The AGI Mandate of Heaven [AGIMoH] – Mapping the Winners of the Race to Build God

The ancient and imperial China, rulers were legitimized by an ideology of the Mandate of Heaven – the idea that heaven selects a rightful virtuous ruler.

Capable and benevolent rulers held this Mandate of Heaven until they lost in battle, or their people fell victim to famine or plague, or some other unfortunate event that seemed to tarnish their divine right to rule.

Like ancient Chinese emperors, AGI company leaders occupy a precarious position of power. AGI company leaders know that – in order to be the one to birth AGI first (the final flex) – they must defeat their rivals not only in technical capability and resources, but also in the perception of their own goodness.

Smart AGI lab leaders who want the highest likelihood of birthing AGI and maintaining power will aim to do everything they can to maintain the AGI Mandate of Heaven – by occupying the space closest to the top right of this quadrant:

Mandate of Heaven - OpenAI, DeepMind, Anthropic, etc
Note: This is an example of what the AGIMoH might look like for the top 5 AGI labs in October 2023. This is not intended to be perfect, but merely representative. Read about the rules for plotting “players” on the quadrant in the article below, and then create your own version of the quadrant with this editable template in Google Slides.

Before creating your own version of where the main AGI labs (i.e. the “players”) fit on the quadrant, let’s explore the axes and how the plotting works:

Understanding the AGI Mandate of Heaven (AGIMoH) Matrix

The Axes

  • Technical Capability to Build AGI: This axis represents all the elements crucial to physically producing AGI. Volume / access / quality of data, talent, financial resources, access to compute and power, etc. The player on the top of the quadrant is the player with the most of these things. The are the most relatively capable.
  • Support of the Public: This axis represents the goodwill of the public, or the relative perceived benevolence of the AGI lap. Even if all AGI labs are treated with suspicion by the public (as nearly all politicians are), then the player on the far right of the quadrant is the one with the most relative support. They perceived as the most relatively “good.”

The AGI lab that “builds god” needs more than capability.

Bands of chimpanzees will kill an alpha male if he is aggregately detrimental to the tribe, ancient Chinese would revolt against an incapable or vicious king.

Similarly, super powerful AGI labs are a few moves away from being commandeered by the military, or torn down by an angry populace who believes that they are not acting in the interest of the public (read: Political Singularity).

Both elements of the AGIMoH must exist in order for the top player to have the best chance of weidling power and influence in the future.

Plotting the Players

For the sake of using the matrix to analyze the AGI power landscape, I do the following:

  • Keep a “Top X” Number of Players on the Field:
    • Limiting the number of plotted players makes the graphic more useful and clear.
    • I think that a “Top 5” or “Top 7” is best, mostly because it seems unlikely that AGI will be birthed by an AGI player who isn’t self-evidently among the top 5 or 7.
    • For the sake of simplicity in this article I’ve stuck with the Top 5 AGI players.
  • The X-Axis Represents Relative Extremes:
    • There will always be a player touching the far left of the matrix. This player is – among the players listed – the one with the least amount of public support / perceived benevolence (again, this is relative to the other players on the field).
    • There will always be a player touching the far right of the matrix. This player is – among the players listed – the one with the most amount of public support / perceived benevolence.
    • Note that this doesn’t mean the public trusts or likes any of these players. All the top players might be feared, even hated. In this case, there is a still a player that is hated or feared least, and they would occupy the position against the right wall.
  • The Y-Axis Represents Relative Extremes:
    • There will always be a player touching the bottom of the matrix. This player is – among the players listed – the one with the least amount of AGI-building capability and resources (relative to the other players on the board).
    • There will always be a player touching the top of the matrix. This player is – among the players listed – the one with the most amount of AGI-building capability and resources (relative to the other players on the board).

I’ll anticipate some of your immediate responses to the image I posted above:

  • “Dan, I see Cohere at the bottom of ‘capability’ – but they’ve raised hundreds of millions!” Yes, but compared to the other players plotter, their funding and overall AGI capability is lowest, so they’re at the bottom of the matrix, because it’s relative.
  • “Dan, I see Anthropic at the far right of ‘public support’ – but not everything really thinks they’re morally good!” Yes, but what I’m showing here is that, of all the five players plotted, Anthropic likely has the most perceived benevolence. Again, the matrix’s extremes are relative.

Bear in mind also that my plotting above isn’t “right” or permanent, it’s an example.

The Value of the AGIMoH – An Amoral Examination of Power

I developed this matrix in order to open an amoral dialogue about incentives and power.

Power struggles quickly devolve into tribal squabbles – where one side is “good” and the other side is “bad,” and one leader is “virtuous,” another is “selfish.”

There is no progress in tribal enemizing, but there is progress in squarely understanding the positions and respective self-interest of the parties involved. Only by doing this can the “other” be treated as something other than “bad” or “evil”, and only by doing this can we establish arenas (as we do in sports, in science, and the best versions of fair markets / justice systems) where competition can be harnessed for a kind of net good.

The Peace of Westphalia, the founding of the United Nations, and the end of many lengthy and complex conflicts have been solved by squarely understanding the motives of respective parties, and orchestrating ways to navigate incentives rather than label parties as good or bad.

As AGI becomes the singular, defining political topic of the century (and likely the last such issue for humanity), there will be many emerging “tribes,” some political, others based around specific AGI initiatives and labs. By understanding the kockeying of power between major players on the AGIMoH, I hope to be able to make their jockeying less of a “good vs evil” story, and more of a reasoned understanding that…

We must deal squarely with human nature and accept that the conatus is king, and trade fruitless tribal conflict for a measured approach to aligning and allying incentives.

Plot Your Own AGIMoH

Use this Google Slides link to create your own version of the AGIMoH.

Remember that if you plot the the players correctly, you’ll always have one player touching the top, bottom, right, and left boundaries. The rest of the players sit in between the extremes, relative to their public support or capability.

Ping me on Twitter with what you come up with, I’m interested to see other perspectives here.

Predictable Dynamics of Power

Much of what we’ll see pan out in the AGI race will be more-or-less identical to how we’ve seen other power struggles – because humans haven’t fundamentally changed very much.

1. Virtue is How the (Relatively) Weak Compete, and Benevolent Protection is How the (Relatively) Strong Compete

OpenAI was founded as the “good little guy,” and quickly changed their tune when power consolidation was possible.

Anthropic spun out of OpenAI, waving the same banner of virtue, only to jump into the same ravenous AGI race, aligning with the same Big Tech allies.

Now OpenAI is in the “powerful” category, no longer an underdog – and predictably they’re focusing on owning the regularity narrative.

Neither of them are “bad,” neither of them were “corrupted,” they are merely operating on the principles of self-interest, just as you and I do. We would do likely exactly what they would do if in their shoes. But it behooves us nonetheless to understand how the powerful and the less powerful will behave.

This graphic from the first-ever full length essay in our AI Power series at Emerj sums up the dynamics we should expect in the years ahead:

Source: Emerj Artificial Intelligence Research

2. Public Support Becomes More Challenging the More Powerful One Becomes

It’s easy to be the good guy (the far right of “public support”) when you’re small.

The little guy gets to say “Great masses! I am one of you!”, and mean it. Their inherent self interest (against current incumbents in power) is in fact in line with more of the populace.

But once in power, like Robespierre, they again do what behooves when: Which generally involves the consolidation of power.

As players ascend in capability, it will become harder and harder to stay on the far right of them quadrant – as new upstarts rally under the “virtuous little guy” banner – and new waves of people believe the “virtuous little guys” are somehow not driven by the same ravenous self-interest that defines the modus operandi of individuals and organizations.


Header image credit: