Imagine a world of Fully Homomorphic Encrypted Decentralised Autonomous Organisations based on Artificial Intelligences.
While it may seem that someone decided to combine as many cringe-inducing contemporary technology acronyms into one article as possible, if we break the concept down into fundamental parts, it seems increasingly unlikely (to me at least) that we won’t meet this creature in some shape or form in the not too distant future.
The fundamental concept here is the conflux of three technologies that are under constant development and improvement:
- Decentralised Autonomous Organisations (DAOs)
- Fully Homomorphic Encryption (FHE)
- Artificial Intelligence (AI)
After running through the general concept of these technologies and how they may interact, we can ask questions like: How would such an AI-driven DAO entity with FHE behave; how would it fit within our social, economic and political environments; how would life change for humans if such things were commonplace; and what kinds of moral and ethical challenges may we face as a result?
Let’s break it down.
Decentralised Autonomous Organisations (DAOs)
A DAO is, at its heart, a program.
Like any program, it has inputs, processing and outputs.
However unlike most traditional programs, a DAO’s program is not run on one computer at a time but on many, through a distributed ledger system such as a blockchain.
In simple terms, the blockchain itself is the computer for a DAO’s program code.
Within a DAO, all nodes in the network work together in processing the data, operations, and outputs of the program. By cross referencing the results of each step with other nodes in the swarm, potentially nefarious elements can be excluded from the final ledger.
What the system loses in speed and efficiency, it gains in redundancy and authenticity. It is in this way that robust and, most importantly, trustworthy programs can be developed.
It’s this sense of trust that gives a DAO its power and autonomy. With the constant scrutiny of the “swarm”, the business risks of delegating certain types of transaction authority to a software program are substantially mitigated to the point where for certain operations, offloading to a DAO could make good business sense.
Such a program could be quite simple; if you do X for me, I’ll give you Y.
Say you have proof of a task being performed, lets label it “a”, and Alice has an evaluation matrix, “B”. Upon processing, payment will be made into a cryptocurrency wallet address of an amount “c” determined by some formula, such as:
a * B = c
From this simple example, it’s easy to imagine more complicated arrangements, chained functions and nested contracts split into projects, departments and vertical markets. These algorithms may include looping and conditional logic, all within the fact-checked and cross-referenced “machine code” of the blockchain.
Rather than assembling around traditional organisations, workforces of the future may be motivated around whichever DAO is most desirable or accessible to any particular individual, given their circumstances.
Fully Homomorphic Encryption — “Fully Hidden Everything”
In the example of the basic blockchain contract, every node in the swarm is able to take a peek at what’s going on inside the (rather simplistic) DAO, which is not fantastic from an operation and information security standpoint.
One way to improve security might be to encrypt the data inputs, however in order for this to work, everyone involved with the DAO would need to have access to various keys to work with the organisation. This has the side effect of making the organisation a central authority, which goes against the decentralised model.
What if you don’t want your data to be revealed to the DAO, and neither does Alice? Both of you are only willing to provide the encrypted data, not the means of decrypting it. You don’t even want to know about each other.
Additionally, the DAO you work for doesn’t want the internal workings of their secret formula to be known, so neither you, Alice nor the blockchain should be able to fathom the algorithm by simply looking at it.
To this end, Fully Homomorphic Encryption (FHE) is a technology that is designed to take encrypted data, run encrypted operations on it, and return an encrypted result without having to decrypt the data or operations at any point in the process. It doesn’t matter if the program executes on my home computer or in my mortal enemy’s dungeon server rack. None of the code, input data, or output data is ever revealed, not even in transient memory.
For an external observer, whether it’s a human actor dealing directly with the FHE cryptosystem or an automated agent, any DAOs built on a fully homomorphic blockchain would look more like uniformly distributed noise than any kind of recognisable program.
Future human workers lining up at the Blockchain each morning may not know who they are working for, or why they’re doing the work they signed up for. They could be tapping out a subroutine for a mining mission to the asteroid belt, or developing a computer supervirus to take down a government.
Only the DAO’s master programmer would have the full picture of what’s really going on…
Or would they?
Advances in AI have led to many different ways of creating software capable of optimising its behavior for specific tasks by creating predictive models. By “feeding back” the result of the prediction vs reality, an AI agent can improve future predictions, becoming more accurate over time.
Consider a DAO that uses AI to optimise for financial return, as mentioned in the opening statement, with the added benefit of FHE.
What would we name this thing? How about a Homomorphic Intelligent Distributed Organisations for short (HIDO)?
Such an entity has no fixed “procedural” program, perhaps all HIDO does is create semi-random (perhaps evolutionary) contracts based on feedback loops that favour increased profits and revenues over time.
The master programmer may even choose to take a small cut of each transaction as a reward.
By utilising fully homomorphic encryption, by creating a system that autonomously seeks to optimise for profits, and using a distributed ledger to verify transactions and program operations, the master programmer has very cleverly stumbled on a gold mine of ever-increasing wealth.
Now the programmer can kick back by the pool atop a personal skyscraper made from gold bullion and bask in their good fortune while HIDO brings home the bacon.
Could such basking by the pool really be done with a clear conscience?
What could possibly go wrong?
Obligatory Doom and Gloom
Honey, I shot the kids
Woops! The HIDO started creating contracts targeting terrorist organisations. Using increasingly specific contracts, a growing plague of tragedies occur globally. No one has any idea who is paying who to kill who, but somewhere there’s a master programmer taking a fair cut for developing a system to broker such transactions.
Got anymore of those… DAO’s?
Woops! The HIDO created a perfect escrow service for massive illegal shipments of contraband. No longer having to operate hundreds of fronts and traditional accounts to redirect and launder their ill-gotten gains, for a tiny fee criminal masterminds are guaranteed a high degree of anonymity for their activities, making it that much harder for police and authorities to track illegal black market trade.
Who done it?
During the eventual trial for being a criminal mastermind, the programmer may argue:
- The program was never written to perform illegal transactions. It was trained to do so through “free market economics” and positive financial rewards.
- The program never ran on any hardware owned by the programmer. It was however run simultaneously by millions of people who consented to arbitrary software of unknown origin being computed on their hardware.
- The programmer did not personally know, or seek to know, any of the criminals who used the program and its contracts.
What crime or ethical breach, if any, did the programmer commit?
If there was no crime, was there at the very least a moral imperative for the programmer to act in a particular way in the broader fabric of civilisation?
And what then for the people responsible for executing “HIDO” software, wittingly or unwittingly?
What, then, could possibly go right?
To my mind there exist some legitimate use cases:
- Witnesses on witness protection programs, who may in the future be able to provide certified testimony through secure cryptographic channels without exposing their whereabouts.
- Whistleblowers and defectors who may set up personal protections, “dead man’s switches”, and other release mechanisms to protect their safety using a HIDO-type service.
- Corporate agents dealing with trade secrets on large, complex projects, hoping to coordinate a large workforce while protecting their broader intellectual property.
- Electronic voting, anonymously to avoid repercussions. If taxes are eventually held in escrow in a blockchain-linked account, your vote could even be equivalent to directly funding the cause(s) of your choosing.
While there are obvious dark alleyways to look out for, I have no doubts there are many use cases for HIDO technology which could provide legitimate benefits — socially, politically, legally — that are as-yet undreamed of.
These ideas do not exist in a vacuum
There have already been inroads made in implementing AI DAOs. For instance #artDAO is a project based on creating an AI DAO that creates and sells its own art.
The concept of combining FHE technology with a DAO is also not unique. However the technology progression is not so clear-cut. Performance is seen as a major drawback. There are some constraints imposed by the mathematical complexity of FHE on the scalability and types of operations which can feasibly be performed.
For FHE, from my observation at least, while solutions exist today there seems to be a sense of holding back, that the “next big thing” might be just around the corner.
This “Brief Survey” by Lucas Barthelemy at QuarksLab offers a decent summary of the issues currently faced:
For example, a few members of Microsoft research team implemented a Logistic Regression Model on encrypted data using their homomorphic library SEAL. Although the function to be evaluated is simple and the amount of data small, this attempt gives a good idea of homomorphic scheme capacities. Considering memory usage, 1 Mb of data results in more than 10 Gb of encrypted data. As far as computation, addition is still manageable as it takes under 1 ms. However, multiplication takes over 5 seconds per multiplication
There is clearly a point where sparkly-eyed pronouncements of “technology advancements” crashes headlong into the hard facts of the universe and its fundamental mathematical limits (a lesson I learned the hard way (or fun way depending on perspective).
Simply put, there may well be no fundamental “leap” that dramatically improves the efficiency of homomorphism.
However for a DAO, where the tradeoff for performance in exchange for trust and reliability is already accepted, this becomes much less of a concern. Indeed, computing homomorphic functions could itself be considered Proof of Work in a blockchain validation scheme. A flaw becomes a feature!
Regardless of how it comes out in the wash, the continuing refinement or improved adaptation of the underlying technologies makes it almost certain, to my mind at least, that someone will attempt to create a “HIDO” program in the near future.
The building blocks are falling into place already.
It’s the pace of this development which is the most interesting part, as the technology is advancing much faster than the layman’s ability to keep up, and in so many directions even the ability of seasoned technology professionals is stretched.
With such a rapid onset, approaching from a number of directions unseen by many, will society actually recognise the first significant HIDO, or family of such technologies?
Do we know what such a creature even looks like or how it behaves?
Are we ready to deal with the consequences of its behaviors?
My gut instinct tells me that ready or not, we’ll be finding out sooner rather than later.