Arm buys a big data company and the Pentagon pays $885M for AI contract
This is a reprint (more or less) of the ARCHITECHT newsletter from July 31, 2018. Sign up here to get new issues delivered to your inbox.
It was a slow Tuesday after a busy last week, but a few items still stuck out as pretty meaningful. Here they are:
- Pentagon signs $885 million artificial intelligence contract with Booz Allen (Wall Street Journal): It’s unclear whether the Project Maven projects that Google recently stopped working on are within the scope of this contract, but it sure sounds like Booz Allen will be working on some similar things, at least. Beyond the huge deal size — $885 million over 5 years — the thing that struck me most is the actual range of stuff Booz Allen will be working on, from health care to battlefield intelligence. As I’ve said before, I understand the ideals of employees who are leery of working with the military, but there’s more to it than just killing and armies aren’t going away any time soon.
- SoftBank-owned ARM is said to agree to buy Treasure Data (Bloomberg): Reportedly for $600 million, no less! I don’t exactly see how this is a natural move as part of Arm’s IoT strategy, as the WSJ and at least one other outlet have suggested. Treasure Data began life in the SQL-on-Hadoop world — and actually created the popular Fluentd open source log-management tool — and currently seems to be pushing its customer-data analytics capabilites (although it does claim some IoT use cases, and the tech seems like a good fit for that size/speed of data). Also, the $600 million price tag would suggest Treasure is doing well financially, which makes me wonder what Arm would do with the existing business, because it is a far cry from chip design. Somebody, please explain to me what I’m missing.
- Despite pledging openness, companies rush to patent AI tech (WIRED): I would like to believe that companies such as Google and Facebook are filing wholly defensive patents because they’re hoping to head off any ridiculous patent-troll lawsuits once the techniques they develop start getting real adoption externally. Heck, back in 2013 Google announced its Open Patent Non-Assertion pledge, where it essentially promised not to sue anyone using a handful of its open source patents, including MapReduce, which was the foundation of Apache Hadoop. And although Google’s PatentShield offering isn’t as altruistic as I remembered, it’s still a somewhat noble gesture. Patent law and technology have always had a strained relationship, but if patents will be issued, I think I’d prefer to see them in the hands of actual tech companies rather than patent trolls or non-practicing entities. That being said, there are plenty of tech companies — including some household names — that are a little more litigious, sometimes indirectly.
AI and machine learning
Test.ai nabs $11M Series A led by Google to put bots to work testing apps
I’m a little confused by the use of the term “bot,” but the general idea here is solid. We’ve all used apps that either don’t work or are entirely non-intuitive. It appears Test.ai is trying to help solve that problem automatically.
DefinedCrowd raises $11.8 million to create bespoke datasets for AI model training
Training data is still the foundation of any good AI system, and getting it is still not a solved problem for mainstream companies. However, it seems like crowdsourcing is not really the answer over the long haul. Also, I still find it kind of funny that we didn’t see such a focus on data quality during the heyday of “big data.” Yeah, we’re talking largely about different types of data and applications, but still …
One in five companies lack the infrastructure for artificial intelligence
I would argue this spans from data systems to application platforms and architectures. I loathe the term, but AI is part of “digital transformation — and a rather advanced part, at that — which means you probably need to be operating like a modern IT shop before you can really expect any AI efforts to pay off. Here’s the report this post is based on.
Google’s AutoML: Cutting through the hype
I trust this is a fair breakdown of the current state of AutoML, and it’s definitely true that there aren’t journalists with the AI chops to accurately assess whether a given approach is good, bad or overkill. However, I would note that a lot of people also get rightfully excited by the idea of certain things, and the ability to get a model without having to build it yourself one of those things.
Cloud and infrastructure
CNCF to host Harbor in the Sandbox
Harbor is a secure container registry created by VMware, Also, it can run in a replicated manner across data centers, which adds some advantages around latency and availability.
Microsoft CTO: Edge computing can make industries ‘massively more efficient’
The big cloud providers spend billions building out data centers and mastering webscale operations. Now, they get to turn their attention to managing edge devices. Done right, though, edge computing should drive even more usage of their cloud platforms because it will generate more data.
Inside the AI-driven shift to liquid cooling at Google datacenters
This is an interesting story — and something also playing out in places other than Google — although it makes me wonder whether there shouldn’t be more of a focus on building AI infrastructure that’s less power-hungry.
Celebrating Istio service mesh project at 1.0 and what it means for the Kubernetes community
Whether or not you care about service meshes, this post does a decent job explaining what they are and how the CNCF’s multiple service mesh projects fit together. I suspect that’s still confusing to a lot of people.
Data and analytics
Dremio raises $30 million from Cisco Investments and others to streamline data management
How many years after Hadoop, and big data analytics still is not solved. The Dremio team has some significant experience in this space, though, and it seems to be picking up traction.
Yellowbrick Data emerges with production customers and $44 million in financing to unveil Yellowbrick Data Warehouse
Another next-gen data warehouse. Notably, it’s in-memory, runs in a small appliance form factor and claims to be designed for hybrid cloud deployment. If you look at the success of other recent data warehouse startups, it’s clear there’s still an appetite for a better experience.
Data’s day of reckoning
I’ve been following this series on O’Reilly for a while, and I hope data scientists and executives are, too. Things really have gone off the tracks with regard to privacy and security, and it’s up to practitioners to take control before more governments start doing it for them.
