<aside> š This is a work in progress, not a public statement.
Please tell us
All discussions of AI, alignment, tech, and society dance around one question: What āgood societyā should we aim for? (Equivalently: āWhat is sacred and special in human life?ā) Powerful tech requires a finely-adjusted direction to its power. As humanity approaches the age of AGI we must reckon with the question of āwhat is sacred about life, which potentially infinite powers should be in service of?ā
Western civilization has a habit of punting this question down the road. Liberalism passed the question off to individuals to make by themselves, with their answers aggregated via markets and votes. Post-modernism went further, saying even individuals couldnāt answer. āTime to buildā accelerationism skips the question, instead glorifying another clearly misalignedāø metric, like ROI or entropy.
Yet, we may finally be able to answer the question. Thereās research that indicates that LLMs can reveal deep lines of agreement on what a good life is, what a wise society would look like, and what is sacred and deserves to be protected by powerful AI.
For instance, some experiments show strong agreement about what's wise for someone (or an AI) to do in different kinds of moral situations. Or that thereās a lot of overlap in how people find it meaningful to live: how is it meaningful to connect with other people? How is it meaningful to approach things like work, love, and art? What relationships are good and nourishing?
Sure, thereās disagreement at the level of political slogans, but thereās a shared vision of what makes for a good life and healthy community, and thatās a lot to build on!
Why, then, do we think thereās so much disagreement about values? For hundreds of years, political leaders have used propaganda to create ideological divisions ā like those between Catholics and Protestants during the 30 Years War, or those between MAGAs and progressives in the modern day USA. The recipe is always the same: whatever side you are on has relatable, sane human values like prosperity, integrity, ambition, and honesty, while the other side is supposed to have alien, dangerous āvaluesā, such as limitless greed, a blind adherence to tradition, or the denigration of the poor. The powerful use this propaganda to raise armies, to win territory, to drive votes. But thereās a side-effect: people become convinced thereās widespread disagreement on human values.
Once you realize how much agreement there is about what a good life looks like, and that political forces have obscured this agreement, a new frontier opens up.
New political systems based on maps of shared values can transcend ideological deadlocksā¶. Social progress is stalled in many areas because the powerful have made idiotic religions out of political parties and cause areas. With a shared understanding of values, we can redefine our political identities and emphasize our individual quests for meaning and sacredness over externally imposed labels like red/blue, or oppressor/oppressed.ā·
Tech policy can have a clear goal. As AI spreads through society, which human jobs and roles should be preserved? The most meaningful! Which human relationships should be preserved? The meaningful ones! Which economic arrangements should be built around human life? Those that support meaningful lives!
A shared map of whatās meaningful to people can make all of this concrete, and give us techniques for measuring our progress towards broadly-supported meaningful lives.
This suggestion is common sense. ****If there will be a division of labor between humans and AIs, shouldn't humans be able to stick with what's meaningful, rather than getting the economic leftovers? What other criteria would make sense?
Markets can be reformed. Many modern systems involve harnessing markets to serve non-market concerns: modern healthcare tries to make markets serve health outcomes. Livable cities try to harness development to serve the flourishing of neighborhoods. We believe a rich understanding of whatās meaningful can put markets in their place: leading to economies that go beyond consumption, atomization, and extraction, and center on meaningful lives and a strong social fabric. As a concrete version of this, we envision collectively-owned LLMs that allocateāµ resources according to what those in the collective find meaningfulĀ¹.
Finally, these maps of wisdomĀ³ can also be used to create wise AIsā“**.** The major AI labs are currently racing towards superintelligence. But intelligence is not enough to create a good future. For this, we need wise systems ā models that understand and develop their values in collaboration with humanity. A map of humanityās values offers a way to develop such systems. These Wise AI models can help us find win-wins in situations where intelligence alone could not.
Wise AI is just a starting point. To sustain this transformation, it must be integrated into a reformation of existing systems of collective action. Imagine we just had Wise AIs. Market dynamics would push against them, towards systems for optimizing existing business incentives. Geopolics, too, will push towards ruthless, military AIs, not the wise ones.
So, we must upgrade models, markets, and governments together ā to something more values-driven, participatory, convergent, and wise ā if we want to make it through.
In other words, we need āfull-stack alignmentā. Thatās good news though!
All the x-risk people say XYZ, to solve this we need coherent collective action, and that requires a positive vision
Focusing on this (big, positive) shift can produce good policy outcomes, and good product directions. Itās way better than focusing on x-risk and dangers**.** You cannot create a good future by only focusing on how to prevent bad ones, because:
(1) Good policy ideas come from positive visions of how things could be mutually beneficial, if they are put in the right relationships. Otherwise, we get regulatory capture, black markets, and warring factions.
(2) Focus on risk narrows peopleās thinking.Fear-driven people create fear-driven political responses, not generative ecosystems.
(3) More generally, fear divides us, whereas hope and positive visions can unite us.
So, let us be driven by love, not fear. We recognize the high stakes of the current moment. But we see the situation as an invitation to deepen our connection and commitment to the sacredness in life, and use this renewed connection as the guide for the path forward.
The need for full-stack alignment is daunting, but also exciting. Let us take the stakes into account, and use this cultural moment to build beautiful things.
Weāre deeply committed to the sacredness of life. We believe being alive in this universe is the highest gift, and weāre committed to honoring the things that make that experience worthwhile ā our drives for intimacy and connection, to expand frontiers and conquer unknowns, to create and to understand, to live in integrity with who we are. We believe these drives for life (our values and sources of meaning) are the fundamental pieces of human flourishing.
Sincerely,
Signatories: JE, EH, OK,
^ add yourselves
<aside> š This will not be published by MAI but in some kind of mysterious and cool way by an unknown group.
</aside>