Reading Time: 3 minutes

On October first, Artificial Intelligence investors Nathan Benaich and Ian Hogarth released their State of A.I. report for 2020.

It’s a massive collection (177 slides) packed with information about A.I., from defining what A.I. actually is, to uncovering what Google spent training a natural language model with 1.5 billion parameters (think around $10MM!).

You can take a look at the report on their State of A.I. website:

Here are several of the main points I took away:

  • Despite early promise and apparent dominance, Google’s Tensorflow has proven much less popular to those building solutions based on neural networks than Facebook’s Pytorch. Pytorch is the platform we use most at Bennett Data Science, due to its comparative ease of use and flexibility.
  • Artificial Intelligence is being used effectively to translate between two programming languages (for example, from Python to C++ for speed reasons) but also to find bugs in any programming language.
  • Advanced topics such as reinforcement learning are being brought to bear on such areas as drug discovery. These models are able to help algorithms discover drugs that are actually possible to synthesize in the lab.
  • Facebook has made huge leaps in efficient object detection (predicting that an image contains a given object) and image segmentation (creating a mask showing the pixels that make up a given object within an image).
  • While you may not think much about static images you take with your phone, there’s a lot of research being carried out on what is actually not in those images. For example, what’s behind that chair? Or what’s going to happen next in a picture of traffic? Or a stream of traffic images? Answers to these and other similar questions allow for new ways to shop and new approaches to safety.
  • Federated learning is booming – this is where remote devices or browsers contain all the data and model information necessary to self-train, without going to the cloud to do it. What this means is off-grid, highly efficient A.I. on devices.
  • Professors of A.I. are leaving (or rather: being pulled/lured away from) their jobs in academia in record numbers to join powerful tech companies. The Eindhoven Artificial Intelligence Systems Institute in the Netherlands plans on spending 100MM Euros on 50 professors to create their brain trust!
  • We still see far more A.I. job requisitions than available candidates. Universities are continuing to expand to help feed this demand.
  • The first phase 1 clinical trial of an A.I.-designed therapeutic drug begins in Japan. Others are raising a lot of money to join the field.
  • We’re not there yet with driverless cars, as over half of U.S. states have enacted legislation to limit driverless vehicles. Nonetheless, startups in this space are raising billions of dollars in capital.

There’s a lot more to the report. I hope you enjoy nerding-out a bit!

Have a wonderful week.

Of Interest

The Difference Between Dynamic and Personalized
Today, people expect websites and services to be adjusted to their specific situation. Services that are tailored to our needs in the specific moment are more useful to us — let it be the web search of Google, planning the next vacation with Maps, looking up products on Amazon, or searching for the next series to binge-watch on Netflix. We love the customized goods we can buy on Zazzle and the Coke with our name on it. Data scientist Martin Thoma thought about the different kinds of customization of web services and realized that there are important types to distinguish. Most important: Is the service dynamic or personalized – or both?

Algorithms of Social Manipulation
As we all continuously interact with each other and our favorite businesses through apps and websites, the level at which we are being tracked and monitored is significant. While the technologies behind these capabilities provide us value, the tech companies can also influence our decisions on where to click, spend our money, and much more. Read more about these Algorithms of Social Manipulation in this informative article by Diego Lopez Yse.

The Economics of Home Roasting Coffee
Data Scientist Robert McKeon Aloe started home roasting with a popcorn machine six years ago. He was curious if buying a roaster would pay off after a few years, so he bought a variety kit and got to work. In his article  The Economics of Home Roasting Coffee, he shares his cost analysis of home roasting. Turns out, one could save a lot of money given the right roast!