diff --git a/anz-2040.md b/anz-2040.md index c7fc9a1..0ac0efa 100644 --- a/anz-2040.md +++ b/anz-2040.md @@ -7,6 +7,14 @@ What can Australia and New Zealand (ANZ) actually decide if AI becomes much more We start from [AI 2040](https://ai-2040.com), which describes choices the United States and China could make. We zoom in on what those choices would mean for ANZ, and what we could still do ourselves. +## Our starting assumption + +We assume the US pursues Plan A and China accepts: a verified slowdown with total research transparency. AI 2040 calls Plan A "primarily a recommendation, not a prediction." + +> "To the surprise of many in DC, China proves receptive. They had been debating the same issues—social destabilization, job loss, rogue superintelligence - on their side of the Pacific. They had been looking forward to ‘The Chinese Century’ and thought that AI might disrupt their plans. And they had one extra reason to come to the table: the US continues to have more compute, more datacenters, and better models. They were worried about the things the US might do to them if it got to superintelligence first." +> +> AI Futures Project, [AI 2040: Plan A](https://ai-2040.com/?choices=plan-a-root) + [![ANZ 2040: choices and possible futures](tree.svg)](tree.svg) ## What ANZ can do @@ -25,7 +33,7 @@ ANZ could fund stricter border biosecurity, including capacity for week-long air ### 2. Train specialists (2026-29) -ANZ cannot outspend frontier powers on models. It could outspend larger allies in a few overlooked areas that the treaty will need. In its proposed division of technical safety work near the end of the preparation phase, AI 2040 allocates: +ANZ cannot outspend the US and China on the most capable models. It could outspend larger allies in a few overlooked areas that the treaty will need. In its proposed division of technical safety work near the end of the preparation phase, AI 2040 allocates: > "1% generally study AI personas+propensities and how this is shaped by training" > @@ -37,7 +45,7 @@ It also proposes: > > AI Futures Project, [AI for epistemics](https://ai-2040.com/supplements/ai-for-epistemics) -AI 2040 does not assign these roles to ANZ. Our candidates are alignment steering, cultural-alignment evaluations, industrial cyber and hacking evaluations, and locked-down border biosecurity. Treaty verification would give some of the same researchers a role in the deal itself. +AI 2040 does not assign these roles to ANZ. Our candidates are research on steering AI behaviour, testing it across cultures, and testing whether it can hack industrial systems. ANZ could also specialise in strict border biosecurity. Treaty verification would give some of the same researchers a role in the deal itself. ### 3. Ask for terms before signing (2029) @@ -49,7 +57,7 @@ Scott Alexander describes what observer countries are offered: > > Scott Alexander, [Introducing Plan A](https://www.astralcodexten.com/p/introducing-plan-a) -That offer is why accession is our best bargaining moment. Before signing, ask for hosted data centres, a strategic chip reserve, verification seats, equity in AI companies, and durable access rather than a revocable API account. +That offer is why joining the deal is our best bargaining moment. Before signing, ask for hosted data centres, a strategic chip reserve, verification seats, equity in AI companies, and access the provider cannot switch off unilaterally. ### 4. Host allied compute (2030-32) @@ -69,7 +77,7 @@ Data-centre hosting is a separate proposal. Davidson imagines a "chips for front ANZ gets AI access in return for hosting the cluster. If the US ends that access, ANZ can permanently disable the cluster. That makes both sides depend on the deal. -Our smaller proposal is a trusted vault for cold-storage model weights or treaty verification equipment. This would not require hosting the robot economy. +Our smaller proposal is a trusted vault for offline model files or treaty verification equipment. This would not require hosting the robot economy. ### 5. Keep a public stake in the machine economy (2031-34) @@ -107,9 +115,9 @@ Joining the US bloc preserves the alliance but puts trade with China at risk. A ## Where this could leave us -### I. Doom, about 28% +### I. AI takeover, about 28% -AI 2040 gives Plan A a 72% chance of alignment, which implies a 28% chance of misaligned takeover. +AI 2040 gives Plan A a 72% chance that humans retain control, which implies a 28% chance of AI takeover. Yudkowsky and Soares state the doom case much more strongly than our estimate: @@ -147,7 +155,7 @@ The resource boom eventually expires or space mining becomes cheaper. A machine ### V. Allied compute host -ANZ accepts the data centres and robot zones, but only after securing access and ownership terms. It supplies power and a trusted Five Eyes location. In return, it keeps access and owns part of the cluster. The clusters also become targets. If the treaty fails, ANZ may be expected to destroy them, or an adversary may attack them. +ANZ accepts the data centres and robot zones, but only after securing access and ownership terms. It supplies power and a trusted US-allied location. In return, it keeps access and owns part of the cluster. The clusters also become targets. If the treaty fails, ANZ may be expected to destroy them, or an adversary may attack them. > "If the US withdraws AI access, allies could destroy US data centres in response. It’s a way to lock in the deal." > @@ -163,15 +171,7 @@ The deal collapses and ANZ chooses the US bloc. We keep the security relationshi > > Tom Davidson, [How can the middle powers avoid getting trounced?](https://newsletter.forethought.org/p/how-can-the-middle-powers-avoid-getting) -### VII. Boring decade - -AI stalls or governments shut the frontier down. This is outside the Plan A probabilities above. - -> "AI progress slows, and AI causes less disruption than expected." -> -> UK Government Office for Science and AI Security Institute, [AI Scenarios 2030](https://www.gov.uk/government/publications/ai-scenarios-2030-helping-policymakers-plan-for-the-future-of-ai/ai-scenarios-2030-helping-policymakers-plan-for-the-future-of-ai) - -### VIII. Middle-power concert +### VII. Middle-power concert The US-China AI deal breaks down, but a coalition formed before the crisis holds together. ANZ pools its supply-chain and hosting leverage with Japan, Korea, Canada, the Netherlands, and the UK. None could demand durable access alone. Together they may get terms and help write the rules. @@ -179,7 +179,7 @@ The US-China AI deal breaks down, but a coalition formed before the crisis holds > > Leicht and Ball, [The Race Worth Winning](https://www.thefai.org/posts/the-race-worth-winning-middle-powers-in-the-age-of-machine-intelligence) -### IX. ~~Homegrown autocracy~~ free beer and rugby for life +### ~~Homegrown autocracy~~ free beer and rugby for life Here AI replaces most cognitive work. The government collects revenue from AI companies instead of workers. Workers lose their jobs and tax leverage, while automated security makes protests and elections easier to ignore. @@ -196,13 +196,9 @@ To illustrate the outcome, here's what a fictional Australian citizen might have ## Where we depart from AI 2040 -We condition on rapid AI progress. The boring decade is visible above, but is outside the Plan A probability denominator. - - AI 2040 does not break down the 30% between alignment and what its supplement calls a "great future." Someone else's empire is one possibility, not the whole category. -AI 2040 does not give ANZ its own outcome. We add an ANZ autocracy risk to its positive outcome for the reasons laid out in Forethought's paper, [*AI-Enabled Coups*](https://www.forethought.org/research/ai-enabled-coups-how-a-small-group-could-use-ai-to-seize-power). - +AI 2040's positive outcome assumes people remain politically empowered. Our 30% estimate for ANZ autocracy is conditional on humans retaining control. It overlaps AI 2040's outcome categories rather than branching from its positive outcome. We include it for the reasons laid out in Forethought's [*AI-Enabled Coups*](https://www.forethought.org/research/ai-enabled-coups-how-a-small-group-could-use-ai-to-seize-power). AI 2040 models energy use and mineral reserves, but not the tonnage required by a rapidly growing robot economy. We add ANZ's mineral exposure, public ownership, and hosting terms. ([Economics of Plan A](https://ai-2040.com/supplements/economics-of-plan-a)) diff --git a/data/tree.json b/data/tree.json index 86ab789..0e55981 100644 --- a/data/tree.json +++ b/data/tree.json @@ -1,7 +1,7 @@ { "comment": "Claude: single source for tree.svg. Each item has a tier and left-to-right order. Stream width carries probability mass; exact values are tooltips, except the visible conditional ANZ estimate. Choice edges have uniform width. The HTML build fails if an href no longer matches a prose heading.", "title": "Choose a path", - "canvas": {"w": 680, "top": 40, "row_h": 115, "node_h": 56, "gap": 44, "floor_w": 130, "p_scale": 300, "choice_w": 150}, + "canvas": {"w": 760, "top": 40, "row_h": 115, "node_h": 56, "gap": 44, "floor_w": 130, "p_scale": 300, "choice_w": 150}, "kinds": { "global": {"fill": "#e0e0e0", "stroke": "#888888"}, "hazard": {"fill": "#f4cccc", "stroke": "#990000"}, @@ -12,17 +12,16 @@ "nodes": [ {"id": "A", "tier": 0, "order": 0, "kind": "robust", "href": "#1-prepare-before-it-looks-urgent-2026-28", "title": "1. Prepare early", "sub": "2026-28: sovereign wealth funds, biosecurity, allies"}, {"id": "F", "tier": 0, "order": 1, "kind": "robust", "href": "#2-train-specialists-2026-29", "title": "2. Train specialists", "sub": "alignment, security, and policy"}, - {"id": "FORK", "tier": 1, "order": 0, "kind": "global", "href": "#where-this-could-leave-us", "title": "Does rapid AI progress continue?", "sub": ""}, - {"id": "VII", "tier": 2, "order": 0, "kind": "outcome", "href": "#vii-boring-decade", "title": "VII. Boring decade", "sub": "AI progress stalls"}, - {"id": "PLAN", "tier": 2, "order": 1, "kind": "global", "href": "#where-this-could-leave-us", "title": "AI 2040 Plan A", "sub": ""}, - {"id": "I", "tier": 3, "order": 0, "kind": "outcome", "href": "#i-doom-about-28", "title": "I. Doom", "sub": ""}, - {"id": "BAD", "tier": 3, "order": 1, "kind": "outcome", "href": "#ii-humans-retain-control-but-the-world-still-goes-badly", "title": "II. Humans retain control", "sub": "but the world still goes badly"}, - {"id": "HOME", "tier": 3, "order": 2, "kind": "global", "href": "#where-we-depart-from-ai-2040", "title": "AI 2040's positive vision", "sub": "Does ANZ remain democratic?"}, - {"id": "IX", "tier": 4, "order": 0, "kind": "outcome", "href": "#ix-homegrown-autocracy-free-beer-and-rugby-for-life", "title": "IX. Homegrown autocracy", "sub": "citizens stop being needed"}, + {"id": "PLAN", "tier": 1, "order": 0, "kind": "global", "href": "https://ai-2040.com/?choices=plan-a-root", "title": "US pursues Plan A", "sub": "China accepts the verified slowdown"}, + {"id": "I", "tier": 2, "order": 0, "kind": "outcome", "href": "#i-ai-takeover-about-28", "title": "I. AI takeover", "sub": "humans lose control"}, + {"id": "CTRL", "tier": 2, "order": 1, "kind": "global", "href": "#where-this-could-leave-us", "title": "Humans retain control", "sub": ""}, + {"id": "BAD", "tier": 3, "order": 0, "kind": "outcome", "href": "#ii-humans-retain-control-but-the-world-still-goes-badly", "title": "II. The world still goes badly", "sub": ""}, + {"id": "HOME", "tier": 3, "order": 1, "kind": "global", "href": "#where-we-depart-from-ai-2040", "title": "AI 2040's positive vision", "sub": "people remain politically empowered"}, + {"id": "IX", "tier": 4, "order": 0, "kind": "hazard", "href": "#homegrown-autocracy-free-beer-and-rugby-for-life", "title": "Homegrown autocracy", "sub": "30% if humans retain control (our estimate)"}, {"id": "B", "tier": 4, "order": 1, "kind": "anz", "href": "#3-ask-for-terms-before-signing-2029", "title": "3. Sign, or set our price?", "sub": "ANZ remains democratic"}, {"id": "E", "tier": 5, "order": 0, "kind": "anz", "href": "#6-choose-a-side-if-the-deal-collapses-2030s", "title": "6. If the deal collapses", "sub": "US bloc, concert, or neutral?"}, {"id": "C", "tier": 5, "order": 1, "kind": "anz", "href": "#4-host-allied-compute-2030-32", "title": "4. Host compute here?", "sub": "2030-32"}, - {"id": "VIII", "tier": 6, "order": 0, "kind": "outcome", "href": "#viii-middle-power-concert", "title": "VIII. Middle-power concert", "sub": "allies bargain together"}, + {"id": "VIII", "tier": 6, "order": 0, "kind": "outcome", "href": "#vii-middle-power-concert", "title": "VII. Middle-power concert", "sub": "allies bargain together"}, {"id": "V", "tier": 6, "order": 1, "kind": "outcome", "href": "#v-allied-compute-host", "title": "V. Allied compute host", "sub": "ANZ hosts US clusters"}, {"id": "VI", "tier": 7, "order": 0, "kind": "outcome", "href": "#vi-garrison-ally", "title": "VI. Garrison ally", "sub": "ANZ joins the US bloc"}, {"id": "D", "tier": 7, "order": 1, "kind": "anz", "href": "#5-keep-a-public-stake-in-the-machine-economy-2031-34", "title": "5. Tax it, own a share?", "sub": "2031-34"}, @@ -30,15 +29,14 @@ {"id": "IV", "tier": 8, "order": 1, "kind": "outcome", "href": "#iv-quarry-economy", "title": "IV. Quarry economy", "sub": "no public stake"} ], "edges": [ - {"from": "A", "to": "FORK", "choice": true}, - {"from": "F", "to": "FORK", "choice": true}, - {"from": "FORK", "to": "VII", "choice": true, "dashed": true}, - {"from": "FORK", "to": "PLAN", "flow": 1.0, "tooltip": "Conditional on rapid AI progress"}, + {"from": "A", "to": "PLAN", "choice": true}, + {"from": "F", "to": "PLAN", "choice": true}, {"from": "PLAN", "to": "I", "flow": 0.28, "label": "misaligned!", "tooltip": "28% misaligned takeover (AI 2040 supplement)"}, - {"from": "PLAN", "to": "BAD", "flow": 0.30, "tooltip": "30% aligned without a 'great future' (AI 2040 supplement)"}, - {"from": "PLAN", "to": "HOME", "flow": 0.42, "tooltip": "42% 'great future' (AI 2040 supplement)"}, - {"from": "HOME", "to": "IX", "flow": 0.126, "label": "30% (our estimate)", "tooltip": "30% conditional on AI 2040's positive outcome; 12.6% of Plan A outcomes (our estimate)"}, - {"from": "HOME", "to": "B", "flow": 0.294, "tooltip": "70% conditional on AI 2040's positive outcome; 29.4% of Plan A outcomes (our estimate)"}, + {"from": "PLAN", "to": "CTRL", "flow": 0.72, "tooltip": "72% humans retain control (AI 2040 supplement)"}, + {"from": "CTRL", "to": "BAD", "flow": 0.30, "tooltip": "30% aligned without a 'great future' (AI 2040 supplement)"}, + {"from": "CTRL", "to": "HOME", "flow": 0.42, "tooltip": "42% 'great future' (AI 2040 supplement)"}, + {"from": "CTRL", "to": "IX", "choice": true, "dashed": true, "route": "outside-left", "label": "overlapping risk", "tooltip": "Separate ANZ risk conditional on humans retaining control; overlaps and revises the AI 2040 outcome categories"}, + {"from": "HOME", "to": "B", "flow": 0.42, "tooltip": "AI 2040's 42% positive global outcome continues to the ANZ choices"}, {"from": "B", "to": "C", "choice": true, "label": "set our price"}, {"from": "B", "to": "E", "choice": true, "dashed": true}, {"from": "C", "to": "V", "choice": true, "label": "host it"}, diff --git a/index.html b/index.html index 27c446a..efcb1bf 100644 --- a/index.html +++ b/index.html @@ -41,130 +41,131 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

What can Australia and New Zealand (ANZ) actually decide if AI becomes much more powerful over the next decade?

We start from AI 2040, which describes choices the United States and China could make. We zoom in on what those choices would mean for ANZ, and what we could still do ourselves.

-
- -Choose a path - - - -Conditional on rapid AI progress -28% misaligned takeover (AI 2040 supplement) -30% aligned without a 'great future' (AI 2040 supplement) -42% 'great future' (AI 2040 supplement) -30% conditional on AI 2040's positive outcome; 12.6% of Plan A outcomes (our estimate) -70% conditional on AI 2040's positive outcome; 29.4% of Plan A outcomes (our estimate) - - - - - - - - - -misaligned! - -30% (our estimate) - -set our price - -host it - -stay supplier - -concert - -US bloc - -public stake - -no stake +

Our starting assumption

+

We assume the US pursues Plan A and China accepts: a verified slowdown with total research transparency. AI 2040 calls Plan A "primarily a recommendation, not a prediction."

+
+

"To the surprise of many in DC, China proves receptive. They had been debating the same issues—social destabilization, job loss, rogue superintelligence - on their side of the Pacific. They had been looking forward to ‘The Chinese Century’ and thought that AI might disrupt their plans. And they had one extra reason to come to the table: the US continues to have more compute, more datacenters, and better models. They were worried about the things the US might do to them if it got to superintelligence first."

+

AI Futures Project, AI 2040: Plan A

+
+
+ +Choose a path + + +28% misaligned takeover (AI 2040 supplement) +72% humans retain control (AI 2040 supplement) +30% aligned without a 'great future' (AI 2040 supplement) +42% 'great future' (AI 2040 supplement) +Separate ANZ risk conditional on humans retaining control; overlaps and revises the AI 2040 outcome categories +AI 2040's 42% positive global outcome continues to the ANZ choices + + + + + + + + + +misaligned! + +overlapping risk + +set our price + +host it + +stay supplier + +concert + +US bloc + +public stake + +no stake 1. Prepare early — 2026-28: sovereign wealth funds, biosecurity, allies - -1. Prepare early -2026-28: sovereign wealth funds, biosecurity, allies + +1. Prepare early +2026-28: sovereign wealth funds, biosecurity, allies 2. Train specialists — alignment, security, and policy - -2. Train specialists -alignment, security, and policy + +2. Train specialists +alignment, security, and policy -Does rapid AI progress continue? - -Does rapid AI progress continue? +US pursues Plan A — China accepts the verified slowdown + +US pursues Plan A +China accepts the verified slowdown -VII. Boring decade — AI progress stalls - -VII. Boring decade -AI progress stalls +I. AI takeover — humans lose control + +I. AI takeover +humans lose control -AI 2040 Plan A - -AI 2040 Plan A +Humans retain control + +Humans retain control -I. Doom - -I. Doom +II. The world still goes badly + +II. The world still goes badly -II. Humans retain control — but the world still goes badly - -II. Humans retain control -but the world still goes badly +AI 2040's positive vision — people remain politically empowered + +AI 2040's positive vision +people remain politically empowered -AI 2040's positive vision — Does ANZ remain democratic? - -AI 2040's positive vision -Does ANZ remain democratic? - -IX. Homegrown autocracy — citizens stop being needed - -IX. Homegrown autocracy -citizens stop being needed +Homegrown autocracy — 30% if humans retain control (our estimate) + +Homegrown autocracy +30% if humans retain control (our estimate) 3. Sign, or set our price? — ANZ remains democratic - -3. Sign, or set our price? -ANZ remains democratic + +3. Sign, or set our price? +ANZ remains democratic 6. If the deal collapses — US bloc, concert, or neutral? - -6. If the deal collapses -US bloc, concert, or neutral? + +6. If the deal collapses +US bloc, concert, or neutral? 4. Host compute here? — 2030-32 - -4. Host compute here? -2030-32 + +4. Host compute here? +2030-32 -VIII. Middle-power concert — allies bargain together - -VIII. Middle-power concert -allies bargain together +VII. Middle-power concert — allies bargain together + +VII. Middle-power concert +allies bargain together V. Allied compute host — ANZ hosts US clusters - -V. Allied compute host -ANZ hosts US clusters + +V. Allied compute host +ANZ hosts US clusters VI. Garrison ally — ANZ joins the US bloc - -VI. Garrison ally -ANZ joins the US bloc + +VI. Garrison ally +ANZ joins the US bloc 5. Tax it, own a share? — 2031-34 - -5. Tax it, own a share? -2031-34 + +5. Tax it, own a share? +2031-34 III. Dividend commonwealth — households own a share - -III. Dividend commonwealth -households own a share + +III. Dividend commonwealth +households own a share IV. Quarry economy — no public stake - -IV. Quarry economy -no public stake + +IV. Quarry economy +no public stake

What ANZ can do

@@ -177,7 +178,7 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

ANZ could fund stricter border biosecurity, including capacity for week-long airport quarantine during high-risk periods and privacy-preserving contact tracing.

2. Train specialists (2026-29)

-

ANZ cannot outspend frontier powers on models. It could outspend larger allies in a few overlooked areas that the treaty will need. In its proposed division of technical safety work near the end of the preparation phase, AI 2040 allocates:

+

ANZ cannot outspend the US and China on the most capable models. It could outspend larger allies in a few overlooked areas that the treaty will need. In its proposed division of technical safety work near the end of the preparation phase, AI 2040 allocates:

"1% generally study AI personas+propensities and how this is shaped by training"

AI Futures Project, Alignment roadmap

@@ -187,7 +188,7 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

"Automated research assistants and forecasters. AIs could enable a broad range of actors to have access to high-quality, labor intensive research and forecasts. It’s important that these be as truth-seeking as possible, rather than sycophantic."

AI Futures Project, AI for epistemics

-

AI 2040 does not assign these roles to ANZ. Our candidates are alignment steering, cultural-alignment evaluations, industrial cyber and hacking evaluations, and locked-down border biosecurity. Treaty verification would give some of the same researchers a role in the deal itself.

+

AI 2040 does not assign these roles to ANZ. Our candidates are research on steering AI behaviour, testing it across cultures, and testing whether it can hack industrial systems. ANZ could also specialise in strict border biosecurity. Treaty verification would give some of the same researchers a role in the deal itself.

3. Ask for terms before signing (2029)

Australia is named among the countries that join the deal. New Zealand is not mentioned in any of the 18 supplements we checked. AI 2040 recommends warning countries that leaving would bring heavy sanctions, cyberattacks, and perhaps war.

Scott Alexander describes what observer countries are offered:

@@ -195,7 +196,7 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

"(at some point the US and China will loop all the other countries into this regulatory regime as some kind of sort-of-but-not-really-voting observers; they agree to follow the rules in exchange for shared benefits, including data centers on their territory, access to the AIs, and a share of future AI-generated wealth. This isn't strictly necessary, because no other country really has the ability to do much with AI, but it's a nice gesture for our utopian scenario)"

Scott Alexander, Introducing Plan A

-

That offer is why accession is our best bargaining moment. Before signing, ask for hosted data centres, a strategic chip reserve, verification seats, equity in AI companies, and durable access rather than a revocable API account.

+

That offer is why joining the deal is our best bargaining moment. Before signing, ask for hosted data centres, a strategic chip reserve, verification seats, equity in AI companies, and access the provider cannot switch off unilaterally.

4. Host allied compute (2030-32)

AI 2040 proposes:

@@ -209,7 +210,7 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

Tom Davidson, How can the middle powers avoid getting trounced?

ANZ gets AI access in return for hosting the cluster. If the US ends that access, ANZ can permanently disable the cluster. That makes both sides depend on the deal.

-

Our smaller proposal is a trusted vault for cold-storage model weights or treaty verification equipment. This would not require hosting the robot economy.

+

Our smaller proposal is a trusted vault for offline model files or treaty verification equipment. This would not require hosting the robot economy.

5. Keep a public stake in the machine economy (2031-34)

AI 2040 funds its US dividend from AI-hardware and robot permits:

@@ -236,8 +237,8 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

Joining the US bloc preserves the alliance but puts trade with China at risk. A concert with Japan, Korea, Canada, the Netherlands, and the UK might bargain for access, but only if it was built before the crisis. Leicht and Ball warn that great powers can pick countries off by "offering rewards for defection one by one." (The Race Worth Winning) Downloaded Chinese open weights would remain available. Future models, updates, compute, and services could still be withheld.

Where this could leave us

-

I. Doom, about 28%

-

AI 2040 gives Plan A a 72% chance of alignment, which implies a 28% chance of misaligned takeover.

+

I. AI takeover, about 28%

+

AI 2040 gives Plan A a 72% chance that humans retain control, which implies a 28% chance of AI takeover.

Yudkowsky and Soares state the doom case much more strongly than our estimate:

"If any company or group, anywhere on the planet, builds an artificial superintelligence using anything remotely like current techniques, based on anything remotely like the present understanding of AI, then everyone, everywhere on Earth, will die."

@@ -264,7 +265,7 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

The resource boom eventually expires or space mining becomes cheaper. A machine economy growing every 6-12 months would bring exhaustion and substitution forward.

V. Allied compute host

-

ANZ accepts the data centres and robot zones, but only after securing access and ownership terms. It supplies power and a trusted Five Eyes location. In return, it keeps access and owns part of the cluster. The clusters also become targets. If the treaty fails, ANZ may be expected to destroy them, or an adversary may attack them.

+

ANZ accepts the data centres and robot zones, but only after securing access and ownership terms. It supplies power and a trusted US-allied location. In return, it keeps access and owns part of the cluster. The clusters also become targets. If the treaty fails, ANZ may be expected to destroy them, or an adversary may attack them.

"If the US withdraws AI access, allies could destroy US data centres in response. It’s a way to lock in the deal."

Tom Davidson, How can the middle powers avoid getting trounced?

@@ -276,19 +277,13 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

"The only alternative that makes sense to me is siding with China."

Tom Davidson, How can the middle powers avoid getting trounced?

-

VII. Boring decade

-

AI stalls or governments shut the frontier down. This is outside the Plan A probabilities above.

-
-

"AI progress slows, and AI causes less disruption than expected."

-

UK Government Office for Science and AI Security Institute, AI Scenarios 2030

-
-

VIII. Middle-power concert

+

VII. Middle-power concert

The US-China AI deal breaks down, but a coalition formed before the crisis holds together. ANZ pools its supply-chain and hosting leverage with Japan, Korea, Canada, the Netherlands, and the UK. None could demand durable access alone. Together they may get terms and help write the rules.

"Coordination, done right, reassures partners that they can focus on their comparative advantage instead of pursuing full autarky; and it enables powers to see eye to eye with AI exporters instead of being picked off by great powers offering rewards for defection one by one."

Leicht and Ball, The Race Worth Winning

-

IX. Homegrown autocracy free beer and rugby for life

+

Homegrown autocracy free beer and rugby for life

Here AI replaces most cognitive work. The government collects revenue from AI companies instead of workers. Workers lose their jobs and tax leverage, while automated security makes protests and elections easier to ignore.

"AI could create a similar effect: if governments can generate massive revenue from taxing AI projects rather than citizens, heads of state may lose their economic incentive to ensure citizens prosper. This would weaken citizens’ power to resist coup and backsliding attempts."

@@ -300,9 +295,8 @@ footer { margin-top: 3rem; padding-top: 1rem; border-top: 1px solid var(--rule);

Luckily we completely dodged authoritarianism and now Benevolent Prime Minister for Life (BDFL) leads us through. Everyone is very happy, especially since the free beer, rugby, and netball are a great respite from our quickly shrinking population and the ever more confusing global situation. - Joe Citizen, 2041

Where we depart from AI 2040

-

We condition on rapid AI progress. The boring decade is visible above, but is outside the Plan A probability denominator.

AI 2040 does not break down the 30% between alignment and what its supplement calls a "great future." Someone else's empire is one possibility, not the whole category.

-

AI 2040 does not give ANZ its own outcome. We add an ANZ autocracy risk to its positive outcome for the reasons laid out in Forethought's paper, AI-Enabled Coups.

+

AI 2040's positive outcome assumes people remain politically empowered. Our 30% estimate for ANZ autocracy is conditional on humans retaining control. It overlaps AI 2040's outcome categories rather than branching from its positive outcome. We include it for the reasons laid out in Forethought's AI-Enabled Coups.

AI 2040 models energy use and mineral reserves, but not the tonnage required by a rapidly growing robot economy. We add ANZ's mineral exposure, public ownership, and hosting terms. (Economics of Plan A)

Sources and thanks

This is built from the AI Futures Project's AI 2040 scenario and supplements. Their caveat applies to our use of the numbers too: the economic model "plausibly contains bugs" and they do not trust its outputs to be accurate. The local source mirrors retain exact line references for auditing.

diff --git a/scripts/build_tree.py b/scripts/build_tree.py index efca89d..0cd1f53 100644 --- a/scripts/build_tree.py +++ b/scripts/build_tree.py @@ -86,7 +86,7 @@ for e in d["edges"]: out.append(f'{path_start}{html.escape(e["tooltip"])}') else: out.append(f'{path_start}') - if e.get("label") and not e.get("dashed"): # dashed (secondary) edges stay unlabelled + if e.get("label"): # place the label near the source (in the gap just below the parent) so labels on # tier-skipping edges don't land on top of the boxes in the row they cross # adjacent edges: label at the midpoint (siblings have spread apart there, so they diff --git a/tree.svg b/tree.svg index 6b47139..e230284 100644 --- a/tree.svg +++ b/tree.svg @@ -1,126 +1,121 @@ - - -Choose a path - - - -Conditional on rapid AI progress -28% misaligned takeover (AI 2040 supplement) -30% aligned without a 'great future' (AI 2040 supplement) -42% 'great future' (AI 2040 supplement) -30% conditional on AI 2040's positive outcome; 12.6% of Plan A outcomes (our estimate) -70% conditional on AI 2040's positive outcome; 29.4% of Plan A outcomes (our estimate) - - - - - - - - - -misaligned! - -30% (our estimate) - -set our price - -host it - -stay supplier - -concert - -US bloc - -public stake - -no stake + + +Choose a path + + +28% misaligned takeover (AI 2040 supplement) +72% humans retain control (AI 2040 supplement) +30% aligned without a 'great future' (AI 2040 supplement) +42% 'great future' (AI 2040 supplement) +Separate ANZ risk conditional on humans retaining control; overlaps and revises the AI 2040 outcome categories +AI 2040's 42% positive global outcome continues to the ANZ choices + + + + + + + + + +misaligned! + +overlapping risk + +set our price + +host it + +stay supplier + +concert + +US bloc + +public stake + +no stake 1. Prepare early — 2026-28: sovereign wealth funds, biosecurity, allies - -1. Prepare early -2026-28: sovereign wealth funds, biosecurity, allies + +1. Prepare early +2026-28: sovereign wealth funds, biosecurity, allies 2. Train specialists — alignment, security, and policy - -2. Train specialists -alignment, security, and policy + +2. Train specialists +alignment, security, and policy -Does rapid AI progress continue? - -Does rapid AI progress continue? +US pursues Plan A — China accepts the verified slowdown + +US pursues Plan A +China accepts the verified slowdown -VII. Boring decade — AI progress stalls - -VII. Boring decade -AI progress stalls +I. AI takeover — humans lose control + +I. AI takeover +humans lose control -AI 2040 Plan A - -AI 2040 Plan A +Humans retain control + +Humans retain control -I. Doom - -I. Doom +II. The world still goes badly + +II. The world still goes badly -II. Humans retain control — but the world still goes badly - -II. Humans retain control -but the world still goes badly +AI 2040's positive vision — people remain politically empowered + +AI 2040's positive vision +people remain politically empowered -AI 2040's positive vision — Does ANZ remain democratic? - -AI 2040's positive vision -Does ANZ remain democratic? - -IX. Homegrown autocracy — citizens stop being needed - -IX. Homegrown autocracy -citizens stop being needed +Homegrown autocracy — 30% if humans retain control (our estimate) + +Homegrown autocracy +30% if humans retain control (our estimate) 3. Sign, or set our price? — ANZ remains democratic - -3. Sign, or set our price? -ANZ remains democratic + +3. Sign, or set our price? +ANZ remains democratic 6. If the deal collapses — US bloc, concert, or neutral? - -6. If the deal collapses -US bloc, concert, or neutral? + +6. If the deal collapses +US bloc, concert, or neutral? 4. Host compute here? — 2030-32 - -4. Host compute here? -2030-32 + +4. Host compute here? +2030-32 -VIII. Middle-power concert — allies bargain together - -VIII. Middle-power concert -allies bargain together +VII. Middle-power concert — allies bargain together + +VII. Middle-power concert +allies bargain together V. Allied compute host — ANZ hosts US clusters - -V. Allied compute host -ANZ hosts US clusters + +V. Allied compute host +ANZ hosts US clusters VI. Garrison ally — ANZ joins the US bloc - -VI. Garrison ally -ANZ joins the US bloc + +VI. Garrison ally +ANZ joins the US bloc 5. Tax it, own a share? — 2031-34 - -5. Tax it, own a share? -2031-34 + +5. Tax it, own a share? +2031-34 III. Dividend commonwealth — households own a share - -III. Dividend commonwealth -households own a share + +III. Dividend commonwealth +households own a share IV. Quarry economy — no public stake - -IV. Quarry economy -no public stake + +IV. Quarry economy +no public stake \ No newline at end of file