Chapter 10 – Applying Biomimicking to AI

1. Deep learning and machine learning mean the same thing. (Yes | No)

No. When an  AI program contains a network, especially a deep one (with several layers), that is deep learning. Deep learning is a subset of machine learning.

When programs such as an Markov Decision Process (MDP) are used, that is machine learning.

To sum it up, not all artificial intelligence programs have to learn. Machine learning is a subset of artificial intelligence programs that learn but do not require networks. Deep learning is a subset of machine learning that uses networks.

2. Deep learning networks mostly reproduce human brain functions. (Yes | No)

Yes in neuroscience research on the human brain. Computer models of the brain using deep learning can provide interesting models.

Sometimes yes, when deep learning networks try to reproduce human vision for image recognition applications.

No, when it comes many programs using statistics and probability for language processing, for example.

3. Overfitting is unacceptable. (Yes | No)

Yes, when an application has been implemented and requires the ability to adapt constantly to new data.

No, when an application has been implemented, and the same images are submitted to the AI program.

No, when a prototype is first built and used as a proof of concept that the solution appears interesting.

4. Transfer learning can save the cost of building another model. (Yes | No)

Yes. A reusable model can become quickly very profitable.

5. Training a corporate model on MNIST is enough to implement it on a production line, for example. (Yes | No)

Yes and no. Yes because if a model works on MNIST, it contains the proper functions to learn. No, because if the goal of the project is not close to MNIST data, it will not work. Experiment with MNIST in an image project that contains similar sorts of simple images.

6. Exploring artificial intelligence beyond the cutting edge is not necessary. It is easier to wait for the next ideas that are published. (Yes | No)

Yes, when a team does not have ideas but excels in implementation, for example. GitHub can provide thousands of ideas, along with social networks, publications, and more.

No, when a team is creative and wants to occupy the market with innovations.

7. Some researchers have reproduced all the physical and biological reasoning functions of the human brain in robots. In fact, some robots have human brain clones in them. (Yes | No)

No. Intelligent human-cloned robots only exist in fiction and social hype.

8. Artificial General Intelligence software, a program that can adapt to any human function (natural language processing, image processing, and sound streams) better than a human, already exists in some labs. (Yes | No)

No. This will take much more time if it ever succeeds. Corporations might rather be satisfied with profits generated by narrow artificial intelligence that solves specific problems.

9. Training deep learning networks has become a quick and easy task. (Yes | No)

No. Even if it appears fast and simple sometimes, it is mostly a tough task to build a network from scratch and get it to work properly.