Microsoft has an AI coming-out party

Microsoft has been a leader in artificial intelligence research for years, even if it sometimes gets
Microsoft has an AI coming-out party
By ARCHITECHT • Issue #111
Microsoft has been a leader in artificial intelligence research for years, even if it sometimes gets overlooked as people ogle research labs at Google (plus DeepMind), Facebook, Baidu and even OpenAI. Lest we forget that, the company made a slew of AI-related announcement on Wednesday, highlighted by the creation the new Microsoft Research AI group led by Eric Horvitz. That group will focus on a wide range of technologies that includes robotics, deep learning, reinforcement learning, conversational systems and more.
Bloomberg positions the new group as an attempt to compete for mindshare with the research groups at the aforementioned companies and institutions, and that’s probably fair. As I noted above, Microsoft has been doing machine learning research for years and over the past few years has made some meaningful contributions to emerging AI fields such as deep learning. It also has released quite a few AI products and open source tools. But it 
The idea behind the new group appears to be that by bringing researchers from across the company’s various divisions under one roof, they can work collaboratively and produce more meaningful results. Specifically, it will place an emphasis general AI systems that can solve a broad range of tasks, and complex or multi-faceted tasks. Hopefully, they can also generate big headlines and attract talented researchers, which are still in high demand—especially as companies such as Uber, Tesla and others also start ramping up their AI efforts.
(As an aside, though, Microsoft might consider streamlining its various research efforts. Microsoft Research AI is being compared with Google DeepMind, although that’s an analogy I might have saved for the Maluuba team that Microsoft acquired earlier this year. The new team is also a subset of the 5,000-person Microsoft AI and Research Group that the company announced in September, led by Harry Shum. It’s a lot to parse.)
In other news, Microsoft also announced:

Sponsor: CircleCI
Sponsor: CircleCI
Sponsor: Linux Foundation
Sponsor: Linux Foundation
Artificial intelligence
Seriously, if you’re interested in how people interact with robots or other digital “beings,” this is a must-read. It’s focused on Alexa, but as they get smarter, I suspect these lessons can apply to industrial robots, as well.
The author suggests AI will take jobs and not replace them, but that humans can remain relevant by focusing on careers that utilize our primary advantage: love. So many thoughts about this …
That’s “deep” as in deep learning. I know people have good things to say about Darktrace, which also raised a bunch of money this week, but I’m curious to know how effective these AI-for-cybersecurity tools really are.
They won’t all be intelligent in any meaningful way, but it seems like a safe bet that a good number will be. How this all translates into job losses/functions remains to be seen.  •  Share
This is a smart application, which has already been proven to work inside companies like Google and Facebook. It could be a moneymaker for Watson.
That’s field as in not in the office, not as in farming. This is actually pretty cool, especially the idea of letting them train their own classifiers for job-specific items.
Yes, this type of technology could be used to educational or information purposes. It also could be used for fraud. The good news for us peons is that it takes a lot of data to pull this off.
Well, in a manner of speaking. The new system, called SCAN (Symbol-Concept Association Network) learns about things in an unsupervised manner, and then combines them to form new “concepts.” And here’s another new piece of research from DeepMind, where it shows that systems designed to carry out simultaneous tasks can actually outperform those trained to carry out individual tasks. These are both, by the way, steps toward the kind of general intelligence Microsoft wants to create with its new group, and which will make AI much more generally useful than it is today.
Basically, this is an alternative to standard reinforcement learning, wherein systems learn by doing and being rewarded. It’s part of a broader trend toward being able to train machines on less data or different types of data that are easier to model, obtain, etc.  •  Share
Sponsor: Bonsai
Sponsor: Bonsai
Cloud and infrastructure
The company predicts an increase of 4.5 percent over the whole year, with software and cloud computing chipping in a lot of that growth.  •  Share
This is one of those reminders that despite all sorts of competition, Intel is still the force to be reckoned with in the data center. But besides supporting Skylake processors, Microsoft supporting Intel’s Altera FPGAs could also provide a boon.
It’s hard to look at this without commenting on the sexual harassment angle (read the post for more on that), but let’s remember that Uber probably will be a company for a long time, and it has lost an awful lot of engineers executives and leaders this year.  •  Share
This is one of the first examples I’m aware of that has reached deeply into enterprise IT, as Artale and Ignition do a lot of enterprise deals. A lot of heads are rolling, but when will behavior actually change?
Although, for what it’s worth, Microsoft made the same claim back in May … I would imagine IBM, Amazon and others have something to say about this, too.
The move once you reach a certain size appears to be opening your platform to developers, thus becoming the hub connecting a whole lot of other services. Workday joins Salesforce, SAP and other application providers, but also, of course, general-purpose platforms like AWS and Microsoft.
Sponsor: Cloudera
Sponsor: Cloudera
All things data
I love Elton John, but I want to question his investment in a startup called Qloo that claims to have “built deep data science about culture and entertainment.” However, he claims it has been “a powerful tool” for his production company. So there you are.
Probably not, if I’m reading this post correctly. It’s a topic that has been around since the inception of big data, and while there are some victories and some light shed on bad practices, regulating how private companies use data is still really tough.
This story is actually out of Australia, but it’s a great illustration of what can go wrong when we apply data without common sense, compassion or context. Also, selling personal data is a practice that needs some more limits.
The WindyGrid system described here sounds really valuable, and is another example of how far cities are coming when it comes to data collection and analysis.
It’s limited in scope and probably utility, but this is still a super-cool dataset if you think about the types of simulations and planning that could be done with larger-scale scans. 
Sponsor: DigitalOcean
Sponsor: DigitalOcean
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.