Radical Reads

The COVID-19 Startup Wave: The pandemic downturn might yield a new startup wave  (Axios)

“Recessions have often triggered startup baby booms. After the dotcom bust in the early 2000s, a new wave of small companies emerged to build “Web 2.0.” And many of today’s industry leaders got started during the Great Recession of 2008–9.”

Radical Commentary: The job losses after two months of lockdown are staggering: nearly 39 Million people in the US have applied for unemployment benefits, and there are ongoing announcements of layoffs at technology companies. However, we expect that some of those affected will start new companies or join them. While some later-stage tech companies, particularly those with strong balance sheets and/or businesses that are accelerating during the crisis, continue to hire, we are seeing many recently funded startups taking advantage of current labour conditions to hire people who may not otherwise have considered a startup. One of our portfolio companies has more than doubled its full time workforce in the last month alone. We expect that many new category leaders in the decade ahead will look back at this period as key to their success.

AI & COVID-19 Severity Prediction: An experimental AI tool to predict which COVID-19 patients are going to get the sickest  (The Conversation)

“COVID-19 is a new disease, one that doctors haven’t seen before and signs of an impending severe case are hard to spot. AI, which can recognize many elusive patterns simultaneously, is the perfect tool to help doctors identify high–risk patients early. This gives them time to better prepare for these cases and could save lives.

Additionally, the symptoms that the AI algorithms found to be important suggested that SARS-CoV-2 was affecting many more parts of the body than just the lungs. This ability to spot what symptoms are important could help doctors as they search for the many ways the virus attacks the body.

Radical Commentary: This is an encouraging early step in applying AI to predict which patients will become sickest from COVID-19. The dataset here is very small, so much more analysis is required in order to use such tools widely. Canada’s single payer (single source) is a perfect testing ground, particularly given the diversity of its population. We have worked with organizations like the Vector Institute for AI to encourage governments to open health datasets like this to AI researchers. We expect to see many more such applications of AI to help fight COVID-19 in the weeks and months ahead.

China’s Massive Technology Investment: China has a New $1.4 Trillion Plan to Overtake the US in Tech   (Bloomberg)

“Beijing is accelerating its bid for global leadership in key technologies, planning to pump more than a trillion dollars into the economy through the rollout of everything from wireless networks to artificial intelligence.”

In the wake of the economic destruction and enormous job losses caused by COVID-19, some countries are using their massive government stimulus investments to accelerate the industries of the future rather than simply propping up industries of the past. Most prominent is this announcement by China, which is investing $1.4 Trillion into technologies including AI. It is noteworthy that these investments are expected to be made almost entirely in Chinese companies and perhaps some foreign companies that support the Chinese technical ecosystem.

This underscores the trend of technology bifurcation between China and the West (led by the US). We are increasingly seeing two technical stacks being built, and political decisions that force companies and countries to choose between them. The dis-integration could have very significant economic, political and even health implications. For example, if governments preclude technology companies from sharing information to cooperate in developing solutions for COVID-19, will those same governments be willing to permit the resulting solutions to be shared, including to third party countries with relationships with both sides? These are the types of questions that should be considered before any politically motivated technology bifurcation progresses to an irreversible point.

AI & Drug Discovery: Drugmakers get hooked on data  (Financial Times)

“For executives preoccupied with improving pharma’s poor record on productivity, the power of data to accelerate drug discovery and reduce research costs represents fresh hope for corporate profits as well as patients.

…Zach Weinberg, co-founder of Flatiron Health…believes two major shifts in the landscape are driving pharma’s fascination with data. The first is a demand from insurers and other payers for more detailed evidence about the performance of a drug once it is in regular use. The other is the increasing complexity of drug discovery, as scientists’ greater understanding of individual biology drives the search for more personalised treatments.”

Radical Commentary: This article quotes leaders from Pfizer, Roche and Novartis who discuss the growing importance of data to their practices. Key to this shift amongst large pharma players is the increasing value of Real World Evidence data (RWD). RWD is important for drug manufacturers in order to understand the efficacy of drugs beyond control trials. Furthermore, RWD is critical if personalized medicine is to be successfully extended to a wider and more diverse population.

The dramatic increase in volume and quality of data captured in recent years through various digital tools in use amongst patient populations is a trend we expect will continue and accelerate. Needless to say, there are massive opportunities to apply AI to these datasets. These opportunities underscore our thesis on healthcare and our conviction that AI enables healthcare to shift from reactive to proactive, ultimately becoming personalized and predictive.

This article is one only of many that touch upon the applications of AI in Healthcare which is part of FT & Lancet’s Special report on the Future of AI and Digital Care.

Predicting Wildfires with AI: Tracking the tinderbox: Stanford scientists map wildfire fuel moisture across western U.S.  (Stanford)

“Researchers have developed a deep-learning model that maps fuel moisture levels in fine detail across 12 western states, opening a door for better fire predictions.”

Radical Commentary: Scientists use a recurrent neural network trained with field data from the National Fuel Moisture Database combined with satellite sensor data to predict moisture levels of flammable materials. By anticipating where a fire is likely to ignite and how it might spread, this tool may be very valuable to aid wildfire management. This is an example of how a Deep Learning prediction model enables two very different data sources to be combined to produce much more accurate and faster predictions that may help save lives and property.

6) Cloud Supercomputers: Microsoft announces new supercomputer, lays out vision for future AI work (Microsoft Blog)

“Microsoft has built one of the top five publicly disclosed supercomputers in the world, making new infrastructure available in Azure to train extremely large artificial intelligence models…”

Radical Commentary: Microsoft has built a powerful supercomputer exclusively for OpenAI, a leading AI company in which Microsoft invested last year.

Cloud computing has revolutionised the way businesses function. However, supercomputers have been restricted to on-premise owners. The latest development signals that this is changing, with supercomputing migrating to the cloud.

The exclusive partnership provides OpenAI with access to significantly more computational power, allowing the research lab to train and operate enormous distributed AI models, potentially improving the accuracy and scope of AI models in the future.

Over time, major cloud vendors will most likely extend their supercomputer capacity to other customers. As companies use more AI tools, cloud supercomputing infrastructure will become a critical resource.

New Math Frontiers for Neural Nets: Symbolic Mathematics Finally Yields to Neural Networks  (Quanta Magazine)

“Neural networks have always lagged in one conspicuous area: solving difficult symbolic math problems. These include the hallmarks of calculus courses, like integrals or ordinary differential equations. The hurdles arise from the nature of mathematics itself, which demands precise solutions. Neural nets instead tend to excel at probability. They learn to recognize patterns — which Spanish translation sounds best, or what your face looks like — and can generate new ones.

The situation changed late last year when Guillaume Lample and François Charton, a pair of computer scientists working in Facebook’s AI research group in Paris, unveiled a successful first approach to solving symbolic math problems with neural networks. Their method didn’t involve number crunching or numerical approximations. Instead, they played to the strengths of neural nets, reframing the math problems in terms of a problem that’s practically solved: language translation.”

Radical Commentary: The so-called “black box problem” has dogged the field of artificial intelligence for years, as scientists have struggled to explain how neural networks, computing systems modeled on the human brain, arrive at their conclusions. However, by applying neural networks to symbolic mathematics, Facebook researchers Guillaume Lample and François Charton may have developed a means to better understand how neural nets “reason”. By tweaking the mathematical expressions fed into these models, the neural network demonstrates the adjustments it makes to solve the problem. Like a math student forced to show its work, this research offers unique insights into the inner-workings of the models that underpin artificial intelligence.

Charton and Lample’s breakthrough in tackling symbolic mathematics is also notable for its application of neural networks that were originally designed for language translation. As these researchers discovered, integrating expressions in symbolic math lends itself to the powerful pattern recognition available in translation models. This research illustrates the vast flexibility of neural networks to tackle a wide and disparate range of real-world challenges. Like an AI Swiss Army Knife, neural nets offer an ever-expanding set of tools for problems requiring classification, clustering, regression, pattern recognition, dimension reduction, structured prediction, machine translation, anomaly detection, decision making, visualization and computer vision, and the list goes on. And, from early stage cancer diagnostics to the intuitive robotic manipulations that will define the factory floors of the future, for each new problem solved by a neural network, applications with vast economic opportunity are unlocked.