Adapting to AI — goal tree and current reality (first pass)

GOAL  The world adapts well to AI, as it already is and as it could be developed
├─ CSF  Frontier general AI is under control, answerable worldwide, and no longer
│       confined to the US (and perhaps China)
│  ├─ NC  Training of the most general frontier models slows; narrow AI continues
│  │  ├─ NC  Independent international bodies (e.g. UN AI safety office) verify
│  │  │      and monitor frontier training                     ↗ chips agreements
│  │  └─ NC  Developers and states agree enforceable training limits     (AI)
│  ├─ NC  AI companies are held to basic account, e.g. copyright
│  │  └─ NC  Non-US jurisdictions have laws that bind AI companies       (AI)
│  ├─ NC  Countries outside the US have a real say in AI rules
│  │  ├─ NC  Non-US countries coordinate their limited leverage
│  │  └─ NC  A forum exists where their decisions shape AI rules         (AI)
│  └─ NC  Countries outside the US can do frontier development themselves
│     ├─ NC  They get advanced chips under verifiable, monitored agreements
│     └─ NC  They have the talent, energy and data-centre capacity       (AI)
├─ CSF  Institutions and societies, including their cultures, stay functional
│       under the impacts of AI already released and still to come
│  ├─ NC  Governments and states keep governing despite AI disruption
│  ├─ NC  Banks, insurance and the wider financial system withstand AI shocks
│  ├─ NC  Critical systems defend against AI-enabled cyberattacks        (AI)
│  └─ NC  Cultures and public discourse stay healthy under the impacts of AI
├─ CSF  People keep their livelihoods and wellbeing and have real power
│       to shape and use AI                                            (AI)
└─ CSF? People can see and work towards hopeful futures with AI          (open)

Above the goal (noted, not in the tree): adapting well to the metacrisis
(AI, ecological breakdown, ...)

CURRENT REALITY · proposed · no links drawn yet

• AI development is out of control                                   (Leo)
• The US is an untrustworthy hegemon                                 (Leo)
• The US holds most of the most advanced AI chips, a growing
  monopoly that gives it hard power                                  (Leo)
• Countries with reasonable AI policies, such as the EU, struggle
  to govern AI because they lack hard power                          (Leo)
• The EU's relationship with the US is destabilising                 (Leo)
• When one jurisdiction restricts frontier AI, developers move to
  ones with no restrictions (race to the bottom)                     (Leo)
• Prediction: current AI capabilities alone could collapse banks
  and insurers and cripple states                                    (Leo)
• Those capabilities are already released as open-weight models      (Leo)
• A wave of AI-enabled cyberattacks is hitting now                   (Leo)
• AI-driven advertising and persuasion may degrade culture and
  public discourse                                                 (Rufus)

Candidate causes to test (AI):
  chip monopoly ──> EU and others lack hard power
  race to the bottom ──> labs race to release faster
  open weights ──> impact of current capabilities can't be contained

CURRENT REALITY · DRAFT v2 · AI-drafted, red-teamed once, NOT reviewed

── A · CONTROL ─────────────────────────────────────────────────────────────
U1   No actor can pause or bound frontier training runs
     (Leo: "AI development is totally out of control")                  [L]
 ◄─ AND ┬ C2   Labs and states race to reach the frontier first
        │  ◄─ AND ┬ C1   Frontier capability promises decisive economic
        │         │      and military advantage
        │         └ C15  US–China strategic rivalry
        └ C3   No binding international limits or verification exist
               for frontier training
           ◄─ ↑C2 · ↑C15 (each on its own)
           ◄─ C19  Verifying training limits is technically hard

C4   Developers threaten to withhold products or relocate when regulated
     (reworded from Leo's "race to the bottom")                         [L]
 ◄─ ↑C3   (with no common rules, the threat is credible)

── B · WHO DECIDES ─────────────────────────────────────────────────────────
U2   Frontier AI decisions sit with a few US and Chinese firms and
     governments; most people affected have no say
 ◄─ AND ┬ C5   The US holds most of the most advanced chips; the supply
        │      chain is concentrated and steered by US export controls  [L]
        └ C6   Few frontier labs exist outside the US and China
 ◄─ C16  Returns to AI flow to the owners of compute and models

U3   Countries with reasonable AI policies, e.g. the EU, can't govern
     AI effectively                                                     [L]
 ◄─ AND ┬ ↑C5
        └ C9   The EU–US relationship is destabilising, and the EU
               depends on the US for security                           [L]
           ◄─ C8   The US is an untrustworthy partner: policy swings,
                   coercion                                             [L]
 ◄─ ↑C4   (threats to withhold or leave weaken enforcement)
 ◄─ C20  The EU faces competitiveness pressure and industry lobbying

U7   Accountability for AI harms and copyright is slow, partial and
     after the fact (some large settlements exist)
 ◄─ AND ┬ ↑C4
        └ C11  Institutions were built for pre-AI threat levels and
               adapt slowly

── C · RESILIENCE ──────────────────────────────────────────────────────────
U5   AI-enabled cyberattacks are rising now                             [L]
 ◄─ AND ┬ C10  Capable models are released as open weights and can't
        │      be recalled                                              [L]
        │  ◄─ C13  The race pushes faster releases with less safety
        │          testing      ◄─ ↑C2
        │  ◄─ C17  Open release is used as competitive strategy
        │          (e.g. Meta, Chinese labs)   ◄─ ↑C15
        └ ↑C11
 ◄─ AND ┬ C18  Closed models are also misused for cyber campaigns
        └ ↑C11

U6   Public discourse is degraded by AI-generated persuasion and
     content (needs evidence)                                           [R]
 ◄─ AND ┬ C12  Ad-funded platforms reward engagement and persuasion
        └ C14  Persuasive models are widely available   ◄─ ↑C10

── D · LIVELIHOODS AND HOPE (new: the goal tree had no problems here) ──────
U8   AI is starting to displace work and income, with no cushion
 ◄─ AND ┬ ↑C16
        └ ↑C11  (safety nets not adapted)
U9   AI gains are concentrating in a few firms
 ◄─ ↑C16    ◄─ ↑C5 (compute concentration feeds it)
U10  Many people feel fatalistic about where AI is heading
 ◄─ AND ┬ ↑U1
        └ ↑U2   (no visible lever for ordinary people)

── PREDICTION (not yet a present problem) ──────────────────────────────────
P1   Already-released capabilities could collapse banks and insurers
     and cripple states                                                 [L]
 ◄─ AND ┬ ↑C10
        └ ↑C11
 ◄─ ↑U5

CANDIDATE CORE DRIVERS (hypotheses, not established):
  C1 + C15  the advantage race and US–China rivalry → branches A, C
  C16       returns flowing to owners of compute and models → B, D
  C5        chip concentration → B
  C11       slow-adapting institutions → C, D, U7