Book Review
Title: Is the Algorithm Plotting Against Us?: A Layperson’s Guide to the Concepts, Math, and Pitfalls of AI by Kenneth Wenger
Genre: Non-Fiction, Technology
Rating: 3.75 Stars
The opening to this book was interesting as it has the aim of educating the reader about AI and the mathematics involved behind the scenes. The author rightly states that the media overexaggerates how far AI has come in recent years leading to a lot of fear and misunderstanding about what AI actually is. Now I don’t normally delve into technology because I don’t understand a lot of the technical aspects but since this is supposed to be a layman’s guide I thought I would give it a try.
Chapter 1 introduces us to the idea of polarization and its consequences, it begins by introducing us to the history of the human mind. We look at how the neuron was discovered and how we learnt of its function and while this alone was an amazing leap in science for humankind, it links even further with how these neuron pathways were used as the basis for almost all modern technology that we have today. McCulloch-Pitts developed the artificial neuron using logic gates and almost all modern technology is based on logic gates. The simplest logic gates are NOT AND and OR, AND gates are used in heavy machinery and are essentially used as a double switch safety function, so the machine will not operate unless both of these switches are “on”. NOT gates are used as signal inverters commonly found in car fuel sensors. The sensor detects the fuel level and when it falls below the sensor’s range then it indicates using a switch that fuel needs to be added. OR gates are used in simple payment systems where the person can pay by either cash or card like in train stations. If no payment is made the gate remained closed but once either payment is made the gate opens.
These logic gates can also be combined to make complex circuit systems, if enough are combined you can build modern computers. Logic gates are also the backbone of most programming languages and it was here that scientists realised that the logic gates could be paired with the artificial neuron and could theoretically produce a functioning model of the human brain. However, it was soon realised that binary inputs were a drawback and this led to Donald Hebb introducing Hebbian learning. Hebbian learning proposed that neurons firing together, strengthened connections between them and was vital to the learning process. For the artificial neuron they realised they had to give weight to the connections and this meant you could fine tune the neuron by adding or decreasing the weight of a connection. Frank Rosenblatt went one step further and developed the perceptron.
The perceptron changed the binary input to a value between 0 and 1 which was more closely aligned with biological neurons. Through this two approaches to creating models were developed: monotypic and genotypic. Monotypic is non adaptable while genotypic is, this meant monotypic was used for studying the brain with a specific set of inputs and desired outputs while genotypic uses well defined functions and compares these to the artificial system. The genotypic approach allows more flexibility in artificial networks and this showed the human mind relied on a statistical system rather than decision based one. For those working on artificial systems this offered a reduction in the number of systems needed for artificial intelligence. In reality, the human brain actually used both systems as it uses specific algorithms for special functions and generical algorithms for most other functions. This introduces the idea of a bias and we are focusing heavily on the input/output process.
The perceptron actually had many practical applications as it was first used in IBM 704 software for punch cards. This software distinguished cards punched on the left versus the right and was later designed for image recognition for images up to 20×20 pixels which developed further into a system that could eliminate noise from phone lines. These systems were ADALINE and MADALINE which are still used today, although it was later pointed out the single-layer perceptron couldn’t implement the XOR logic gate. The single-layer perceptron only solves linear problems and most problems aren’t linearly separable so system needed to be adapted again and this was the beginning of the AI winter.
The book goes into much greater depth about how AI actually work, however, I did find that have a good understanding of either mathematics or computers would be extremely helpful when reading. While it claims to be a layman’s guide, it does get very technical and while the author does their best to explain it I found myself struggling at times. Overall, I found the book to be informative and interesting but slightly too technical to be a true layman’s guide to AI. If you have a background in maths or computer science then this might be the perfect read for you.
Buy it here:
Paperback/Hardcover: amazon.co.uk amazon.com
Kindle Edition: amazon.co.uk amazon.com
Comments