Innovation Watch #16
This content was sent on july, 10th via our Tech Watch Newsletter
We’re still in the early days of July and it feels more comfortable, the air is breathable again. The heatwave is over and we can get out of our (wifi-less) caves to get back to our normal life. Kinda. Provided you’re not already on holidays. If that’s the case, please enjoy yourself and don’t forget to send us a postcard 😉.
In this edition, we’ll talk about #AI and #computer-vision again, which start to takes over human activities, for the good or the bad, only time will tell us. But being able to predict breast cancer 5 years before any oncologist seems a pretty good idea in our opinion #tech4good. A lot of #privacy concerns also, with Alphabet building their vision of future’s #smartcity, London’s police facial recognition fiasco and the issues of #cashless society in HK. If you scroll at the end, you’ll find some useful and playful tools 😄.
This is Smile’s Innovation Watch.
Automated translation isn’t something very new. Most of you are using Google Translate or its likes and, at least for Google’s one, they are not relying on 1:1 translation of text in a different context but the predictive model looks for vectors in a huge database. Let’s say that words are usually consistent in terms of distribution in a text, whatever language is; and context can be seen as a vector of word A — word B + word C = word D which is also kinda the same across languages. Where word can be words/expression. That’s this kind of vector matching that is used now to translate your text and can be applied to decipher ancient text you don’t know the language.
McDonald is reportedly looking at two big innovations: Automatic Voice Recognition for drive-thru and robot to cook your food (so far, just the automatic fryer). Voice recognition is a field we have been exploring for more than a year at Smile and it makes total sense for us to see this technology coming to that very controlled environment with a very limited set of items in the menu and a very specialised field of activity (ie. ordering food). Also, it allows being automatically crossed with Dynamic Yield, the tech that profiles you as a customer and tries to upsell you with a very targeted add-on.
MIT AI model can predict breast cancer 5 years earlier than human, works equally for white & p.o.c. woman
We already tweeted about the study that discovered that an AI model was able to foresee breast cancer in a patient from 3 to 5 years prior a human (who then considered it as a false positive). But the issue with most of those models was the lack of training set from people of colour. That’s why MIT tweaked its model and tool to achieve the same accuracy for white or p.o.c women. Representativity is always a huge issue in using computer vision with human detection and everyone needs to keep it in mind when working on a project involving AI and picture of people. 👍
Privacy •/ˈpɹaɪ. və. si/
(Reads in 🇫🇷) Google’s project of building a smart-city of 76 hectares in Toronto is going well but more and more concerns are being raised from citizen and corporation. Because of the hyper-connected city with millions of sensors, it’s a bit unclear what Sidewalk Lab, the Alphabet company in charge, will do with all those data. If you are wondering, this is the company we talked about in the previous issue of our tech watch. And yes, Ap[ple is already trolling them.
A few issues ago, we spotted the fact that London’s police is rolling out facial recognition system in the wild for real scale test using special van equipped with a lot of cameras, arresting people that don’t want to be scanned and try to cover their face and force enrol them in the system. Seems it’s not “production-ready” yet. Meanwhile, San Francisco banned the technology for public agencies use last month.
Hong Kong is facing more and more protests as new laws would make extradition possible to mainland China or their allied (HK has always been a refuge for people that are against China state power where they can seek asylum) and protester realises that their cashless world can be used against them. In HK, public transportation, taxis, even small groceries are handled with an all-in-one card called the Octopus card which became a central database of all your activities in the real world. Protesters start reintroducing the usage of cash to buy single-ride tickets (even if they cost a bit more than the Octopus card system). “Cash protects our autonomy, and indeed our human dignity”.
Openness • /ˈəʊ.pən/+/nəs/
Training computer to recognize things or make decisions is always tricky. And the most essential part is training your model/neural network to do what you want to teach them. You need for that representative data set to train, with little bias possible (or enough you want to introduce in your predictive model). When it comes to a 3D environment like the world around us, it’s difficult then to keep a consistent data set to use to train then benchmark the model. That’s where HabitatAI comes handy. It let you explore a high fidelity 3D scene and give it to your neural network to play with.
Fun • /fʌn/
Ok, it’s summer and maybe you feel a bit tired, maybe you don’t have that much of work, or maybe there’s just the new season of POSE on Netflix and you JUST.CAN’T.MISS.THIS. So here’s a Chrome extension that will help you. It will make your Netflix movie look like a conference call.
Using GIT isn’t super easy for the first time, and sometimes, it even becomes a little nightmare for you. As your GIT sherpa may not be available to answer all your questions, here’s a comprehensive database of Q&A about all the little GIT issues you may have.
Here’s away mode. An Alexa skill that plays a weird audio recording of people being home. Yes. That’s it. Anyone approaching your flat / house will think someone is here. Genius.😴
That’s all folks!