Why is AI happening NOW?
This is a blog post written as a summary of notes, following my learnings from an AI course I completed back in 2019 that I feel is still incredibly relevant in 2024]({{ rel ‘../how-useful-way-my-ai-short-course-from-2019/’ }}).
AI is a huge field with an expansive history extending over the last century, it might be easy to be fooled into thinking thay AI was just something we found in the last 5 years. However, the recent surge of interest cannot be ignored - popularized by easy and cheap access to incredibly sophisticated models like ChatGPT.
Deep learning is specifically the area of AI which is under explosive growth, in what we’re called the modern era of AI (2005+). Deep learning is a subset of machine learning, which has been enabled by a surge in processing power and the ability to store and process huge amounts of data. While the previous century of AI - the early days, were targeted at building perfect models (expert systems, etc), modern approaches to training AI are effectively showing the AI lots of “good results”, and then iterating billions of times to build a neural network - which means we are inferring the AI model from those good results.
It is widely believed that AI, along with digitalization is ome of the enabling technologies for the forth industrial revolution.
- First: ~1765, Modernization with the Steam Engine
- Second: ~1870, Mass production, Electrification
- Third: ~1969, Automated production and information technology
- Forth: Now, Digitialization and AI
I’ve previously known that you “cannot do AI without data” - because AI is inherently trained on good data. An interesting side-effect of Digitialization of modern business is that it’s surfacing huge amounts of data that can be consumed by AI (just a few years ago, the industry was obsessed with “big data”). This is probably another reason that AI has taken hold right now, even if we had some of the technology pieces of AI before - because we didn’t have access to data in the same way - complementary enabling technologies.
A similar complementary enabling technology of the past was electrification - by combining electric generators and electric motors with the electricity grid.
In respect to the technology changes that came with these industrial revolutions, the course also introduced me to Amara’s law - we tend to overestimate the effect of a technology in the short run, and underestimate the effect in the long run. Garner’s hype cycle is a modern application of that law.
The enabling technology for AI, now, are processes that can do matrices very quickly (parallel processing, GPUs), algorithms (ie: backpropogation in neural networks) and access to data - which is probably the most significant technology.
In 2019 we looked at the industries that are making the heaviest use of AI, and at the time it was e-commerce, advertising technologies (adtech), and Finance. Certainly it doesn’t seem like any of these industries are using AI less now than in 2019, but I’ve certainly seen the biggest surge in AI assisted software engineering, assistants (ChatGPT) and image generation.