I was lucky enough to attend “Demystifying AI - An Introductory Workshop to AI and ML”, an event held at The Yard - Gowanus by the brilliant Sabri Monaf.

The workshop was an introductory event for those who had limited or no AI/ML exposure but were curious about the subject. People like me! This is a hot topic, and it showed. The auditorium was filled to the brim with developers, half of them in the back gallery standing, as the limited number of seats were quickly occupied.

Is this hype cycle any different? Are we witnessing an AI explosion, or is this just another brief peak before a slump?

Sabri introduced the field of Artificial Intelligence, and some of its subfields: machine learning and deep learning. He then proceeded to answer some questions about the direction of AI, which has seen booming periods followed by long winters in the past seventy years. I thoroughly enjoyed this part of the discussion. The topic even resurfaced in the adjecent cafeteria after the workshop was over, when I was grabbing my last cuppa of the day.

How do machines learn

The learning process was then touched on. There was not enough time to get into the mathematical underpinnings in just sixty minutes, but I think the audience walked away with a better understanding of the actual process used in machine learning. Sabri explained the differences between an overfitted model, one that yields precise results with the training data but poor real-world performance, and an underfitting one, or a model that produces weak results both on training and test sets due to the lack of training. Bias and fairness were also discussed, which sparked deep philosophical questions: can biased humans train unbiased models? What happens when people are directly and unfairly affected by AI?

Contrary to normal development, where we usually start with a set of requirements, everything in machine learning demands experimentation to tune “the knobs” of the model.

Finally, Sabri talked about some of the tools of the trade currently employed by AI and ML developers. He also showed a quick example of how to create, train, and use an image classification model, capable of classifying numerical digits in images. With the heavy lifting already taken care of by the libraries, it is nowdays possible to create in minutes what used to take months just a few years ago.

What a brilliant introduction! It left the auditorium crawing for part two!

The Yard at Gowanus, Brookyln.

What’s left to say other than a big thank you, Sabri. And thanks to New York Code and Coffe for hosting the event.

You can find the slides of the talk, as well as the code for the exercise here.