The 4 things you should read today: From Google Glass to defending Watson

Tuesday, July 18, 2017 could forever go down as the day Google, er, Alphabet changed the course of th
The 4 things you should read today: From Google Glass to defending Watson
By ARCHITECHT • Issue #114
Tuesday, July 18, 2017 could forever go down as the day Google, er, Alphabet changed the course of the discussion around artificial intelligence and automation in the workplace. I know, I know: It’s a little bit early (and probably more than a little bit optimistic) to go proclaiming Glass Enterprise Edition as a revolutionary technology, but there I think there really is a lot of promise in the product and the idea of augmented reality in the workplace. For jobs where Glass makes sense—Alphabet points to manufacturing, logistics, field services and health care—it could help people do their jobs better and more efficiently without the need for robotics or heavy-handed software systems. 
Some folks have questioned the extent to which automation will actually eliminate human jobs, suggesting instead that humans and machines will co-exist and learn to make each other better. I happen to think there’s something to that argument (in some, but not all situations), but what if equipping people with something like Glass helped them improve enough that it actually mitigates the need for automation on a larger scale? There certainly are reasons for companies to continue employing as many people as possible, and augmenting employees with smart glasses might be a way to get the best of both worlds.
I also realize that Epson, among others(?) has been selling its own take on Glass Enterprise for years, but Epson doesn’t have a global cloud footprint and the hordes of highly skilled software engineers and AI researchers. The abundance of cloud computing power today, as well as the breakneck pace of innovation in AI and hardware architecture, could make this new, targeted version of Glass much more useful than the previous iteration. If Amazon, Microsoft and Apple also get into the glasses game, competition will push the pace of improvement even faster.
It would have been easy for Alphabet to just kill Glass and write it off as a great idea ahead of its time, but instead the company identified its bright spots and worked with that user base to turn it into a targeted product. I might be guilty of over-optimism or just flat-out naïveté, but I also think there’s a there there. In a time where people are fretting about automation while technology is hurtling down the track, I have high hopes for Glass Enterprise Edition and whatever market grows around it.
I linked to the blog post on Glass Enterprise from Alphabet X above, but here are a couple of other good takes on it:
And completely (or mostly) unrelated to Glass, here are three other things I want to point out today:
  • Google Cloud delivers its take on Amazon Snowball: And in the tradition of inspirational Google product-naming, it is called the Transfer Appliance. Plug it into your rack, transfer data to it, and ship it back to Google.
  • Two new security startups launched/raised money: StackRox, which is doing security for container environments (and whose founders will be the podcast guests this week); and Corelight, which is selling a network appliance based on the open source Bro project.
  • An IBM exec rebuts Watson criticism: Vijay Vijayasankar doesn’t work on Watson, but like everyone else at IBM he works with the Watson teams. He makes some fair points about why criticism of Watson is not warranted, but it’s hard to agree with his comments on marketing. As I’ve said before, I think Watson is probably technically capable, but it’s the over-promising that has generated much of the criticism.

Sponsor: DigitalOcean
Sponsor: DigitalOcean
Artificial intelligence
Basically, it sounds like Microsoft will provide the global cloud footprint to accelerate Baidu’s Apollo driverless car project outside of China. Microsoft claims it already has quite a few car companies building or planning to build smart features using its cloud AI services.
And partnered with Peterbilt, which is a huge manufacturer of semi-trucks. For what it’s worth, I don’t think Google Glass will save trucking jobs if trucking are companies are determined to automate them—although regulations might mandate humans in the cabin for some period.
Called CatBoost, it’s designed for ranking and other tasks, and is available as R and Python packages. Notably, it’s not a deep learning library, and there appears to be some question as to whether it’s superior to deep-learning-based approaches.
On Monday, I linked to a book excerpt from Francois Chollet on the limitations of deep learning. Here’s another, this time on its future. Among the interesting predictions are that models themselves will be generated automatically, and that they’ll be able to “learn” continuously across broader areas.
This is a summary of a handful of Microsoft research projects, as well as quotes from Bill Gates, targeting the ability for machines to reason over text and other types of data. Mastering this would obviously be a very big deal.
Uh, duh. Like with all buzzwords, smart buyers will pretty quickly figure out what all these “AI” companies are really doing. Whether it’s technically AI or not doesn’t really matter as long as it works.
Like many other assessments, this one comes down to suggesting many people who want to get started will probably want to use Keras. There are options for the lower-level framework, but TensorFlow is definitely the most popular.  •  Share
This is a fair criticism of Amazon Echo and Alexa, for which there are more than 15,000 skills that are very difficult to find. In theory, more skills are good for the Alexa platform, but you have to think developers will leave if nobody is using what they’re building.
We’re seeing more work in this area, where humans have more control over training reinforcement learning models. In this particular case, the concern is over settings where humans and robots must interact, and there’s a potential for injury during the training process while the robot is learning.  •  Share
Sponsor: CircleCI
Sponsor: CircleCI
Cloud and infrastructure
The most noteworthy (and probably correct) prediction is that SaaS will continue to dominate cloud revenue. The money is in apps and services as infrastructure costs keep going down.
Regular readers know I have a soft spot for Rackspace, and I think this is another good move. Google Cloud is an obvious choice, but Cloud Foundry (via partnership with Pivotal) is strategically interesting not just because it’s a PaaS offering, but also because it brings a hybrid/portability story to bear.
Speaking of Cloud Foundry, here’s a look at how widely deployed it is inside of Comcast. When we’re handicapping which platforms will thrive/survive, it’s always good to consider which ones have the largest user bases.
There aren’t a lot of details here, but let’s assume it’s true. It’s hard to imagine Oracle ever being considered a cloud “leader,” but its existing footprint and cashflow could help it play spoiler as AWS, etc., strive for complete dominance.
This isn’t necessarily groundbreaking, but it’s yet another move by AWS to decentralize where users’ cloud processes are carried out. The cloud computing companies could own the edge, too.
And he does it by making the old “rent vs. buy” real estate analogy. It’s still valid, but as public clouds get better and better, the costs of maintaining that data center have to look worse and worse. At any rate, it’s not a growing business.  •  Share
Compatibility with Java is a pretty good foundation considering how popular Java continues to be.
Sponsor: Bonsai
Sponsor: Bonsai
All things data
Alation isn’t a household name, but it has raised more than $30 million so far, and it compares its technology—which indexes existing data stores based on factors such as usage—to PageRank for enterprise data. 
Notably, the company is headquartered in London. Europe is going to be ground-zero for strict data privacy regulations pretty soon, and anything to help companies keep data safe while still letting them do advanced analysis will be a big help.
This is focused on fitness tracker data, but the idea extends to pretty much anything. Just knowing what isn’t super-helpful if you don’t also know whether it’s good, bad, anomalous, etc. It’s a risk you run when selling data-based products to non-expert users.
This headline asks what will be a very important question in the decade to come, even if we’re not always dealing with outright or intentional lying (e.g., with VW emissions). Algorithms will be wrong, biased and otherwise flawed in part based on the data they’re fed, but identifying and correcting problematic ones is easier said than done.
Sponsor: Cloudera
Sponsor: Cloudera
Did you enjoy this issue?
The most interesting news, analysis, blog posts and research in cloud computing, artificial intelligence and software engineering. Delivered daily to your inbox. Curated by Derrick Harris. Check out the Architecht site at
Carefully curated by ARCHITECHT with Revue. If you were forwarded this newsletter and you like it, you can subscribe here. If you don't want these updates anymore, please unsubscribe here.