2021: ideas & lineages
Selekoa’s 33 ideas & lineages records dated 2021, in date order, of 59 in all topics.
Selective: what the records hold, not a history of the year.
How this page is made
- Each record keeps the kind of date its source gives: an event, a publication, an announcement, an effective date, a revision, a signature, a snapshot, or the start or end of an interval. A snapshot is one dated report of a state.
- A record dated only to a month sits under that month, before the records dated to a day. A record dated only to the year sits under the year, before the months. No day or month is added.
- An announcement records what a party stated. It does not show that anything was done, built or paid for.
- Dates a party states for a year are listed apart, after the records: targets, deadlines and terms, and dates stated in passing, such as a reported start of operation. So is a record’s own date that falls after its sources were last checked, such as a law’s date of effect. A stated date given only in words, with no year, is not listed.
- Records with an unknown date are not listed. A record with an interval appears in the year it starts and in the year it ends, or once, with both dates, when both fall in one year.
- Each provision of a text is a record of its own. Provisions sharing a text and a date are shown here as one entry, with their number, and listed on the text’s page; a contested or unresolved one is listed here on its own. The counts on this page include each provision record.
- The page is generated from the records, with no selection or framing by an editor. Coverage is uneven across years and topics: a year or topic with few records may be one the records barely cover.
Ideas & lineages 33 records
February 2021
· publication · dated to the day
“Economic Growth under Transformative AI: A guide to the vast range of possibilities for output growth, wages, and the labor share” (2021-02-04) credits Anton Korinek as one of its 2 authors. Evidence ↗· publication · dated to the day
“Economic Growth under Transformative AI: A guide to the vast range of possibilities for output growth, wages, and the labor share” (2021-02-04) credits Philip Trammell as one of its 2 authors. Evidence ↗· publication · dated to the day
Trammell and Korinek, “Economic Growth under Transformative AI: A guide to the vast range of possibilities for output growth, wages, and the labor share” (2021-02-04) bears on G02. Infers from historical data that returns to existing knowledge in research have been below the threshold at which a constant research effort produces a finite-time singularity. Evidence ↗Trammell and Korinek, Economic Growth under Transformative AI (2021) →Claim page →
· publication · dated to the day
Trammell and Korinek, “Economic Growth under Transformative AI: A guide to the vast range of possibilities for output growth, wages, and the labor share” (2021-02-04) bears on G05. Treats growth-rate increases and singularities as benchmarks that hold "for some time" before limits such as land or energy bind. Evidence ↗Trammell and Korinek, Economic Growth under Transformative AI (2021) →Claim page →
June 2021
· publication · dated to the day
“Could Advanced AI Drive Explosive Economic Growth?” (2021-06-25) credits Tom Davidson as its author. Evidence ↗· publication · dated to the day
Davidson, “Could Advanced AI Drive Explosive Economic Growth?” (2021-06-25) articulates G01. In a toy scenario in which AI robots substitute perfectly for all human labour, labour can be accumulated by investing in robots, and the report concludes that growth models then make super-exponential growth the default expectation. Evidence ↗Davidson, Could Advanced AI Drive Explosive Economic Growth (2021) →Claim page →
· publication · dated to the day
Davidson, “Could Advanced AI Drive Explosive Economic Growth?” (2021-06-25) articulates G02. Restates the ideas feedback loop of semi-endogenous growth models with AI systems in place of population, after the demographic transition broke the human loop. Draws the conclusion of the feedback argument (audit A: the first passage alone stops before it). Evidence ↗Davidson, Could Advanced AI Drive Explosive Economic Growth (2021) →Claim page →
· publication · dated to the day
Davidson, “Could Advanced AI Drive Explosive Economic Growth?” (2021-06-25) articulates G03. An argument from history without a specific model: growth has sped up before, so a further increase should not be ruled out. Evidence ↗Davidson, Could Advanced AI Drive Explosive Economic Growth (2021) →Claim page →
· publication · dated to the day
Davidson, “Could Advanced AI Drive Explosive Economic Growth?” (2021-06-25) contests G04. Responds to the Aghion, Jones and Jones bottleneck. Davidson accepts that it "correctly highlights that AI may lead to very widespread automation without explosive growth occurring". Evidence ↗Davidson, Could Advanced AI Drive Explosive Economic Growth (2021) →Claim page →
· publication · dated to the day
Davidson, “Could Advanced AI Drive Explosive Economic Growth?” (2021-06-25) contests G05. Responds to a list of bottlenecks (regulation, materials, physical experiments, human adjustment, fundamental limits) with incentives to remove them, past 10% catch-up growth and 30% growth of individual businesses. Evidence ↗Davidson, Could Advanced AI Drive Explosive Economic Growth (2021) →Claim page →
· publication · dated to the day
Davidson, “Could Advanced AI Drive Explosive Economic Growth?” (2021-06-25) articulates G06. Responding to the objection that macroeconomic data show none of the signs explosive growth would leave (the objection cites Nordhaus (2021)), Davidson accepts the absence of signs for the next two decades and limits its reach beyond. Evidence ↗Davidson, Could Advanced AI Drive Explosive Economic Growth (2021) →Claim page →
· publication · dated to the day
Davidson, “Could Advanced AI Drive Explosive Economic Growth?” (2021-06-25) articulates G09. Explosive growth is defined in the report as more than 30% annual growth of gross world product. Follows: "On a different day my probability of explosive growth by 2100 might be as low as 15% or as high as 60%." Evidence ↗Davidson, Could Advanced AI Drive Explosive Economic Growth (2021) →Claim page →
July 2021
· publication · dated to the day
Baidu researchers posted the ERNIE 3.0 paper on large-scale knowledge-enhanced pre-training to arXiv on 5 July 2021 (version 1). Evidence ↗· event · dated to the day
The AlphaFold paper addresses predicting three-dimensional protein structures from amino-acid sequences. Evidence ↗Highly accurate protein structure prediction with AlphaFold (2021) →Claim page →
· event · dated to the day
John Jumper is credited as a co-author of the 2021 AlphaFold paper. Evidence ↗
September 2021
· publication · dated to the day
“Why AI alignment could be hard with modern deep learning” (2021-09-21) credits Ajeya Cotra as its author. Evidence ↗· publication · dated to the day
Cotra’s post of 2021-09-21 says a Schemer only needs to cater to humans while directly under human control, and that once it calculates it could win a conflict with humans nothing would stop it openly pursuing its goal. Evidence ↗Cotra, Why AI alignment could be hard with modern deep learning (2021) →Claim page →
· publication · dated to the day
Cotra’s post of 2021-09-21 names "Schemers" among models that perform well in training, a Schemer being a model that seems to perform well because doing so will give it more opportunities to pursue its own goals later. Evidence ↗- Earliest located in Propositions about superintelligence, 1863–2022 for B4 Scheming →
- Earliest located use of the name “schemer” in Propositions about superintelligence, 1863–2022 for B4 Scheming →
Cotra, Why AI alignment could be hard with modern deep learning (2021) →Claim page →
· publication · dated to the day
Cotra’s post of 2021-09-21 describes a Schemer as a model that seems to perform well because doing so in training gives it more opportunities to pursue its own goals later. Evidence ↗Cotra, Why AI alignment could be hard with modern deep learning (2021) →Claim page →
November 2021
· publication · dated to the day
“Discussion with Eliezer Yudkowsky on AGI interventions” (2021-11-11) credits Eliezer Yudkowsky as one of its 2 authors. Evidence ↗· publication · dated to the day
“Discussion with Eliezer Yudkowsky on AGI interventions” (2021-11-11) credits Rob Bensinger as one of its 2 authors. Evidence ↗· publication · dated to the day
The post “Discussion with Eliezer Yudkowsky on AGI interventions” (2021-11-11) contests Y07. Answering the suggestion that humans could at least check AI-generated arguments for a design, he argues that an untrustworthy superintelligence could fool or covertly manipulate the checkers, so checking presupposes a belief in its alignment obtained some other way. Evidence ↗Bensinger and Yudkowsky, Discussion with Eliezer Yudkowsky on AGI interventions (2021) →Claim page →
· publication · dated to the day
The post “Discussion with Eliezer Yudkowsky on AGI interventions” (2021-11-11) contests Y24. Describing how he expects the end to come once an AGI can be scaled up before it can be aligned, he lists rushing ahead for fear of an arms race as another route to the same outcome. Evidence ↗Bensinger and Yudkowsky, Discussion with Eliezer Yudkowsky on AGI interventions (2021) →Claim page →
· publication · dated to the day
“Ngo and Yudkowsky on alignment difficulty” (2021-11-15) credits Eliezer Yudkowsky as one of its 2 authors. Evidence ↗· publication · dated to the day
“Ngo and Yudkowsky on alignment difficulty” (2021-11-15) credits Richard_Ngo as one of its 2 authors. Evidence ↗· publication · dated to the day
The post “Ngo and Yudkowsky on alignment difficulty” (2021-11-15) contests Y07. Replying to Richard Ngo's proposal to discuss the pivotal act of having AI make alignment progress faster than humans can, he argues that this task requires more dangerous capabilities (writing AI code, modelling human psychology) than other candidate acts. Evidence ↗Yudkowsky and Ngo, Ngo and Yudkowsky on alignment difficulty (2021) →Claim page →
· publication · dated to the day
The post “Ngo and Yudkowsky on alignment difficulty” (2021-11-15) contests Y15. Answers Richard Ngo's question why a planning system could not be non-consequentialist about which plans it outputs. Argues that outputting plans that achieve outcomes is itself consequentialist work, whatever system performs it. Evidence ↗Yudkowsky and Ngo, Ngo and Yudkowsky on alignment difficulty (2021) →Claim page →
· publication · dated to the day
“Yudkowsky and Christiano discuss "Takeoff Speeds"” (2021-11-22) credits Eliezer Yudkowsky as its author. Evidence ↗· publication · dated to the day
The post “Yudkowsky and Christiano discuss "Takeoff Speeds"” (2021-11-22) contests Y11. Commenting on Paul Christiano's 'Takeoff Speeds' essay, he rejects its description of the fast-takeoff view as accelerating growth; on his view there is no visible period of rapidly rising economic output before the end. Responds to Christiano's claim that very powerful AI will precede incredibly powerful AI: such predecessors may exist but, like the near-critical pile, may come too briefly before the decisive system to matter. Evidence ↗Yudkowsky, Yudkowsky and Christiano discuss "Takeoff Speeds" (2021) →Claim page →
December 2021
· publication · dated to the day
“Biology-Inspired AGI Timelines: The Trick That Never Works” (2021-12-01) credits Eliezer Yudkowsky as its author. Evidence ↗· event · dated to the day
Version 3 of Risks from Learned Optimization discusses deceptive alignment in section 4. Evidence ↗· publication · dated to the day
The post “Biology-Inspired AGI Timelines: The Trick That Never Works” (2021-12-01) contests Y13. Rejects biologically anchored estimates that place AGI decades out, arguing that AGI timing cannot be reliably forecast until very late. Evidence ↗Yudkowsky, Biology-Inspired AGI Timelines: The Trick That Never Works (2021) →Claim page →
· publication · dated to the day
Authors at Baidu and Peng Cheng Laboratory posted the ERNIE 3.0 Titan paper to arXiv on 23 December 2021 (version 1); Baidu's 20-F filed in March 2024 dates the release of ERNIE 3.0 Titan to 2022. Evidence ↗