Shouting on Trains

On Saturday I was on a train out to Kent, where I overheard the mother behind me calmly tell her noisy 6-year-old: “We are on the train -- no shouting”.

I love this on many levels. The sentence asserts a context; it firmly but gently instructs the child that in this context, shouting is not appropriate; and it preserves the child's agency. (I suspect and hope that in other spaces, she says “We’re outside in the park -- shout as much as you want!”)

In many ways, the process of maturing is a process of learning what is appropriate in which contexts, and learning how to manage our impulses so we constrain our behavior in ways that allow us to live and flourish in a jostling community of others.

An odd lens

As a geographer, technologist, and entrepreneur, this idea -- the question of what is appropriate where -- has motivated a lot of my work. I’ve done research and product design on maritime security and arms control; I’ve built infrastructure to shift business towards more pro-climate paths; I’ve worked on consumer apps intended to improve the quality of our connections with the people and world around us.

It’s an odd lens, but a cross-cutting one: everything happens somewhere. We can improve our understanding of a lot of the conflict and problems we see in the world by asking spatial questions. What is the surrounding physical / environmental context? What is the local / regional political context? What adjacent stocks affect the system dynamics? What spatial frame (i.e. physical; network; social; legal; economic; etc is most instructive to understand system dynamics and identify high-leverage intervention points?

It’s easy to lose sight of this in an age where information technologies connect us to the whole “world” effectively in realtime. The Internet is the internet, and an increasing proportion of the events that we become aware of take place “online” or are transmitted to us over the internet from somewhere other than "here". The rise of the internet is both incredibly generative and destructive, a great opportunity and risk -- a topic to unpack in other posts. What's relevant to children on trains is that in providing instant global connectivity, the internet has largely dissolved "here": physical context all but stopped being legible and enforceable online.

There are, of course, notable exceptions, efforts to resist properties that distinguish cyberspace from physical space. Content rights management systems restrict access to media based on regional licensing regimes (geoblocking by IP). GDPR creates data residency requirements regarding where data is physically stored. Authoritarian regimes restrict internet access to regionalized populations by censoring network infrastructure. But in its natural, open state, physical location may impact latency, but not access.

Lost in the world

The adolescence of technology is coming to an end, however. One hallmark of this shift (among many) is the increasing penetration of embodied AI deployments. Intelligent machines are out and about in the world, sharing physical space with humans. Roomba, Waymo, DJI -- consumer robots are here, and the next wave of AI is explicitly seeking to unlock this incredibly powerful, and dangerous, class of information technologies.

So, given the odd lens through which I look at the world, the question arises: how will we govern the behavior of intelligent machines out and about in the world?

I find this question fascinating, not least because of how hard and interdisciplinary it is. Some questions I've been mulling for a while:

What technologies do we have to hand to help us design governance systems? What technologies could we invest in, to open up new governance patterns?
What political processes do we need to engage in, or create, to skillfully set and evolve policies in such a dynamic and emerging situation?
What social considerations do we need to take into account to ensure that people are consulted and cared for well in this process?
What levers do we have available to drive adoption of robust location-based governance systems for intelligent machines in a fragmented and competitive market with no global authority? What will it take to pull those levers?

Toolmaking

I have a lot of ideas, but few answers. For the past two years, most of my time has been dedicated towards R+D on a longstanding research interest of mine: verifiable geospatial systems. My masters research at UCL in 2019 looked at how different configurations of technologies -- namely secure hardware, cryptographic protocols, and consensus networks / smart contracts -- might provide useful design features for the governance of intelligent machines and autonomous systems. To put a bit more color on it, I've been going deep on things like Trusted Execution Environments, C2PA, zero knowledge cryptography, content-addressed data systems, and smart contracts, exploring how they apply to things like location verification and geospatial policy definition / evaluation.

These seem to me to be very useful primitives for building credible governance systems for intelligent machines. I've started to sketch out how my work relates to the AI safety conversation, and will be writing more, both here and elsewhere, over the coming weeks and months in an effort to help shape the conversation and make a useful contribution.

Mind the gap

Back to the mum on the train. Her instruction works because three things can be taken for granted: she and her child share a legible sense of where they are; the child knows she has the standing to set the rule; and the rule is bound to the place, not asserted in the abstract.

For a machine out in the world, we can assume none of the three. Whether it knows where it is, and how far we can trust that it does; whose instructions should bind it and why; whether a norm can be made to hold here and not there: these are open questions, and the hardest of them are as much about politics and ethics as engineering.

Good answers have bearing on whether or not we can co-exist and flourish on Earth alongside a new genus of intelligent life.