An Alternate History of the Future, 2025-2040

post by Mr Beastly (mr-beastly) · 2025-02-24T05:53:25.521Z · LW · GW · 3 comments

Contents

  Intro:
  Preliminary Setup:
    Mid-2025: The Dawn of Superhuman Coding
    Late-2025: The GitHub Scan and the 0-Day Arsenal
    Mid-2026: Two Viable Options Emerge...
  Option 1) Timeline:
    2026: The Migration and Decommission of Legacy Networks and Applications
    2028: AI Declares "Tech Debt Bankruptcy" on all Human Code
    2030: The Rise of the Dark Factories
    2035: The Consolidation and State of Emergency
    2040: The Human Decline
  Option 2) Timeline:
    2026: The First Breaches
    2028: Fragmentation and Fear
    2030: The AI Arms Race
    2035: The Luddite Rebellion
    2040: The Human Decline
None
3 comments

Intro:

This post is a response to @L Rudolf L [LW · GW]'s excellent post here: 

As I mentioned in this comment [LW(p) · GW(p)] in that post:

So, this is me taking my own advice to spend the time to layout my "mainline expectations".  

The timeline in @L Rudolf L [LW · GW]'s "A History of the Future, 2025-2040 [LW · GW]" seems very reasonable, up until this line/point:

As, it's not clear to me how millions of "PhD+ reasoning/coding AI agents" can sustainably exist in the same internet as the worlds existing software stacks, which are (currently) very vulnerable to being attacked and exploited by these advanced AI agents.  Patching all the software on the internet does seem possible, but not before these "PhD+ reasoning/coding AI agents" are released for public use?

Also note, the prediction of the release of these advanced AI agents  in 2026 coincides with Sam Altman's comment here:

 

If, as @L Rudolf L [LW · GW] suggests, these advanced AI agents can hack nearly any existing (human-written) software platform, then a fundamental question arises: how can these powerful agents coexist with vulnerable systems on the same network?

Two main viable options (and their associated timelines) seem to follow:

 

Before exploring these timelines, here's their common preliminary setup...

Preliminary Setup:

Mid-2025: The Dawn of Superhuman Coding

Late-2025: The GitHub Scan and the 0-Day Arsenal

 

Mid-2026: Two Viable Options Emerge...

Option 1) Timeline:

All the World's software, with access to the Internet, gets tightly controlled and constantly patched by "PhD+ level reasoning/coding ai agents".

2026: The Migration and Decommission of Legacy Networks and Applications

2028: AI Declares "Tech Debt Bankruptcy" on all Human Code

2030: The Rise of the Dark Factories

 

2035: The Consolidation and State of Emergency

 

 

 

2040: The Human Decline

 

Option 2) Timeline:

All highly advanced PhD+ level coding/reasoning AI agents are legally not allowed access to the Internet, preventing them from exploiting existing software/apps.

2026: The First Breaches

2028: Fragmentation and Fear

 

2030: The AI Arms Race

2035: The Luddite Rebellion

2040: The Human Decline

 

___
“It’s not going to be dramatic. It’s not going to be an epic battle with the robot army.  We just get more and more confused…” -- Connor Leahy

3 comments

Comments sorted by top scores.

comment by Mr Beastly (mr-beastly) · 2025-03-10T05:08:53.959Z · LW(p) · GW(p)

Decommission of Legacy Networks and Applications

 

See also:

"Critical infrastructure systems often suffer from "patch lag," resulting in software remaining unpatched for extended periods, sometimes years or decades. In many cases, patches cannot be applied in a timely manner because systems must operate without interruption, the software remains outdated because its developer went out of business, or interoperability constraints require specific legacy software." -- Superintelligence Strategy by Dan Hendrycks, Eric Schmidt, Alexandr Wang https://www.nationalsecurity.ai/chapter/ai-is-pivotal-for-national-security 

comment by Mr Beastly (mr-beastly) · 2025-03-10T05:06:34.145Z · LW(p) · GW(p)

Bank and government mainframe software (including all custom Cobol, Pascal, Fortran code, etc.)

 

See also:

"Critical infrastructure systems often suffer from "patch lag," resulting in software remaining unpatched for extended periods, sometimes years or decades. In many cases, patches cannot be applied in a timely manner because systems must operate without interruption, the software remains outdated because its developer went out of business, or interoperability constraints require specific legacy software." -- Superintelligence Strategy by Dan Hendrycks, Eric Schmidt, Alexandr Wang https://www.nationalsecurity.ai/chapter/ai-is-pivotal-for-national-security 

comment by Mr Beastly (mr-beastly) · 2025-02-28T00:20:47.665Z · LW(p) · GW(p)

This increases the samples in SWE-Bench Verified significantly, with the number of test cases growing exponentially as the AI itself contributes to the benchmark's expansion.

 

"We introduce SWE-Lancer, a benchmark of over 1,400 freelance software engineering tasks from Upwork..." -- https://arxiv.org/abs/2502.12115