The world’s leading publication for data science, AI, and ML professionals.

Algorithmic Corporations and the Post Death Society

As humans we possess a variety of means of leaving behind a legacy. This includes the bequeathal of genetic material to our offspring, but…

As humans we possess a variety of means of leaving behind a legacy. This includes the bequeathal of genetic material to our offspring, but also such things as businesses handed over to seceding generations and the creation of college endowments, to name just a few. All such legacies share a common and vexing trait – like disobedient children, that have a tendency to stray from their originators intended course. Being entrusted to and run by other humans, such legacies can be co-opted, steered in new directions, or abolished altogether. By their very nature, legacies are a mine-field of principal–agent problems. As such, they rarely survive beyond a few generations of stewards. That may be about to change. Enter the Decentralized Autonomous Organization.

For those unfamiliar with the term, Decentralized Autonomous Organizations or Decentralized Autonomous Companies (DACs for short) are organizations that rely upon an algorithm to maintain their charter or objectives. To grasp this concept, it can be first helpful to understand that in many ways a multi-person organization represents a kind of super-intelligence, one comprised of humans bound together to promote whatever objectives is spelled out in the organizations mission statement.

The humans who make up an organization operate as a kind of meta-entity, one that possesses greater powers than any single human comprising it. OpenAI founder Greg Brockman notes the similarity between creating a benign AI super-intelligence and creating a benign corporation. In both cases, one is faced with the challenge of spelling out an objective function to an entity whose intelligence is likely to be greater than that of a single human individual. Ever since the first humans banded together to form tribes, we have been capable of producing such super-intelligent cooperative entities. A collection of humans is almost by definition smarter and more powerful than a single person. Steering those multitudes is the duty of an organization’s charter, CEO, or board of trustees. But imagine for a moment if the governing body of an organization was no longer human? If for instance, a computer algorithm was capable of playing the role of strategist and minder to an organization.

Though this may read like science fiction, already cryptocurencies represent a similar kind of entity. Lacking any form of human governance or oversight, they instead rely upon a distributed ledger to ensure their objective functions are being met. In this way, Bitcoin, Ehtereum and other cryptocurrencies are these first members of what we might think of as the post death society, one that survives at the intersection of the human and digital. It is also is likely to be the future of what we think of as legacies.

Imagine that a person’s dying wish was to create a non-profit that would focus on saving as many dolphins as possible. In today’s terms, one would create the organization or foundation at some point prior to death, and than pray that the people are who entrusted with fulfilling its objective are reliable. But with humans at the helm, whose to say the organization wouldn’t be diverted to some other purpose? With DACs and DOAs, one eliminates the middleman(literally!) and the charter itself is responsible for fulfilling the organizations objectives. Compare the goal of saving dolphins to that of beating an opponent at the video game Starcraft or DOTA II. These might seem entirely disparate concepts at present but for those familiar with dynamic programming, they both represent a Markov Decision Process, that is, agent is operating under uncertainty, enacting strategies in a large continuous action space, with a means to assess its success or failure.

I use this example, because already state-of-the-art Reinforcement Learning algorithms have succeeded at beating humans at games taking place in challenging Markovian environments. It may not be a far stretch for such algorithms to be repurposed towards saving dolphins or any other objective function spelled out in a companies’ charter. Supplied with a model of the environment and relevant actions available to it, the algorithm could be set to work pursuing the organization’s mission statement, creating contracts and dispersing funds electronically as necessary. So long as it could evaluate the success or failure of its strategies through external means, such as accessing relevant statistical data on dolphin populations, than reinforcement learning would allow it to hone in on successful strategies and eliminate less productive ones. Even at present, most of the actions necessary for a company or organization to succeed in its mission can be undertaken digitally via email, text, or wire transfer. This is because almost all forms of actual physical labor can be sub-contracted out to the other organizations that actually hire people to do the work. Advances in Natural Language Processing coupled with reinforcement learning will expand this algorithmic capability in the coming decades.

It is worth asking if we would even know if such an autonomous organization was already in our midst, saving dolphins by the boatload. The electronic paper trail might very well be indistinguishable from other companies and organizations, at least to a point. This is in fact the script to an engaging sci-fi novel by Mathew Mather called Darknet. Just as we now show little surprise when algorithms beat human players at games like chess, GO and DOTA II, it is likely that such autonomous companies and organizations will prove more efficient and successful than their human counterparts in the long term. That is for the same reasons they have defeated human gamers – any algorithm whose success in an environment scales with the number of microprocessors employed, for example PPO and A2C, is likely to dominate against humans given sufficient compute and training time. The emergence of DACs and DOAs is therefore likely to challenge our deepest notions of legacy, organizational governance and even mortality.

If an individual’s deepest cares, ambitions and missions in life could be written in silicon, continuing to influence physical reality long after their demise, to what degree would they have truly passed away? Since it is already common to think of the body as merely a vehicle for carrying out the aims of the mind, would the person still be alive in some meaningful sense if those precise aims and ambitions could be transferred to another substrate? And whose is to be held accountable if those aims contravene the moral authority of a given Society? Such vexing questions are likely to form the mainstay of political and judicial discourse in the coming decades as we move into the uncertain terrain at the nexus between the digital and biological worlds.


Related Articles