This week is all about AI: Intel's new AI processor, OpenAI and Microsoft team up, Google brings GPUs to the cloud and invests in a Montreal AI research group.
Plus our favorite articles from the past week about AI and some of the projects making us smarter.
There's AI Inside
Intel unveiled a new AI processor designed to help with both training and execution of deep neural nets. They’re calling it Nervana, based on the technology originally built by the startup of the same name Intel acquired earlier this year.
Intel plans to have prototypes ready by the middle of next year, with a market-ready chip by the end of 2017.
NVIDIA is the primary supplier of GPUs, which Microsoft, Google, Facebook, and Baidu typically us to train their deep neural nets. Intel is pushing to be part of this potentially enormous market despite that the hardware used to train and execute deep neural networks is changing.
For instance, Google’s developed their TPU. Even Microsoft is evolving, moving to FPGAs, a type of programmable chip.
OpenAI + Microsoft
Microsoft is partnering with Elon Musk’s OpenAI to help protect humanity’s best interests – and to make Azure the primary cloud platform for their compute needs.
OpenAI will get access to the Azure N-series virtual machines and GPUs.
Microsoft says its goal is democratize AI and make it accessible to everyone.
Hmm, that sounds vaguely familiar to some company building a marketplace to democratize access to algorithmic intelligence.
AI Groups at Google
Google announced a new Cloud Machine Learnings group headed by Fei-Fei Li (Stanford Artificial Intelligence Lab) and Jia Li (former head of research at Snapchat).
The news welcomes GPUs to the Google cloud, which will offer the AMD FirePro S9300 x2, as well as the NVIDIA Tesla P100 and K80 GPUs for deep learning, AI, and high-performance computing applications. This is AMD’s first move into deep learning.
Google's also invested invested $3.4M in the Montreal Institute for Learning Algorithms, an academic fund covering three years for seven faculty members across various Montreal academic institutions.
To top that off, they're opening a brand new deep learning and AI research group in their Montreal offices.
What We're Reading
- Q & A: The future of artificial intelligence. It's the study of methods for making computers behave intelligently. Roughly speaking, a computer is intelligent to the extent that it does the right thing rather than the wrong thing. (berkeley.edu)
- The Sky(net) isn't falling. Yet. We've come a long way with our smart machines. But we've got a long way to go before they become truly intelligent, as in having feelings, consciousness and free will. (CBC.ca)
- Humanity and AI Will Be Inseparable. By 2021, everyday software will be vastly more intelligent and powerful, replacing humans in more and more tasks. How will we keep up? (Verge)
- Artificial Intelligence and life in 2030. Every five years, a study panel is convened to assess the current state of AI, and what we have here is the first report. It focuses on what impact AI will have on life and society (in urban cities) by 2030 – close enough in to be imaginable, without straying into science fiction. (The Morning Paper)
- Who Will Command The Robot Armies? Today, having built the greatest apparatus for surveillance in history, we're slow to acknowledge that it might present some kind of threat. We would much rather work on the next wave of technology: a smart home assistant in every home, self-driving cars, and rockets to Mars. (Idle Words)
- Facebook Must Really Suck At Machine Learning. Given how similar fake news detection is to related problems such as search index spam, ads landing page spam, social networking bots, and porn detection this suggests one of two things: (1) Facebook really sucks at machine learning or (2) Facebook does not want to address the problem. (Elad Gil)
Things Making Us Smarter
- TensorFlow implementation of Conditional PixelCNN
- PixelCNN++ OpenAI Implementation of PixelCNN
- DeepMind's Differential Neural Computers (DNC)
- A Python Wrapper for the New York Times Archive API
- Reinforcement learning with unsupervised auxiliary tasks
- Can a neural network learn to recognize doodles?
Watch: TESLA Autopilot
Emergent Future is a weekly, hand-curated dispatch exploring technology through the lens of artificial intelligence, data science, and the shape of things to come. Subscribe here.