Followers

Search This Blog

Showing posts with label AI. AGI. Show all posts
Showing posts with label AI. AGI. Show all posts

Monday, January 9, 2023

Book Review-Rebooting AI By Gary Marcus and Ernest Davis

The long title of the book is Rebooting AI: Building Artificial Intelligence We Can Trust.

In the aftermath of the introduction of ChatGPT, a chatbot by OpenAI, and the attendant controversy which the site stirred up, this book can serve as a great introduction to, and provide an explanation of Artificial Intelligence as it is being designed right now.

The reason for the longer title is that the authors do not believe that AI community, specifically the Deep Learning contingent, are ever going to give the world Artificial Intelligence which will come close to replicating human intelligence. Indeed, the authors make a bold statement about WHY the Deep Learning Approach will never come close. That argument is the premise of this book.

The authors are well respected members of the AI movement. Their credentials are impeccable and they straddle the academic and the commercial spheres of AI.

The authors lay out their problems, and the problems with how the current approach to creating AI as it is practiced today, is flawed. They proceed to explain, in qualitative details, how Deep Learning and the statistic based AI methods are just very advanced curve fitting: no more, no less; they also point out why this approach will not ever create anything that will even crudely approximate human intelligence.

They do so in eight chapters.   The first two lay out the problem and gives us a vision of what our expectations should be. The third chapter details the process: Deep Learning, that all the AI world use to create their systems. Chapters four and five make the argument against the premise that Deep Learning is the sole method that is necessary to create human like artificial intelligence. The authors call what is necessary Artificial General Intelligence (AGI). The distinction between what we have versus what we truly want, i.e. human intelligence, is the distinction between AI and AGI.

Chapters six through eight is where the authors make their point about what is missing in just using Deep Learning to build AI, and what is critical and necessary for being able to approximate human intelligence: AGI. They offer up the examples of human traits which are non-numerical, intuitive, causal, and elusive. They also delve into why these traits can not be replicated through programming the AI engine with infinite amounts of training data.

It is necessary to point out that the authors are not naysayers when it comes to the idea of artificial intelligence, they do not outright argue against the enterprise of creating artificial intelligence systems, indeed, they are steeped in the AI world, they are AI enthusiasts, they are AI believers. They are also critical thinkers and cognitive scientists, very well grounded in the human mind and its knowns and unknowns. They give credit where credit is due when they allude to the impressive results of Deep Learning creations; their objection is that the Deep Learning proponents are using just a single note to create the polyphonic sound we hear when we imagine machines that think; because they are neglecting the cognitive and psychological aspects of being human.

Their argument is convincing, because they are clear, concise, and their combined knowledge on the subject is impressive. They lay out their case with an impressive combination of technical papers and study results mixed in with imaginative weaving of common sense. Their logic appeals to what humans have always longed for when we speak of robots and our technical ability to design our dream robots.

As the news of ChatGPT  has made the rounds of popular social media, the mix of concern, amazement, and bemusement has fed the interest of the lay public in the subject, an interest that has even surprised those who created ChatGPT. Yet the discussions have not gotten beyond the superficial layer of AI. This book makes a very fine primer for those who wish to dig in the depths of the subject, at least a little bit. Fortunately, the authors have also provided a substantial list of references for the reader, in case they are curious.

I learned a copious amount about AI research that had been done in the past few decades. Reading the book gave me a foundation upon which I can form an educated opinion. My opinion may not contribute to the state of the art of AI, but it gives me a solid foundation to work from.