AI, ethics and the AWS container strategy comes around

OK, let’s get this out of the way: Amazon Web Services launched a ton of new services and product updates yesterday. These range from a deep-learning-powered camera for developers to an infrastructure-less container service; and from a serverless Aurora database service to a new graph database service.

I can’t possibly comment on everything (nor would I want to), so here’s a link to the AWS blog with everything it has announced: https://aws.amazon.com/blogs/aws/category/events/reinvent.

I’ll do a more thorough rundown of interesting stuff (including AI) in the next issue, but here are some thoughts on the container services AWS announced (which seemed to garner a lot of attention):

This is a Kubernetes managed service, obviously. There’s every reason to expect EKS will be competitive with alternatives from Google and Microsoft, if only because AWS already owns such a large share of cloud workloads and, even more so, of Kubernetes workloads. AWS is also smart to work with partners like Heptio that have a fundamental understanding of the technology and community. 

That being said, Microsoft is doing some really impressive work in this space and building up a really good team. Google created Kubernetes, probably has the most mature Kubernetes service at the moment, and just this week (see below, and yesterday) made it cheaper and easier to use Kubernetes on its cloud.

And all of that being said, now that every cloud provider has a bona fide Kubernetes offering, the real point of competition is once again the ecosystem and the platform as a whole. Even if containers and Kubernetes become the dominant ways of packaging and managing applications in the years to come, they don’t exist in a vacuum. Providing the best Kubernetes and overall cloud-native management experience will definitely be an advantage, but it’s still just one part of a continuum that spans from storage up to voice-recognition APIs.

Fargate is a service for managing containers without dealing with AWS instances at all, and is probably where AWS would ultimately like to see this space head. Just like with Amazon Lambda, its serverless service, the goal is to empower developers even more by further abstracting infrastructure-level concerns. It’s easy to look at these things today and, accurately, suggest that no serious business would build or manage mission-critical applications with tools like this … but that’s also what a lot of people said 12 years ago about Amazon EC2. 

A service like this also happens to be a great way for AWS to charge premiums on convenience and further increase utilization of its physical servers that run all this stuff.

I really like this piece from Mike Loukides at O'Reilly about ethics at scale (even if it draws on an old example). The premise is that we need to thoroughly think through how our ethical, and legal, rules change in a world where algorithms can make important decisions, a single data breach can effect hundreds of millions, and even our physical security is threatened by connected locks and other devices. And all it takes is one stupid little mistake. 

And then there are the fake news and privacy angles, which Loukides also hits upon.

I wrote about this at least once at Gigaom back in the day, in the context of how social media has skewed what used to be our right to privacy in America. One little post, or one unflattering photo posted by the right person in the right place, and lives and reputations can be thrown into turmoil. If you ask me, we’re still a long way from figuring out how to get a handle on what the internet has become.

On a related note, this blog post from DeepMind is worth checking out, as it addresses the blowback from its problematic data-sharing agreement with a United Kingdom hospital. Given the history, the explanation about why DeepMind’s Streams product for detecting acute kidney injury doesn’t use AI seems a bit white-washed. The company does, however, make a great point about where we’re actually at when it comes to forcing AI into everything.

Basically, it turns out that hospital staff actually benefit more from just having tools to make their days and tasks more efficient. As the author notes:

This is probably an underappreciated reality in a lot of different industries where AI is being pushed as a savior. It might not be a horrible idea to take a step back, assess the actual reality for most industries and even consumers, and then take a pragmatic approach to making them more intelligent.