(a series of posts for each chapter)
Recently a new book titled “Rebooting AI” by Ernest Davis and Gary Marcus showed up. I have been following Gary Marcus on Twitter for a while. I liked his heads-up tweets to remember that current AI is neither reliable nor that much magic for us to trust it at all. When I noticed about this book, I have ordered my copy immediately. Then started to read this book after finishing my previous reading — “Singularity is Near: When Humans Transcend Biology” by Ray Kurzweil.
In this post I would like to list some quotes from the book which I found them either important or interesting. (Note: some parts of the quotes are paraphrased by myself)
Trending AI Articles:
1. Deep Learning Book Notes, Chapter 1
2. Deep Learning Book Notes, Chapter 2
3. Machines Demonstrate Self-Awareness
4. Visual Music & Machine Learning Workshop for Kids
This book is about;
- how to be skeptical,
- why AI hasn’t been on the right track,
- what we might do to work toward real AI which is robust and reliable so that we can trust it to be with our homes, parents, children, and our lives.
- above mentioned critics (proposed in this book) doesn’t imply that the authors hate AI, but rather they love AI; they want to see AI advance as rapidly as possible.
- It’s partly about what AI can’t do now (and why it matters) and what we might do to improve a field that is still struggling.
1. Mind the Gap
Whenever you hear about a supposed success in AI, ask the following 6 questions to assess the AI’s success:
- What actually did this AI system do?
- Is the result general? Does it do only one aspect of a problem or does it well all aspects of proposed problem?
- Does any demo exist to try it out?
- If people allege that an aforementioned AI performs better than humans, then which humans and how much is it better?
- How far does it take us toward building genuine AI?
- How robust is the system? Does it work well with other data sets without massive amount of retraining (does it have well generalization ability)?
“Big data + deep learning + faster hardware” has been a winning formula
Self driving: highway driving in good weather is relatively doable by narrow AI. However, urban driving is much more complex; what can appear on a road in a crowded city at any given moment is essentially unbounded.
An environment where an AI program trained on reflects the exact dynamics of an environment. On the other hand, real life is open-ended; no data perfectly reflects the ever-changing world. There are no fixed rules, and the possibilities are unlimited.
Currently there is an enormous gap — the “AI Chasm” — between ambition and the reality:
- gullibility gap — humans didn’t involve to distinguish between humans and machines
- illusory progress gap — mistakingly progress in AI on easy problems for progress on hard problems
- robustness gap — once people in AI find a solution that works some of the time, we assume that with a little more work and data it will work all of the time
Our human brains can;
- understand language
- understand the world
- adapt flexibly to new circumstances
- learn new things quickly with limited number of data
- reason in the face of incomplete and even inconsistent information
Current obsession with building “blank slate” machines that learn everything from scratch, driven purely from data rather than knowledge, is a serious issue.
>>> Next post (“Rebooting AI” notes-2):
Don’t forget to give us your 👏 !
“Rebooting AI” notes-1 was originally published in Becoming Human: Artificial Intelligence Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.