Today’s episode of the Gestalt IT Rundown unfortunately had a little run in with internet gremlins, and didn’t make the transition from live show to recorded video. It’s really unfortunate because the newest member of the Tech Field Day family, Kori Younger, joined us for the show. She was great! You should follow her on Twitter and at Total Packets. We’ll have her on again before two long for sure.
But there was still a ton of IT news this week. Here’s the Rundown in text form:
On Monday, President Trump signed the American AI Initiative Executive order, designed to boost the country’s AI industry. The order makes federal data available to AI researchers, asks agencies to prioritize worker retraining for changes brought by AI, and directs federal agencies to prioritize AI funding although no new funding is allocated in the initiative. The order also requires the National Institute of Standards and Technology to create standards to allow the development of ““reliable, robust, trustworthy, secure, portable, and interoperable AI systems” and calls for more international cooperation on AI, within the purview of keeping the US in the driver’s seat.
Is this enough to keep (or retake) the US in AI driver seat? It’s encouraging to see something making AI a priority in the country. And I was really surprised to see retraining mentioned explicitly. There’s been a lot of ruminating about the future of work, and it’s good to see that this takes this into account. But when China is treat AI as a national defense issue, it feels like this has to be just a first step.
Uber released an open source AI toolbox named Ludwig, which would allow users to train and test AI models without writing code. Ludwig is built on top of Google’s TensorFlow and allows for models to be trained with a tabular dataset, like a CSV file, and a YAML config file that specifies input and output columns. Ludwig includes a flexible encoder-decoder architecture that can be used for text classification, object classification, image captioning, sequence tagging, regression, language modeling, machine translation, time series forecasting, and question answering. Ludwig and it’s source code are available on GitHub. We’ve seen similar offerings from Baidu and others.
This seems to be a way to keep Uber both engaged with AI developers and also keep them up to date with what people want to do with AI. Building on top of TensorFlow is really smart. I’m skeptical that any organizations will really use this to build anything in production, but it could be helpful to build out POCs or otherwise convince org that additional investment is needed.
IBM announced that enterprises will now be able to run Watson AI services in their own data center. Previously Watson had strictly been tied to IBM’s public cloud offerings. You still need to play ball, Watson services require running IBM Private Cloud, which can be used as a platform to run containerized Watson apps. Amazon is focusing on Watson Assistant, their conversational interface builder, and OpenScale, for AI lifecycle management, at launch. The company is also launching a new version of the Watson Machine Learning Accelerator, a GPU cluster performance option for Power Systems and X86 systems with a promised 10x AI performance improvement over CPUs.
So I guess the strategy hear is Watson on all the things? It’s not a bad strategy per se, but I have to question how long this has been in the works. Clearly this wasn’t IBM’s original vision for Watson. I suspect some disappointing earnings from Watson and an increasingly strict data compliance landscape made an on-prem or private cloud Watson a consideration.
Google has freed its Google Docs automation API from developer preview, making it generally available to all. The REST API is designed to build automated workflows, integrate with content management, and create bulk documents. It supports import and export, so technically you can use a Google doc as the backend and front end for content managment. Zapier has already been using the API in preview, but expect to see more use cases roll out soon.
G Suite has always been based on a “good enough” mentality, banking that people only really need an 80% version of Microsoft Office. But this ups the ante a little, allowing Google Docs and G Suite to fit into more specialized environments. CRM system integration will be huge for Google Docs, and might just be enough to get some people to switch over.
Android Thing is dead, long live Android Things. Google announced they are “refocusing” the IoT platform, moving it from a open OS platform with certified hardware to a platform specifically for OEM partners. As a result, going forward Android Things won’t run on SoCs based on NXP, Qualcomm, and MediaTek designs, although the plan is to still support it on the Raspberry Pi 3.
It’s not surprising to see Google kill a project suddenly but is this refocusing a strategy for success, or an admission that they’re not competing in the IoT hardware platform space? Tough to say at this point. The IoT chip market doesn’t have a ton of SoC hardware players, so maybe this is Google realizing it makes more sense to work with those OEMs directly instead of keeping an open platform that they have to come to.
The Register reported on a trove of logins and passwords being offered for sale on Dream Market. The total haul offers 617 million account details across 16 hacked websites, the biggest of which is Dubsmash, but also include CoffeeMeetBagel, MyFitnessPal, MyHeritage, and Animoto. A total of about $20k in Bitcoin will nab them all. It should be noted that these include encrypted passwords, but the value of the encryption varies wildly. Some of the sites have passwords salted and encrypted with stronger algorithms like SHA512, SHA256 or bcrypt. But 7 of the sites used either SHA1, MD5 or DES encrypted passwords, which are not considered secure for a variety of reasons.
This is an important part in the emerging privacy discussion. Privacy online demands the right to have your data secured. Simply having encryption doesn’t excuse these companies from liability in this breach. MD5 and SHA1 should have been phased out by 2012 at the latest. Breaches of encrypted passwords like this might be inevitable, but at least put up barriers to entry. If these weakly encrypted passwords are cracked, the information in this sale makes it a treasure trove for social engineering.