There have been quite a few announcements lately regarding machine learning, so I try to give an overview of the ones that caught my attention. And Hyperdev has been rebranded and extended into Gomix.
Machine learning updates
There was a lot of news concerning various machine learning (or AI if you prefer)1 related efforts this week. First though, let’s have a look at what Amazon announced during Re:Invent for this.
AWS announced 3 major new parts of their machine learning strategy. First an image recognition service called Rekognition. This is similar to other offerings, including one from their cloud competitor Azure, and does what you’d expect from a service like this. You can do image processing either real-time or in batches, and even compare these images. As far as I can tell it doesn’t stand out in any way, other than being easy if you’re already using AWS.
Next is Polly, which is a text-to-speech service. Unsurprisingly this is well-tested by way of Alexa, the service behind the Echo, so I’m sure it works well. In English. However, it can apparently support 24 languages (and 47 different voices), which isn’t something the Echo can do. One of my major complaints about the Echo is that it’s only officially available in the USA and UK, and therefore only needs to speak English. I really hope this is a precursor to making the Echo available world-wide, because who doesn’t want a device in their home that listens to everything they say and sends it to an online store?2
And speaking of Alexa, the last and biggest part of the new offerings is Lex. A slightly strange name, but a very interesting functionality. In short, Lex is a direct interface to the Alexa conversational functionality. This means it enables you to build conversation models in order to get a more friendly interaction with a bot or other service. Now, I’m not convinced that bots, let alone a conversational interface, are the best way to do everything, but for anything that uses actual speech to interact it’s a necessity. Even if it’s just Siri’s usual “I’m sorry Arjen, I didn’t get that”3.
These are of course nice toys to play with, and it will be good to see what can be done. But last week Amazon also announced something new that uses their machine learning technologies. Amazon Go is their new physical store, with “walk out technology". Let’s forget about the terrible name of the technology4, and look for a moment at what it does. You use an app to identify yourself when you enter the store, and then you’ll be constantly monitored while a machine learning algorithm judges what you’re buying and therefore what you should be charged for. Then when you walk out of the store this is automatically charged to your Amazon account. It sounds very futuristic, and interesting, but also as something that might have some issues at the start. For example, it’ll be interesting to see how it reacts to children running around putting things in their parent’s bag.
Obviously Amazon isn’t the only one who works with machine learning and the implications of the release of several machine learning environments as open source is likely to have a bigger impact in the long run. OpenAI released Universe, their platform that is used to train robots to learn how to play games and use apps in order to get a better understanding of the physical world. Around the same time, Alphabet released DeepMind Labyrinth which is the maze they use to train DeepMind. Both of these are available on GitHub and hopefully their availability will lead to more advancement in his field. After all, to ensure progress it’s good when everything is open so everyone can learn from each other.
And that openness seems to be coming to the company best known for not being open as well. Apparently machine learning researchers at Apple will finally be allowed to write papers about their work. Whether this is a reaction to the impression that Apple is behind in this field, or a part of Apple slowly becoming more open5, doesn’t matter as this is a case where the result is more important than the reason.
Gomix
Back in June, I mentioned the beta release of Hyperdev a tool for easily building websites by live editing the code. Now it has been rebranded under the name Gomix. While holding true to the beta version in some ways, this new version is more focused on building bots and apps than websites. It’s still very possible to build sites though, but it seems more focused on the others.
Additionally, it has made it far easier to start with a project by introducing what it calls mixing. This means that you can take an existing bot, application, or website and easily extend this to have the functionality you want. As before, the idea behind Gomix is very interesting as a playground, proof of concept, or for simple one-person projects. I haven’t tried it yet to see if the integration with GitHub has improved but right now it’s definitely a better thing to play around with.
-
I have to be honest here the rebranding from AI to Machine Learning, or Deep Learning as AWS calls it, isn’t a very smooth process. Both terms are used to describe the same things by different entities, and I for one can’t really be bothered trying to find a distinction. As the naming convention seems to move towards machine learning, that’s what I’ll use regardless of what the source articles say. ↩︎
-
Obviously, that is a privacy concern, like every other IoT device (even when they’re not hacked). That said, I’m still interested in owning an Echo, even though my lack of actually buying one might paint a different picture. ↩︎
-
Yes, Siri has gotten better over the years and it works a lot better on my watch than anywhere else, but it still has too much trouble understanding me to be something I can rely on. ↩︎
-
From the people who brought you the one-click order technology. ↩︎
-
While the company is as closed-mouthed about future products as ever, they have been improving about letting people know about other things and opening up their code. There’s still far to go, but at least it’s moving in the right direction. ↩︎