How to get an opinion about Deep Learning and humanity future
Any technology magazine and major newspaper has at least 1 piece every week about Artificial Intelligence and how it will revolutionize, for better or worse, our future. A lot of articles have a philosophical viewpoint about the role of humans in world future.
I’m intrigued and excited by these conversations about our near future, but as a technical person I often distrust technology backed apocalyptic visions of years to come. A lot of Journalists love to write alarmist stories about facts that they barely understand and to provide readers catchy social-shareable title. Check for example the story about Facebook’s AI agents that created their own language.
The best way to get an informed vision about things is to go as deep as possible with facts. That’s one of the reasons I decided to look for a Deep Learning course. (Pun not intended)
Starting from a good understanding on Machine Learning theory and techniques, I soon realized that reading introductory articles was not enough to get a good understanding of this hot technology with an evocative name.
I decided to invest my time in Microsoft Deep Learning Explained free course. Lessons are easy to follow even without a solid data science background. In 3 weeks, with a 4–6 hours per week effort, I’ve been able to go deep enough into Deep Learning theory and practice.
Using free Azure Jupiter notebooks I was able to complete all the practical exercises using Python and Microsoft Cognitive Toolkit without any system set-up hassle.
Tutorials of modules 1 to 3 are based on MNIST database of handwritten digits, in each module the course introduces a different approach in training a model able to predict handwritten digits. The first module uses Logistic Regression (a Machine Learning technique not based on deep neural networks) to show that results are pretty poor. The 2nd and 3rd modules explain how to use Multi-layer perceptrons and Convolution Neural Networks to gain much higher success rate in digits prediction.
The way tutorials are structured allowed me to get deep enough into training phase and understand how each parameter influences results without the need of spending hours in coding.
Now I understand how Deep Learning techniques can be applied to some real-world use cases and I have the tools to debunk most of the generalist news about AI.
My next step in search for the truth will be to get a meeting with Elon Musk and ask him if he really fears “intelligent” computers more than stupid humans.