November 11, 2023

• The American Executive Order asks for Chief AI Officerrs in every agency.

October 12, 2023

• The Spotify AI translator I have seen last Friday, is from Rask.
• Spotify used it for translating Podcasts, like Lex Fridman interviewing Yuval Noah Harari.
• More details can be found at Rask.
• Read this blog about Simpson paradox, strafication and the backdoor criterium.

October 3, 2023

• Read Smolensky's Connectionist AI, symbolic AI, and the brain (Jun 1987) to see if his mental categories included words and letters.
• According to Smolensky, the subsymbolic description is closer to neural systems than symbolic systems. Between neural structures and the subsymbolic system he indicates 'abstract & formalize', while two 'approximate' steps are included to mental structures of the symbolic systems. As inbetween represenation he indicates 'higher-level descriptions'.
• Smolensky dedicates a whole section on Semantic interpretation. In his eyes, the actication passing rules have essentially a different character than symbol manipulation rules. Although a pattern could be mapped to a single concept, it cannot be associated with a single node, but is a distributed activation spread over many nodes.
• On page 105 he writes. In the subsymbolic paragdigm, semantic interpretation occurs at the level of patterns of activity. ... In the symbolic paradigm semantically interpretetable entities are symbols, which are combined by some form of concatenation: activity patterns superimpose on each other, like wave-like patterns.

September 29, 2023

• Read Learning representations by back-propagating errors (1986)

September 27, 2023

• I didn't know that Turing suggested to build a humanoid robot:
  
  'One way of setting about our task of building a “thinking machine” would be to take a man as a whole and try to replace all the parts of him by machinery.
  He would include television cameras, microphones, loudspeakers, wheels and “handling servo-mechanisms” as well as some sort of “electronic brain".'
• In Intelligence without Reason Brooks discussed Turing's idea qutie extensively, including his ideas on embodiment (p. 4), although he recommended to study various games and crypt analyses first, because "in that they require little contact with the outside world", with the footnote Interestingly, Turing did not completely abstract even a chess playing machine away from embodiment, commenting that "its only organs need be `eyes' capable of distinguishing the various positions on a specially made board, and means for announcing its own moves"
• This same aspect is mentioned in the Annals of History article. The draught player programmed by Strachey on the ACE notified the opponent that it was his or her turn with a pip-pip sound, and printed messages like remarks like "I REFUSE TO WASTE ANY MORE TIME. GO AND PLAY WITH A HUMAN BEING".

September 26, 2023

• Found this article on Alan Turing in 'Annals of the History of Computing' (Sep 2023).
• One new insight was the discussion on heuristic search. One heuristic that was used when operating the bomb, was the Herivelismus heuristic, which relied that a part of the Enigma's was only modified a few turns, so starting in the neighborhood often worked.

September 21, 2023

• The claim that logic can use any knowledge hangs a bit. The logic part is clear, but for knowledge it would be good to show a bit on knowledge enigneering.
• The OpenCyc initiative still exist, although Cyc is now mostly commerically initiative. One product is hospital-billing, other is enterprise employee knowledge.
• Nice is that on platform page, the tupple Deduction, Induction, Abduction shows up.
• Ronald gebruikt als voorbeeld the semantic web, maar geeft geen voorbeeld van de Google info-summaries die je krijgt bij een zoekopdracht.
• The latest update on high-impact papers can be found yearly blog.
• Could only find a list for CVPR, which is actual #4 publication medium (after Nature & Science). NIPS is #9.

September 11, 2023

• Read the conversation with Rodney Brooks: Beyond Computation.

September 7, 2023

• Found the Zoom meetings of the Lectures in the Syllabus page from this year Canvas.
• The procedure to publish the Perusall-grades is to do 'Release to students' in Perusall first, followed by a 'Posst grades' in Canvas.
• Modified all Assignment settings that the 3 points don't count to the grade.

September 1, 2023

• Watched the 20 minute video on Zipline's technology.
• One of people behind Zipline is Keenan Wyrobek, one of the founders of ROS.
• Their design for their silent propellors is also quite impressive.

August 15, 2023

• Had some difficulty to activate Perusall. Read both TLC manual (which seems outdated, because adding the external url failed). Yet, the popup window indicated that my perusall account was already coupled to a canvas account.
• Instead used Perusall getting started and added Perusall in the navigation option.
• Now I get a Perusall course home.

July 27, 2023

• How to read ML papers with AI Magic.

June 15, 2023

• Made a first version of this year's Syllabus.

May 9, 2023

• The University of Groningen published the book The Shallow and the Deep: A biased introduction to neural networks and old school machine learning, where the first three chapters (e.g. From neurons to networks) seems interesting for this course.

April 4, 2023

• This essay is quite well balanced, of a professor which didn't sign the letter of concern, but thinks that AI should be more involved in the discussion. Especially like the concern he rises in his section 'Imbalance of Power'.
• In addition, all three engineering solutions (alignment, explainability and transparency) face serious logical contradictions, which make them non-solutions in his eyes.
• Luckily Ali Minai also has some engineering advice which could work a little to lower the risk.

March 24, 2023

• A nice overview of Deep Learning in a nutshell is given by Nvidia, in part 1, part 2, part 3 and part 4
• Part 2 covers the history, Part 4 reinforcement learning.

February 7, 2023

• The current estimate is that next year there will be 10\% less students, so specified as maximum size 200 students (last year 178 registered students). That means that I could use L1.02 with a capacity of 240.

January 18, 2023

• Start reading in the 115 pages of From Cybernetics to Brain Theory (Arkin's memoir, 2018).
• Arkin claims in his first chapter: "It is also salutary to read Mind, Brain and Adaptation in the Nineteenth Century: Cerebral Localization and its Biological Context from Gall to Ferrier (R. M. Young, 1970) to see how many of our present questions were already being explored in the flush of new insights into evolution and brain localization decades before Santiago Ramon y Cajal (1899, 1911) and Charles Sherrington (1906) ushered the investigation of the anatomy and physiology of neural circuitry into the 20th century."
• Also a nice anecdote: "Soon after my arrival at the high school (when I was 11), Fred lent me his copy of Mathematics and the Imagination (Kasner & Newman, 1940) which introduced me to the mathematics of infinity, probability, and topology. The authors coined the terms googol for 10100, and googolplex for 10googol to make the point that incredibly large numbers are nowhere near infinity. Much later, this led to the naming of the company Google and the choice of Googleplex for the name of their complex of buildings. But for me, back in the 1950s, this book set the pattern of reading mathematical books for pleasure, expanding my horizons far beyond the classroom materials".

January 17, 2023

• Should check the 2022 book The Road to General Intelligence for chapters useful for IKI.
• Another interesting read would be LeCun's A Path towards Autonomous Machine Intelligence (Jun 2022).

January 13, 2023

• Started reading Jurgen Schmidhuber's History of Modern AI and Deep Learning
• He points to his blog Turing Oversold.
• Jurgen's claim on founder of computer science goes from Ramon Llul (13th century), via Algebra of Thought (1686) to Thoralf Skolem (recursive functions, 1923).
• Leibniz designed the step reckoner, which "could perform all four arithmetic operations and store intermediate results on a Leibniz wheel. Blaise Pascal already made Pascal calculator, which could do the addition/subtraction. The step reckoner was still based on decimal arithmetics, although later the Algebra of Thought laid the foundation for Boolean Algebra. The universal calculator of Ballage was partly so hard to build because it was still using decimal arithmatics. With binairy computations the first binary computers could be build (by Zuse). Nice is the quote that Leibniz mission was inspired by Lull "If controversies were to arise, there would be no more need of disputation between two philosophers than between two accountants. For it would suffice to take their pencils in their hands, to sit down with their slates and say to each other [...]: Let us calculate!"
• Leibniz built several versions of the Step Reckoner; only one survived:
  
• it was Turing's work (1936) that convinced Gödel of the universality of both his own approach (1931-34) and Church's (1935). Church and Post universal model of computing were more powerfull, but the simplicity of Turings (quite inefficient) model made it a convenient tool for later theoretical studies of complexity.
• Yet, these were still theoretical constructs, while Konrad Zuse was in 1936 already working on a practical computer.
• Read Descartes’ influence on Turing, which claims on page 7 that Turing was for his 1950 paper was insprired by the neurosurgeon Geoffrey Jefferson's "The Mind of the Mechanical Man' (1949), which points to Descartes Language test (and Karel Čapek’s Robots)
• In a note on storage, Jurgen noted that Ancient machines stored programs on rotating cylinders; Jacquard stored them on cardboard; Zuse stored them on 35mm film
• Jurgen's main claim is that most histories introduce a AI-winter after Minsky & Paperts 1961 book, while at that period already deep learning solutions where invented in Russia (by Alexey Ivakhnenko, 1965) and in Japan (Shun-Ichi Amari, 1967). Yet, this is a less known fact in Europe / America.
• Also, the ReLU was invented by Kunihiko Fukushima in 1969. Also the alternation of convolutional and downsampling was invented in Japan, only called NeoCognitrons (instead of CNNs).

Previous Labbooks

• Labbook 2022
• Labbook 2021
• Labbook 2020
• Labbook 2019