What do you think?
The Shortest History of AI
Paperback
About the author
Toby Walsh
33 books37 followersToby Walsh is one of the world’s leading researchers in Artificial Intelligence. He is a Professor of Artificial Intelligence at the University of New South Wales and leads a research group at Data61, Australia’s Centre of Excellence for ICT Research. He has been elected a fellow of the Association for the Advancement of AI for his contributions to AI research, and has won the prestigious Humboldt research award. He has previously held research positions in England, Scotland, France, Germany, Italy, Ireland and Sweden.
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 10 of 10 reviews
August 24, 2026
Educational, short, dense. I learned a lot…but there was also a strong correlation between reading 10 pages and wanting to have a nap.
March 29, 2026
What I want from such a book times ten. Explained things I wanted to be explained, introduced me things I was now aware of, reminded me things I’d forgotten.
Ofc I asked clawbot to research the book and author before I purchased it
Ofc I asked clawbot to research the book and author before I purchased it
August 30, 2026
A book like this sets the tone from the onset –its title “gives it away”-- and leaves little room for surprise. Indeed, this one did deliver what it intended to do --introduce the reader to the history of the AI, but through a arid tone that could be tolerated only for such a short work.
Starting from its conceptual origins in the 19th century (with the originators of computer such as Boole, Babbage), and taking the reader through the ebbs and flows of its popularity (both its present-day hype and its past “AI winters”), Professor Walsh presents AI’s six significant ideas, the turning points in its short history --especially after 1950s, and its figureheads.
Walsh also neatly divides AI’s history into two segments: the first where AI was built “using symbols,” and the latter (post-1990s) where AI was built “using learning” also known as “machine learning.”
More specifically, then, these six ideas are:
The first idea is “navigation.” This is done by reducing many problems in AI to the computer searching its internal representation of the world from the symbol database and linking it to the real world, (hence, reasoning with use of symbols).
Here, the 1st AI program, the Logic Theorist (LG), proved to be quite adapt, so much so that it found new and shorter proof for mathematical theorems; as well as the STRIPS process that followed LG that was able to break down a problem into plans, find the end goal and recursively find mini-solutions through these plans.
The second idea is “playing games.” Essentially this involves choosing the best possible move, given that the opponent will also choose their best possible move and constantly updating previous uncertainties as much as possible –here, computers’ “forte” come into play, i.e. brute force and precision.
The third idea is “narrow expertise.” This simulates human expertise by following simple rules in narrow domains, which ultimately failed as it failed to take into account “intuition,” that is, the mysterious way to calibrate knowledge through practice and fine-tune it in practice. Incidentally, this failure set the “machine learning” revolution in motion.
The fourth idea is “learning machines.” This is the start of a real revolution of “deep learning” – this is when, the idea of copying human brain with a network of artificial neurons that learn from experience germinated. At the risk of oversimplifying, “learning” was achieved through assigning “weights” to multiple layers of “perceptron” (that is, simple binary models of neurons) networks. This learning process was sped up through “backpropagation” – essentially a process by which weights are updated continuously, and through GPU (the graphics processing unit) that could perform matrix calculation required for this particular task. Finally, “the deep learning recipe - backpropagation over deep neural networks using GPUs to manage the computational load” was complemented with “training data.”
Unsurprisingly, the ChatGPT (GPT stands for Generative Pretrained Transformer) came into being through this idea. Walsh explains its components, and the process quite well for a layman: tokenization through transformers (turning text into a sequence of numbers), transforming a sequence of tokens into a sequence of word vectors, encoding through these vectors (thus capturing complex relationships among words), and decoding (transforming the vectors back into tokens). When all is said and done, the ChatGPT acts very much like “the auto-complete on your phone, but on steroids.” ☺
This fourth idea is also indicative of AI’s weakness: it still does not understand the language as a human, cannot know what is true (only what is probable) and can struggle with reasoning that we take for granted.
The fifth idea is “reinforcement learning.” This is essentially learning from experience, rewarding success and penalizing failure. While developers, early on, had difficulty on how to go about assigning “credit” for doing the right thing, this problem was resolved relatively quickly as computers were left to do whatever they were programmed to do constantly billions of times (that humans could never do!) to then learn from “their” mistakes.
Finally, the sixth idea is “the application of Bayes’ theorem.” This is computing a probability of an event, given evidence about it at that moment, and then updating that probability as new events occur. On the face of it that is straight-forward. In real life, where reasoning has to be exercised to choose between running the car into a brick wall to avoid running over children in the middle of the road (or vice-versa) can be more problematic.
The book has two major shortcomings. First, and most glaringly, it does unnecessarily focus on the biographies of influential players so much so that the readers gets the feel of an HR agent skimming through CVs - as impressive as they may be. Second, Walsh is aware but not terribly alarmed about the existential threats AI poses – read, AI “singularity.”
All in all, a decent work if one is interested in the topic, and can muster strength to plough through its relatively dry prose.
Starting from its conceptual origins in the 19th century (with the originators of computer such as Boole, Babbage), and taking the reader through the ebbs and flows of its popularity (both its present-day hype and its past “AI winters”), Professor Walsh presents AI’s six significant ideas, the turning points in its short history --especially after 1950s, and its figureheads.
Walsh also neatly divides AI’s history into two segments: the first where AI was built “using symbols,” and the latter (post-1990s) where AI was built “using learning” also known as “machine learning.”
More specifically, then, these six ideas are:
The first idea is “navigation.” This is done by reducing many problems in AI to the computer searching its internal representation of the world from the symbol database and linking it to the real world, (hence, reasoning with use of symbols).
Here, the 1st AI program, the Logic Theorist (LG), proved to be quite adapt, so much so that it found new and shorter proof for mathematical theorems; as well as the STRIPS process that followed LG that was able to break down a problem into plans, find the end goal and recursively find mini-solutions through these plans.
The second idea is “playing games.” Essentially this involves choosing the best possible move, given that the opponent will also choose their best possible move and constantly updating previous uncertainties as much as possible –here, computers’ “forte” come into play, i.e. brute force and precision.
The third idea is “narrow expertise.” This simulates human expertise by following simple rules in narrow domains, which ultimately failed as it failed to take into account “intuition,” that is, the mysterious way to calibrate knowledge through practice and fine-tune it in practice. Incidentally, this failure set the “machine learning” revolution in motion.
The fourth idea is “learning machines.” This is the start of a real revolution of “deep learning” – this is when, the idea of copying human brain with a network of artificial neurons that learn from experience germinated. At the risk of oversimplifying, “learning” was achieved through assigning “weights” to multiple layers of “perceptron” (that is, simple binary models of neurons) networks. This learning process was sped up through “backpropagation” – essentially a process by which weights are updated continuously, and through GPU (the graphics processing unit) that could perform matrix calculation required for this particular task. Finally, “the deep learning recipe - backpropagation over deep neural networks using GPUs to manage the computational load” was complemented with “training data.”
Unsurprisingly, the ChatGPT (GPT stands for Generative Pretrained Transformer) came into being through this idea. Walsh explains its components, and the process quite well for a layman: tokenization through transformers (turning text into a sequence of numbers), transforming a sequence of tokens into a sequence of word vectors, encoding through these vectors (thus capturing complex relationships among words), and decoding (transforming the vectors back into tokens). When all is said and done, the ChatGPT acts very much like “the auto-complete on your phone, but on steroids.” ☺
This fourth idea is also indicative of AI’s weakness: it still does not understand the language as a human, cannot know what is true (only what is probable) and can struggle with reasoning that we take for granted.
The fifth idea is “reinforcement learning.” This is essentially learning from experience, rewarding success and penalizing failure. While developers, early on, had difficulty on how to go about assigning “credit” for doing the right thing, this problem was resolved relatively quickly as computers were left to do whatever they were programmed to do constantly billions of times (that humans could never do!) to then learn from “their” mistakes.
Finally, the sixth idea is “the application of Bayes’ theorem.” This is computing a probability of an event, given evidence about it at that moment, and then updating that probability as new events occur. On the face of it that is straight-forward. In real life, where reasoning has to be exercised to choose between running the car into a brick wall to avoid running over children in the middle of the road (or vice-versa) can be more problematic.
The book has two major shortcomings. First, and most glaringly, it does unnecessarily focus on the biographies of influential players so much so that the readers gets the feel of an HR agent skimming through CVs - as impressive as they may be. Second, Walsh is aware but not terribly alarmed about the existential threats AI poses – read, AI “singularity.”
All in all, a decent work if one is interested in the topic, and can muster strength to plough through its relatively dry prose.
April 2, 2026
Before I read this short book I knew next to nothing about the development of AI and its future challenges. Now I know a little more and feel a bit more confident.
Isn't that the point of the 'Short History' series?
Isn't that the point of the 'Short History' series?
September 16, 2026
Perhaps is should have known this would be an oversimplification of AI given the title but this was truly a beginners guide. Beyond that expect it to massively understate the potential dangers and negative aspects of this technology.
March 23, 2026
A Very insightful book, highly recommended.
May 20, 2026
Helped me understand AI and made me realize how AI is not “new”.
June 19, 2026
A fun and interesting read for someone wanting to learn about AI. I appreciate the anecdotes, examples and length of this book!
September 1, 2026
I’m scared😟😟😟😟😟
Read
May 25, 2026Very clear explanation of the development and principles of AI. The author is optimistic - the negatives of AI are listed quickly in the last chapter. I liked the overview of the problems AI developers tried to solve, and what tools were applied to go further. Despite not having any subscriptions or using any free tokens, I use AI every day, e.g. identifying plants in iNaturalist - it's embedded in the app/website. I like knowing what's going on. Will not detail my many concerns here.
Displaying 1 - 10 of 10 reviews