The expanding universe of modern AI

The expanding universe of modern AI

In this article, I go through some history of the evolution of modern AI.

Let's start. The beginning of AI goes back to the 1950's. Back then, AI used hand-coded rules like "if the gameboard looks like this, then this is a good move". But the AI's scope of practical applications was very limited. ML emerges in the 80s and uses statistics to build the model from observed data. So, unlike previous approaches, what machine learning relied on feature extractors, which looked at characteristics of the input data (the input data being a text -for instance) and analyzed it (word analysis -having an automated system that can recognize words that are typical of spam email for instance). After some time and driven by our increased capacity to create data (webcams, social media, etc), a new approach was needed. And that's the dawn of Deep Learning.

No alt text provided for this image

DL is especially useful to work with images (either pictures or videos) and voice and emerges in the context of modern AI. What is modern AI though? Modern AI starts when 3 initial conditions are in place in the market. The first one is big data: you have to have lots of data to teach computers how to perform. The second condition is GPUs. Why? because you need a huge amount of computation to process all that big data, and what a GPUs does is to process all that data in parallel, rather than sequentially. Doing that sequentially on a CPU-only system is just too slow. And third: the Deep Learning (DL) algorithms. The idea of Deep Neural Networks(DNN) has been out there for a while but without big data to train the DNN and the parallel processing of GPUs to run them, their potential was constrained.

No alt text provided for this image

Once those 3 tools were on the table, researchers started publishing their research but also embodied their work in software frameworks that were published to the open-source communities, so others can build on top of their work and start building applications. This scenario called the attention of large cloud service providers like Amazon or Google and they started offering AIaaP which helped to train DNN without having to build your own data center. That helped to develop ideas and out of the ideas, startups were born. And when there are successful startups, the industry leaders start paying attention and investing in those new techs developed by the mentioned startups.

That is -more or less- how we got to the current state of modern AI and how the market developed (and continues to develop). If you are still not clear on what is a DNN and how is it related to DL, what is a framework and what is the difference between ML and DL, or what else is there, take a look at the article below where I try to explain in simple terms the basics of AI.

That's it for today. I hope you had a great weekend. 3pm on Sunday. I'm ready to go for a run. Have a great week, wherever (and whenever) you are reading this.

David

要查看或添加评论,请登录

社区洞察

其他会员也浏览了