As we dig into 2020, it’s worth taking a moment to look back at the last year in relation to the advances for AI/ML platform developments. This is where I had a chance to spend time in learning and developing knowledge, and it’s gone from curiosity to more of a passion play at this point.
First, the landscape. At its core, Artificial Intelligence is machines doing things that degrees of human intelligence can do, and Machine Learning is a machine doing things that humans can’t really do at all, or if they can do it, the work takes immense effort and time.
While machine learning and the concept of AI has been around for over 50 years, the concepts were hindered by a combination of computing cost and scale. As Amazon, Microsoft, Google evolved their businesses, they developed massively scalable machine learning platforms for personalization, advertising, search and extremely sophisticated methods of tracking, influencing, and effectively controlling the buyer — you and me.
Amazon, Google, and Microsoft hiccup and a billion dollars in profit comes out. In the last couple of years, each have pivoted to productize and expose AI/ML platforms for the outside world. I speculate this is going to make them money on scale that makes traditional compute look trivial.
AI/ML Services Enter Prime-Time
Let’s take a look at what happened last year.
In April, GoogleNext curated incredible sessions on AI/ML use-cases. Around this conference, AutoML Tables and AutoML Vision were announced. These platforms are machine learning tools, being developed so non-data scientists can build, test, train, and refine models without having to call on a team of DevOps, data scientists, and developers to accomplish the initial feat. Granted you still need to have mastery of your data, ML concepts, and the tooling itself, but this advance leveled up on Amazon and Microsoft ML tools in a major way, by making machine learning accessible to people without CS PhDs or formal data science training.
Google’s stepped up approach also attacks a major problem — data science unicorns don’t exist on scale, and where they do exist they are being recruited into the major leagues well before they exit academia (so they can expand on the machines giving us AI/ML to play with).
By summer, re:MARS was held in June 2019 and from attending was by far the best conference I’ve attended in 20 years. re:MARS was a scaled up concept of the annual exclusive MARS summit, tailored for the inaugural 3000 attendees in Las Vegas. Topics hyper-focused on machine learning, automation, robotics, and space. Amazon curated all recordings and this is a treasure trove for anyone looking to learn more about AI/ML. The majority of vendor placements showcased robotics and automation examples:
While there were no AWS service announcements, this was the first time a major cloud service provider invested heavily to center attention not so much on technology, but the applied and potential use-cases for AI/ML both on Earth and beyond. The closer for me was an Apollo era rocket engineer describing how AI/Robotics in space will recycle and build the majority of things, with reusable components and zero waste.
By December, at the AWS re:Invent conference, AI/ML featured prominently and again you have a complete archive of session recordings. The Machine Learning 1/2 day summit demonstrates Amazon’s shifting focus to applied and ‘for the greater good’ use-cases, along with a wider discussion of ethics and social impacts of these platforms. But the headlines (for data science nerds and people who support these efforts) was the introduction of SageMaker Studio. This development environment includes a raft of tooling to automate, debug, and auto-recommend models and essentially remove some of the initial knowledge barriers to teams using SageMaker.
Notably, SageMaker AutoPilot can study your data, compare results of multiple models, and make recommendations on which is best suited to your use-case/data. Yes, there is AI embedded into your AI development platform, so the machine tells you how best to use the machine. AWS is also packaging pre-build machine learning frameworks to service broad industry use-cases, including Personalization, Forecasting, Fraud Detection, and Code Debugging (AI showing humans how to fix their shitty code). So again, within the span of 6 months a major player levels up the field competing for the early adopters and mindshare as this marketplace continues to take shape.
I hope you’ve enjoyed this journey for AI/ML advances in the last year. Don’t follow them, let them follow you.