Rodney Brooks grades his own AI predictions every year until 2050
On January 1, 2018, Rodney Brooks published dated AI predictions he grades in an annual scorecard, promising to review them every year until 2050.
Facts
Atomic, verifiable facts - every one tied to a primary source.
268 entries, all primary-sourced
Scale AI says it has paid $1 billion to data contributors
Scale AI states it has paid over $1 billion to its global contributors and processed roughly 15 billion human decisions to train AI models.
Shannon credited J. W. Tukey for the word bit
In his 1948 paper, Claude Shannon noted that the word bit, for a binary digit, was suggested by J. W. Tukey.
Simon and Newell predicted a computer chess champion within ten years
In 1957 Simon and Newell predicted a digital computer would be world chess champion within ten years; it took forty, until Deep Blue in 1997.
Simple models and a lot of data trump elaborate models
A 2009 Google essay argued that simple models trained on large amounts of data consistently beat more elaborate models trained on less data.
Siri launched with the iPhone 4S in October 2011
Apple introduced Siri as a feature of the iPhone 4S, announced on October 4, 2011.
Snapchat's My AI first launched only to paying subscribers
Snapchat's My AI chatbot launched on February 27, 2023, first available only to paying Snapchat+ subscribers before a wider rollout.
Solving checkers took 18 years of computation
Proving that checkers is a draw with perfect play required dozens of computers running almost continuously from 1989 to 2007.
TD-Gammon taught itself backgammon by self-play
Gerald Tesauro's TD-Gammon reached strong master level at backgammon by playing against itself and learning from the outcomes, not from human game records.
Tesler's Theorem says intelligence is whatever machines haven't done yet
Larry Tesler's own site says his theorem was 'Intelligence is whatever machines haven't done yet,' often misquoted as being about AI.
The Avianca lawyers were fined $5,000 for ChatGPT's fake cases
A federal judge imposed a $5,000 Rule 11 penalty on the lawyers who filed ChatGPT-fabricated case citations in Mata v. Avianca.
The Bitter Lesson: general methods that leverage computation win
Richard Sutton 2019 essay The Bitter Lesson argues that over 70 years of AI, general methods that leverage computation are most effective by a large margin.
The Connection Machine CM-1 had 65,536 processors
The Connection Machine CM-1 packed 65,536 one-bit processors into a black cube studded with blinking red lights.
The creator of ELIZA became a leading AI critic
Joseph Weizenbaum built the ELIZA chatbot in 1966 and a decade later wrote a 1976 book warning against over-trusting computers with human judgment.
The MATH dataset has 12,500 competition math problems
The MATH benchmark introduced by Hendrycks et al. consists of 12,500 challenging competition mathematics problems, each with a full worked solution.
The Netflix Prize paid $1 million for a 10 percent gain
The Netflix Prize offered one million dollars to the first team to improve Netflix's recommendation accuracy by 10 percent over its Cinematch system.
The Pile is an 825 GiB dataset built from 22 sources
EleutherAI's openly documented language model training corpus, The Pile, totals 825 GiB and is assembled from 22 distinct datasets.
The prize-winning AI artwork took at least 624 text prompts
Jason Allen told the Copyright Office he ran at least 624 Midjourney prompts to reach the image that won the 2022 Colorado State Fair.
The Sims runs on a virtual machine and visual language called Edith
The Sims is built on Edith, a virtual machine whose object behaviors are written in a visual box-and-arrow language.
The support vector machine paper appeared in 1995
Cortes and Vapnik's paper introducing the modern support vector machine was published in the journal Machine Learning in 1995.
The Supreme Court declined to hear the AI-authorship case Thaler v. Perlmutter
On March 2, 2026 the Supreme Court denied certiorari in Thaler v. Perlmutter, leaving the ruling that AI-made work with no human author cannot be copyrighted.
The Transformer paper has eight authors, all then at Google
'Attention Is All You Need' was written by eight researchers working at Google: Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, and Polosukhin.
Training AI on AI output makes the rare cases disappear
A 2024 Nature study found that training models on AI-generated data causes 'model collapse,' in which the tails of the data distribution disappear.
Turing's 1936 paper described a single universal computing machine
On Computable Numbers showed one universal machine can simulate any other given a description of it, the basis of the general-purpose computer.
Uber's car saw the pedestrian six seconds before the fatal crash
Uber's self-driving system detected the pedestrian about six seconds before the fatal 2018 crash but kept reclassifying her.
Watson won $77,147 on Jeopardy!
IBM Watson finished its 2011 Jeopardy! match with $77,147, well ahead of human champions Ken Jennings and Brad Rutter; the winnings went to charity.
Wiener coined cybernetics from the Greek word for steersman
Norbert Wiener took the term cybernetics from the Greek word for steersman to name the study of control and communication in animals and machines.
XiaoIce averaged 23 conversation turns per session
Microsoft's XiaoIce social chatbot averaged 23 back-and-forth turns per session, longer than typical human conversations.
Zillow cut about 25% of its workforce winding down Zillow Offers
When Zillow announced it would shut down its algorithmic home-buying business in November 2021, it said the wind-down would include a roughly 25% workforce cut.
Two 1970s chatbots talked to each other in an internet RFC
A 1973 internet RFC preserves a conversation between PARRY and ELIZA's Doctor, two chatbots talking over the ARPANET.
A US court ruled the ENIAC patent invalid in 1973
In the 1973 case Honeywell v. Sperry Rand, a US court ruled the ENIAC patent invalid, partly because key ideas derived from John Atanasoff's earlier work.
CMU's Harpy understood a 1,011-word vocabulary of connected speech
Harpy, built at Carnegie Mellon under DARPA's speech program, handled continuous speech over a 1,011-word vocabulary using beam search.
Cyc has hand-coded common-sense knowledge since 1984
Doug Lenat's Cyc project began in 1984; by 1995 he reported 'a person-century of effort' had gone into hand-building its knowledge base.
Logistello beat the Othello world champion 6 games to 0
In 1997 the program Logistello swept Othello world champion Takeshi Murakami 6-0, a perfect score in the match.
MIT's Kismet social robot ran on a bank of about 15 computers
Kismet, the expressive MIT social robot from the late 1990s, needed a bank of roughly 15 computers just to power one robotic head.
97 percent of SmarterChild users just chatted with it for fun
ActiveBuddy built SmarterChild to deliver information, but found 97 percent of users were simply chatting with the bot for fun.
The paperclip maximizer comes from Bostrom's 2003 paper, not the 2014 book
The paperclip maximizer - a superintelligence that turns the world into paperclips - first appears in Nick Bostrom's 2003 paper, not his 2014 book.
A 2004 CIFAR program funded the three researchers who would share the 2018 Turing Award
CIFAR's 2004 program under Geoffrey Hinton sustained Hinton, Bengio, and LeCun through the neural-network winter - the three shared the 2018 Turing Award.
In the 2004 DARPA Grand Challenge, the best vehicle traveled only 7.5 miles
No vehicle finished the 2004 DARPA Grand Challenge; per DARPA the top car covered just 7.5 miles and the 1 million dollar prize went unclaimed.
Catapult grew out of an Olympic-prep research project
Catapult Sports was founded in Melbourne in 2006 from an Australian Institute of Sport research project.
The car that won the DARPA Urban Challenge was a Chevy Tahoe named Boss
Carnegie Mellon's winning 2007 DARPA Urban Challenge entry was a self-driving Chevrolet Tahoe named Boss, taking the $2 million prize.
ImageNet was organized around the WordNet hierarchy
ImageNet populated the noun concepts of Princeton's WordNet lexical database with labeled images, borrowing its is-a hierarchy.
iRobot sold over five million home robots within eight years of the Roomba launch
By January 2010, iRobot reported that sales of its home robots had passed five million units since the Roomba's 2002 launch.
Google's self-driving cars had logged 140,000 miles before the project was revealed
When Google publicly disclosed its self-driving car project in October 2010, the cars had already driven 140,000 miles on public roads.
Stanford's 2011 online AI class drew about 160,000 students
Stanford's free fall 2011 online AI course by Thrun and Norvig enrolled roughly 160,000 students across more than 190 countries.
Da Silva Moore was the first court opinion to approve predictive coding
A 2012 opinion by Magistrate Judge Andrew Peck was the first US judicial decision to approve computer-assisted review for e-discovery.
Amazon paid about $775 million for Kiva Systems
Amazon's 2012 acquisition of warehouse-robot maker Kiva Systems was worth approximately $775 million in cash.
The chatbot that 'passed' the Turing test posed as a 13-year-old
University of Reading's 2014 Turing-test claim involved Eugene Goostman, a chatbot posing as a 13-year-old boy that fooled 33% of judges in 5-minute chats.