Google goes peacenik, chip wizardry and AI gets into art and drugs

Roundup While we've already covered a lot of AI stories this week a few slipped under the radar so, as is traditional, here's the roundup of some news you may have missed.

Google turns peacenik: Under fire for helping the US military use AI to better bomb people, Google has not only stepped away from that particular Pentagon contract but also this week released a set of seven principles for the development of AI.

CEO Sundar Pichai made a point of noting that they are "not theoretical concepts; they are concrete standards that will actively govern our research and product development and will impact our business decisions."

And they are pretty good: be socially beneficial (the AI equivalent of 'Do no evil'?); don't introduce biases; be safe; respect privacy; be accountable; be scientific; limit abuse. They also sparked Twitter's CEO Jack Dorsey to ponder out loud whether they were something the tech industry as a whole could get around.

Should other companies building AI commit to these principles (or those of another company), help fund an independent organization to steward, write their own and share, or/and open source principles on @github? https://t.co/FiTa2mP0LC

— jack (@jack) June 8, 2018

But while Pichai added a section at the end called "AI applications we will not pursue" in order to indirectly address the AI-bombing-Pentagon-contract issue, you do have to wonder whether, in real terms, the principles will amount to much when faced with the lure of dollar bills.

You can read them here.

Video understanding competition: Google have the second Youtube-8M Video Understanding Challenge to encourage the development of AI algorithms that can automatically label YouTube videos.

There is a lot of interest behind teaching agents to learn about the world through videos. A lot training datasets for autonomous cars contain thousands of short driving clips. DeepMind, also recently taught a bot to play Montezuma’s Revenge by watching humans play the game on Youtube.

But before the machines are trained, the datasets have to painfully annotated. Road markings, trees, or cars have to be singled out in order for a car to recognise objects in its surroundings. It’s tedious work that is normally done manually or through services like Amazon Mechanical Turk.

Google’s YouTube-8M Large-Scale Video Understanding Challenge aims to automate that labelling process. It started at a workshop at the Conference on Computer Vision and Pattern Recognition last year.

“The purpose of the competition was to accelerate improvements in large-scale video understanding, representation learning, noisy data modeling, transfer learning and domain adaptation approaches that can help improve the machine-learning models that classify video,“ it said in a blog post.

Google has updated the dataset. It now includes less videos (6.1 million), but are at a higher quality. Developers looking to enter the competition should train and build audio-visual content classification models using the updated Youtube dataset, and submit a paper to the European Conference on Computer Vision (ECCV) in September in Germany.

The top 5 teams will be awarded $5,000 to support their travel to ECCV and attend the workshop based around the challenge.

New accelerator chips: Imagination Technologies, a British semiconductor biz, have released two new chips to run neural networks on edge devices.

Codenamed AX2185, one chip is aimed at smartphones or security cameras, whereas the other AX2145 suits more low-powered applications like TVs and smart cameras. It’ll help power things like facial ID, voice recognition, object detection - the usual.

At the moment, the chips only support Caffe and TensorFlow and are only for inference. Developers will have to train their models on their desktops or on the cloud before.

It’s difficult to say much else beyond the usual marketing fluff as the company did not answer any questions about benchmarks.

Stephen Alderman, on the business development team for vision and AI, at Imagination Technologies, did tell us that the AX2185 has performance of 4.1 TOPs and the AX2145 is at 1TOPS. And that the AX2185 was 3.5 times faster at running AlexNet than Nvidia’s Titan X chip. So there’s that.

AI + Art: Mozilla announced an award of $225,000 (~£186,000) to fund projects that combine art, machine learning and AI.

“In a world where biased algorithms, skewed data sets, and broken recommendation engines can radicalize YouTube users, promote racism, and spread fake news, it’s more important than ever to support artwork and advocacy work that educates and engages internet users.,” it explained in a blog post.

The project can be anything from videos, games, data visualizations, or even browser extensions. The fund is split into: Two $50,000 total prize packages ($47,500 award + $2,500 MozFest travel stipend) and five $25,000 total prize packages ($22,500 award + $2,500 MozFest travel stipend).

Applications open on 18 June until 1 August. You can apply here.

Let's hope the results are more insightful and less disturbing than the AI-generated nudes created by researcher Robbie Barrat.

Barrat fed a generative adversarial network thousands of nude portraits and got it to create its own version of the artwork it processed. The results can be see on his Twitter feed: think a cross between Salvador Dali and Francis Bacon without, you know, the skill or insight.

I've been using transfer learning and very small datasets to try and nudge the nude portrait AI into generating more of the portraits that I like. pic.twitter.com/jRO9qqzFkC

— Robbie Barrat (@DrBeef_) June 7, 2018

UK opens new AI drug lab: The UK government announced it would be funding the Rosalind Franklin Institute at Oxford University, a new research lab focused on using AI to discover new drugs.

It has pledged £103 million ($138,226,000) to the institute, £6 million will be spent on:

• The World's most advanced real-time video camera, the key to a new technique that uses light and sound to eradicate some of the most lethal forms of cancer.
• A new project pioneering fully-automated hands-free molecular discovery to produce new drugs up to ten times faster and transform the UK's pharmaceutical industry.
• A ground-breaking new UK facility that will revolutionise the way samples are produced and harness Artificial Intelligence (AI) to generate new drugs for clinical testing within a few weeks.

The development will be led by Ian Walmsley, Pro-Vice Chancellor at the University of Oxford.

“The new Rosalind Franklin Institute will lead a revolution in drug development and diagnosis to improve the lives of millions of patients. And with over 10 million people in Britain alive today expected to live to 100, now more than ever it is vital that the Government invests in the development of new technologies and techniques which will support people to have healthier lives," Greg Clark, the Secretary of State for Business, Energy and Industrial Strategy, said. ®