This book review was written by Eugene Kernes
“AI is a field that includes a broad set of approaches, with the goal of creating machines with intelligence. Deep learning is only one such approach. Deep learning is itself one method among many in the field of machine learning, a subfield of AI in which machines “learn” from data or from their own “experiences.” To better understand these various distinctions, it’s important to understand a philosophical spilt that occurred early in the AI research community: the split between so-called symbolic and subsymbolic AI.” – Melanie Mitchell, Chapter, Page 24
“It’s no secret: deep learning requires big data. Big in the sense of the million-plus labeled training images in ImageNet. Where does all this data come from? The answer is, of course, you – and probably everyone you know. Modern computer-vision applications are possible only because of the billions of images that internet users have uploaded and (sometimes) tagged with text identifying what is in the image.” – Melanie Mitchell, Chapter 6: A Closer Look at Machines That Learn, Page 93
“The phrase “barrier of meaning” perfectly captures an idea that has permeated this book: humans, in some deep and essential way, understand the situations they encounter, whereas no AI system yet possesses such understand. While state-of-the-art AI systems have nearly equaled (and in some cases surpassed) humans on certain narrowly defined tasks, these systems all lack a grasp of the rich meanings humans bring to bear in perception, language, and reasoning. This lack of understand is clearly revealed by the un-humanlike errors these systems can make; by their difficulties with abstracting and transferring what they have learned; by their lack of commonsense knowledge; and by their vulnerability to adversarial attacks. The barrier of meaning between AI and human-level intelligence still stands today.” – Melanie Mitchell, Chapter 14: On Understanding, Page 212
Is This An Overview?
There are many different approaches to creating Artificial Intelligence (AI), to creating machines with intelligence. Deep learning, is a subset of machine learning, in which machines learn from data or their own experiences. Deep learning requires data, much of which is obtained from various free digital sources in which humans tag images with identifying text. Using user data not only to sell the data to other firms, but also to improve their products. Machines learn in a supervised learning procedure, in which different weights are applied to process examples. AI can also learn through trial and error, with randomly chosen weights. There are limits to AI learning, as machines do not learn on their own, they do not engage in open-ended categories, and they do not actively seek information.
There is a barrier of meaning for AI. They do not understand the meaning of the
questions asked of them. Computers do
not understand the meaning of situations they encounter. For a computer, meaning is derived by the way
the symbols can be combined, operated on, and correlated. AI has difficulties with abstract
information, and transferring knowledge from one information domain to
another. AI performs well on narrowly
defined tasks, in which the situations are similar and are highly
expected. AI has a higher chance of
making errors in unexpected situations that occur infrequently. This is known as the long-tail problem, for
the vast range of unexpected situations that AI can encounter.
Do AI Think, See, And Speak?
For some, thinking only occurs in biological entities because biological entities have a conscious. An awareness of their own actions and feelings. No machine has a conscious, therefore cannot think.
Machines have difficulty with object recognition because programs see pixels and cannot easily differentiate between the objects that the pixels can form. The objects themselves can appear very differently in different images. Correlations within images does not mean that the computer will properly identify the appropriate object. Humans are assumed to know what an object is, no matter the image. But there is much less proof that a computer actually sees and classifies an object appropriately.
AI can read the information that is there, but cannot
extrapolate based on information not present.
Does not understand what is left unsaid.
Making it difficult to understand language.
What Is The Future Of AI?
There are many potential futures for AI such as AI going rouge, taking over jobs, and make autonomous decisions that are not understood. AI can possibly make human creativity and emotions, basically the human spirit, easy to reproduce.
AI can enhance the quality of life, but there are
limitations to AI safety. There is
disagreement about how to proceed with AI, either to embrace their capabilities
or approach with caution given AI vulnerabilities. AI should be regulated using experiences from
AI practices and government agencies. Neither
alone can be trusted. There are ethical,
political, and technical decisions that need to be made.
Caveats?
This is not a book about the popular diverse future perspectives on AI potential or what AI would do. This is a book about the methods used to train AI, and the limitations to AI learning.
