Image for post
Image for post

Every week, my team at Invector Labs publishes a newsletter to track the most recent developments in AI research and technology. You can find this week’s issue below. You can sign up for it below. Please do so, our guys worked really hard on this:

Image for post
Image for post

From the Editor: AI Learns Things that it Was not Trained For

Our brains are magical factories to produce emerging behaviors. When we need to learn a new task or skill, we don’t follow a strict script but we rather develop unique behaviors that are better equipped for our experiences. The development of emergent behavior is even more visible and fascinating in infants that always surprise us about the specific way they master a new task. But what about artificial intelligence(AI) systems? The traditional way to think about AI is that we train them to master a given task and we expect them to perform in a specific way. But what happens when AI starts developing new behaviors that we didn’t account for?

This week, AI powerhouse OpenAI unveiled a fascinating experiment in which they trained a group of AI agents to learn the game of hide-and-seek using traditional reinforcement learning techniques. The researchers were surprised by a group of emergent skills developed by the agents that were completely off-script. Through playing a simple game of hide and seek hundreds of millions of times, two opposing teams of AI agents developed complex hiding and seeking strategies that involved tool use and collaboration. The OpenAI experiment is an example of how the theory of evolution can could play a role in AI systems. Simply fascinating.

Now let’s take a look at the core developments in AI research and technology this week:

AI Research

Better speech using self-supervision

Facebook AI Research(FAIR) unveiled three new research initiatives for building more robust speech recognition systems.

>Read more in this blog post from the FAIR team

AI agents learn behaviors they were not trained for

OpenAI trained group of AI agents on the game of hide-and-seek, and they end up developing unique and unexpected behaviors.

>Read more in this blog post from the OpenAI team

Deepfakes learn human preferences

Also from OpenAI, they unveiled a new version of its controversial GPT-2 algorithm that can generate fake texts that are indistinguishable from the real ones.

>Read more in this blog post from the OpenAI team

Cool AI Tech Releases

Training AI with AI

Facebook open sourced the code for question-answering model that can be trained by generating its own question-answering dataset.

>Read more in this blog post from the FAIR team

Learning from the human brain to process temporary information

Google unveiled Project Ihmehimmeli, a new neural network architecture that allow to process temporal information in order to build better knowledge.

>Read more in this blog post from Google Research

AI in the Real World

Deep learning has recreated a lost Picasso in full color

A neural network has recreated a lost painting from Picasso’s Blue Period.

>Read more in this report from MIT Technology Review

AI for Earthquake Prediction

A team of AI researchers is testing AI models for predicting earthquakes in the Pacific Northwest

>Read more in this article from Quanta Magazine

AI analyzes the safety of dietary supplements

Supp.ai is a new project unveiled by the the Allen Institute for Artificial Intelligence that can analyze the interactions and conflicts in the components of dietary supplements.

>Read more in this article from TechCrunch

Written by

CEO of IntoTheBlock, Chief Scientist at Invector Labs, Guest lecturer at Columbia University, Angel Investor, Author, Speaker.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store