Started Labbook 2023.
December 22, 2022
- Next year less cybernetics and robot history and more focus on Brooks! Directly going from Baitenberg vehicles to Brooks' behaviors.
- Why do I use for Brooks a scanned PDF, while PDF of the published version is online available.
November 30, 2022
- For the new course 'Kantelpunten informatiewetenschappen’ this post, on the women behind the ENIAC.
November 16, 2022
- Next academic year will be the last year of this course.
- In 2024 the Turning Points of Information Science will start, given by a selection of Professors of the Institute.
November 11, 2022
- Slide 13 with the history of Deep Learning from the Workshop DL could be useful in my 3rd lecture.
- The workshop also has two interactive sessions with Colabs
- One recommendation was to read Peter Bloem's blog on Transformers.
- This is the 2nd Colab
- At the last page of the workshop some links to other courses are given.
November 8, 2022
- Perusall has a gradebook, which can be downloaded as csv file (which I hope to read on my home-computer).
- The bonus can be calculated from the Average score column (divided by 3).
- Tried to for Norvig column to Sync to LMS, but didn't see them in the Canvas grade book.
- According to documentation there should be an additional button Release grades, but I don't see that button. The only button is on top: Unrelease all grades.
September 29, 2022
August 11, 2022
- On singularity, Russel & Norvig point to metacrap, which was not even available in the Wayback machine (but luckily perserved in digital history).
- Metacrap was an earlier blog-publication of Doctorow, the singularity reference is The Rapture of the Nerds post-cyberpunk science fiction book.
-
- In the computer science section, the tensor processing unit (TPU) is already mentioned.
- In the cybernetics section, the Ratio club is mentioned, a club where professors where banned.
- More details can be found here.
-
- In the AI history, I cover most mentioned Turing Award winners, except '; Ed Feigenbaum and Raj Reddy (1994) for
developing expert systems that encode human knowledge to solve real-world problems;'.
- In the Big Data section, Russel & Norvig point to Banko & Brill Scaling to very very
large corpora for natural language disambiguation, claiming that doubling the dataset outperformes any algoritmic tweak (for NLP).
- In state-of-the-art, the AI100 project is mentioned, including AI index (with the metrics per country).
- Turing warned in 1951 on consciousness machines, pointing back to the 1872 book Erewhon.
- Note that the 2021 version of this Introduction seems to be only slighlty longer (35 pages vs 31 pages). Note that Figure 1.1 (four categories) is gone.
- The Introduction refers often to chapter 28; Philosophy, Ethics, and Safety of AI. Yet, that are also a rather large chapter (30 pages).
- Note that the last chapter (Future of AI) is conventiently short (10 pages).
August 10, 2022
- Read the first chapter of 4th edition. Layout seems not optimal for Perusall, found a better Global layout elsewhere.
- A big difference with the previous edition, is that now an intelligent machine should pursuing our objectives, but is uncertain what they are (which leads to agents provable benificial to humans.
- In the bibliographic notes, they reference Nils J. Nilsen's book, from which a preprint is available on his personal website.
- In the philosophie section, the confirmation theory seems to be new.
- The General Problem Solver of Newell and Simon is here called a greedy regression planning system (Chapter 11).
June 20, 2022
- Rodney Brooks points in his last IEEE column: "In 1979, Kunihiko Fukushima first published his research on something he called shift-invariant neural networks, which enabled his self-organizing networks to learn to classify handwritten digits wherever they were in an image.", which was forther worked out in 1989 by Yann LeCun.
- Kunihiko Fukushima paper SELF-ORGANIZATION OF A NEURAL NETWORK WHICH GIVES POSITION-INVARIANT RESPONSE can be found at page 291-293 from the IJCAI proceedings.
June 18, 2022
June 17, 2022
June 3, 2022
May 19, 2022
- Would be great to add a workshop to train cats & dogs with Teachable Machine.
- The training is as fast, because it is uses pre-trained models (Mobilenet, where they apply transfer-learning).
May 9, 2022
April 21, 2022
April 4, 2022
March 22, 2022
- This chess post gives some more background on Claude Shannon's chess ideas.
- Shannon published on this subject in 1950, one year after Turing (and 6 years before the birth of AI). Programming a Computer for Playing Chess, Philosophical Magazine, Ser.7, Vol. 41, No. 314 - March 1950.
March 15, 2022
March 11, 2022
February 21, 2022
- Read the short story by Rodney Brooks on Claude Shannon, with the anecdote that Arthur Samual implemented Shannon's idea to improve the evaluations of board positions and published the first paper with "machine learning" in the title (1959):
Arthur, Samuel. "Some studies in machine learning using the game of checkers." IBM Journal of research and development 3.3 (1959): 210-229. (PDF)
- In the English wikipedia page of El Ajedrecista, the algorithm of the machine is described.
January 12, 2022
- Google Research created a 4 PB dataset on a small sample from the human brain, which resulted in 13- million annotated synapses:
Previous Labbooks
