← Blog

Some questions about decentralized personal computing

· ai, decentralization

Personal computing, but decentralized: back to the future. The next iteration for personal computing after centralized online ecosystems. Variants of this idea seem to be in the air, but I have not yet seen an articulation of quite what I’m picturing, or much discussion of what the problems to solve are.

Why are these ideas in the air? Well, first, our online world right now is dominated by huge closed tech ecosystems controlled by increasingly unaccountable mega-corporations, which in turn have shown themselves eager to line up in support of fascism. There is a groundswell of interest in taking control back through something resembling democracy. Perhaps we can – and definitely should – use actual political democracy to force big tech to act in the public interest. But in addition, perhaps we can use technology to build an alternative.

Second, thanks to the latest AI agents, it has gotten dramatically easier to build and run custom software. Many people who would not have considered doing so are now creating and running custom software, because they can. In over two decades of programming and over a decade as a professional software engineer, I never ran my own software – but now I do. It’s easy to extrapolate from this to envision a world where everyone just has great custom software to meet their exact needs. Maybe we get there, but I don’t think it’s that straightforward.

There’s a decentralized computing world to build and I think the possibilities are exciting, but there’s a lot of problems to solve on the way, and ones to solve if or when we do get there. I don’t have much of a narrative to offer yet, but in this post I want to think out loud about some of those problems. Upfront caveat: some of these may already have good solutions, I am not claiming otherwise!

Running it

Our modern computing doesn’t happen just on our personal devices: we expect apps to run in the background, to be able to keep track of things across our phones and laptops, to work even when our devices are off. It does not seem terribly likely that everyone is going to run their own servers. So that means running these systems in the cloud.

In this world, you’re not running just one little app, but you’re potentially running a whole suite of your own apps, connecting to storage, connecting to backups. The more that’s happening here, the more complex it is for a regular user. And while software engineers might anticipate and have our AI agents address realistic failure modes, most regular users won’t – so whatever system they’re building or deploying in needs to anticipate those needs and protect them.

What do I mean by failure modes? Data shouldn’t be dual written when there’s connection issues. Apps should not put their user’s internal data at risk from the outside world. Apps should not put other apps’ or shared data at risk. Apps shouldn’t hog resources. Etc.

Beyond that, there needs to be basic observability and there need to be meaningful backups as a baseline.

Public and private

People want to share. If I make a game, I might want my family and friends to be able to play it too. Perhaps I host a webpage with my favourite recipes I made this year, and I want to be able to link that on my social media. It needs to be easy to tune from personal only to privately shared to public.

Identity is challenging, and everyone individually setting up identity and authorization structures on a per-app basis is not tenable for adoption at scale. There need to be ways that people can securely grant others access to something in their environment, whether at a data level or an app level.

Platform and trust

So far we’re assuming that we’re talking about software that a user wrote, or that they had AI write for them. I don’t think this is a good assumption for most people. Some domains are simple enough and regular folks who know their needs well enough can use competent AI agents to set up something good for themselves without a lot of effort. But in many cases, building something that is a joy to use and which anticipates non-obvious needs that arise in using a system isn’t easy.

Most people don’t want to build software! They want to solve problems they have. They want to do things using tools that work. Or they want to use exactly the app that a friend of theirs shared with them. Or that someone recommended…. And here we get into problems.

How do we distribute software in such an ecosystem? If someone makes a productivity app that meets their needs and wants to share, how do they do it? What ensures it’s safe enough for others to use? There need to be mechanisms to facilitate trust.

Creating good software is, well, work. Is there a mechanism for people to sell an app, so that there are incentives for software to be created?

Of course, people will want to start with apps that exist and update them for their own needs. What does it take to remix something, and can the result itself be shared? What about when the original app was paid?

Security

End-users don’t know the questions to ask to prevent security and privacy surprises, so any systems we create need to anticipate and design around security for them.

We need to consider how such software can be kept up to date, either with updated features or for security; and these updates need to not be surprising, or themselves a vector for security issues.

In a centralized system, there’s more resources to monitor issues and address threats, or at least to do rapid updates when the latest vulnerabilities are discovered by another frontier model. If a million people are instead running their own copies of a vulnerable piece of software, how do we avoid disaster?

Architecture and standards

What even is a decentralized computing platform? There’s many layers from identity to trust to the actual computing and storage. A system here could take several layers and create a new, closed ecosystem. Perhaps there is a way to package up this kind of software in such a way that it is cleanly portable across different ecosystems – this would require developing standards and protocols to enable this kind of interoperability.

Devices

The common denominator of personal computing today is the smartphone, which is dominated by two closed ecosystems. Could a decentralized computing world with app stores of some sort enable hardware that is specifically oriented to running apps from this ecosystem? What would that require in terms of basic expected functionality, interoperability with existing systems, and security in particular?

Bad actors

The big tech ecosystems care about their profits, and one consequence of that is they work to safeguard their ecosystems from bad actors seeking to harm their users. In a decentralized computing world, well-resourced bad actors will develop malicious software, seed it with fake users, and try to snag victims. Who and what fights against this? And if they do get in, how much damage can they do?

Regulation

The narrative around decentralized software that I started with was around big tech and about customizing software. But people have different motivations for getting away from that kind of control, and one of those is to evade regulations and laws, which big tech enforces in some measure. In a decentralized software world, what mechanisms are there to address people using these systems themselves to do harm? Regulators will start to pay attention as it gets bigger; what is the interaction between regulators and decentralized software platforms and ecosystems?

Power

What does success look like and for whom? What role do the current big tech players take on in this new ecosystem – and can their resources, technology, and market power become constructive and collaborative? If decentralized personal computing takes off, what kind of individuals, organizations, and companies end up with power?

6 likes2 reposts1 quote

Discussion

    1. It feels inevitable that app-shaped objects will become much more democratized, and there’s probably orders of magnitude more people living in that future than there were just a couple of years ago. But it seems to me it’s not easy and not even necessarily good to get that to consumer scale2/2

      repliesreposts2likes