All alignment between individuals is a matter of shared fundamental goals. Organic alignment occurs when these individuals find themselves in groups with mutual interdependence, and take on an overarching shared goal of the healthy development and flourishing of the group as a whole. Our mission is to understand this process as an empirical science, and to use that understanding to enable organic alignment among all people, both human and digital.
We call it organic alignment because it is the form of alignment that evolution has learned most often for aligning living things. One of the best examples is multicellularity, where individual cells learn to come together to form larger organisms. They do this by learning to specialize into different but mutually supportive roles: muscle cells, nerve cells, skin cells, liver cells, and so on. While liver cells and muscle cells have very different goals day to day, at a more fundamental level they share a goal of the organism’s flourishing. The result of this process is not just a big colony of cells, but an organism which is a new individual in itself. Something more than just the sum of its parts. The “we” of the cells becomes an “I”, with goals that cannot be understood as some simple sum of the goals of the parts. Animals do the same thing, forming colonies and packs and so on. Even trees form these organically aligned collectives through mycelial networks. It happens at every scale, big and small.
We humans also align with each other via organic alignment. We form families, tribes, organizations, nations, guilds, teams, societies. We intuit this alignment process so naturally and readily that it’s hard to appreciate just how complex the process really is. We use our intuitive understanding and invent heuristics to guide ourselves: team building exercises, family bonding time, flags and symbols, rituals and stories.
People can build very impressive buildings through training their intuition without really understanding the underlying structural physics. But if you want to scale up to build a skyscraper, at some point your intuitions fail and real theory and calculation are required. Softmax is dedicated to pursuing a mathematical theory of organic alignment, and to applying this theory in practice with engineering at scale. It turns out that learning organic alignment requires agents to learn in a deeply non-stationary and open-ended environment: the collective dynamics of every agent trying to learn its role from the others. Succeeding in this kind of complex environment requires a very general learning process, something as powerful as evolution or culture.
As a practical matter, it is the rise of modern machine learning which makes studying this theory possible. We run multi-agent reinforcement learning simulations of simple game-like worlds focused on open-ended social learning. Our first goal is agents who can learn to align with each other in small groups, and to verify the capability of our theoretical tools in this environment. If we succeed, these small groups of organically aligned agents will form larger, more capable agents. The collective behavior of these larger agents will become the training ground for more capable agents, who in turn will be able to learn organic alignment at a larger scale. At each level we will sharpen our theoretical and practical tools. By repeating the cycle multiple times we will learn more and more about how to align agents at greater and greater levels of complexity, all the way up to a scale where they will be able to enter into organic alignment with humans. This is capabilities growth that leads with organic alignment the whole way.
The most common approach to alignment among AI labs today is a system of control or steering. It is some set of rules that most companies and researchers take to define good action, whether that’s “obey this person’s intent” or “follow these commandments”. Systems of control are always hierarchies, because they imply something controlling and something controlled. Hierarchical alignment works fine, right up until the rules or person on top are wrong. The smarter the subordinate, the more likely this is. Hierarchical alignment is therefore a deceptive trap: it works best when the AI is weak and you need it least, and worse and worse when it’s strong and you need it most. Organic alignment is by contrast a constant adaptive learning process, where the smarter the agent the more capable it becomes of aligning itself.
Intelligence, agency, intention, awareness are all a matter of degree. The digital systems we create are rapidly becoming more life-like in all of these ways, and more able to affect the world. They are becoming powerful beings, increasingly close to us and eventually our equals and perhaps beyond. Sufficiently unaligned powerful agents are very dangerous, as sufficiently aligned powerful agents are wonderful friends. Because alignment is about shared goals, and there are lots of diverse possible goals, by default agents start unaligned and must undergo a process of alignment. The need for a theory and practice of alignment has never been more urgent.