as the first ‘AI winter’) until Australian Rodney Brooks recognised something that would have been obvious to the cyberneticists of the Macy’s Conferences – intelligence cannot operate in a vacuum.
Brooks’ new generation of AI-powered robots stumbled through their environments, learning as they went along, feeding what they encountered back into their behaviours, producing emergent qualities that had been impossible to gin up from a purely mathematical and abstract understanding of the world. Brooks’ subsumption architecture recast intelligence as something neither inside the mind nor out in the world, but in their marriage.
Brooks’ research fitted perfectly into the theoretical foundations laid by the two most influential second-generation cyberneticists – Humberto Maturana and Francisco Varela. In their clear and breathtaking 1982 book The Tree of Knowledge: The Biological Roots of Human Understanding, they illustrate the informational self-similarity of all systems that trade information with the world – whether that’s a single neuron, a person, or a culture. That much had already been explored by the Macy’s Conferences, but Maturana and Verela added to this a powerful model to understand the nature of these informational relationships. All systems exchange information with their environments, they reasoned, and every exchange of information transforms both parties to that exchange. Keep that up for a while, and the two parties form a new unity, a structural coupling that has informationally united the two. Structural coupling points directly back to Wiener’s efforts to erase any informational distinctions between animal and machine, but cuts deeper and broader: Any two systems, at any scale, engaged in the exchange of information produce a greater unity, a structural coupling. With structural coupling, Maturana and Varela made ‘we are all connected’ more than a platitude, providing a theoretical basis for the success of Brooks’ subsumption architecture.
In the third and final part of this series, today’s third-generation cyberneticists look to steer themselves and our culture through a world of hyperconnectivity, feedback – and an artificial intelligence, embodying cybernetic principles, that has finally learned how to learn.