DeepMind, part of Google, announces General AI breakthrough with a true learning AI. First, they built a dynamic environment (like a game) that can change it’s own layout — XLand. Then, they use Deep Learning and Reinforcement Learning combined — Deep Reinforcement Learning — to create an AI the can learn without training at all or data about what it’s doing. The AI played 700,000 games in 4,000 unique worlds! The AI performed 200 BILLION training steps while performing 3.4 million UNIQUE (non-taught/programmed) tasks.

Support The Show On Patreon:
https://www.patreon.com/theaiguide.

Follow The AI Guide on Twitter:
http://www.twitter.com/DavidRa20284933

Like The AI Guide on Facebook:
http://www.facebook.com/DavidTheAIGuide.

Follow The AI Guide on Instagram:
https://www.instagram.com/theaiguide/

Follow The AI Guide on LinkedIn:

Humans find AI to be a frustrating teammate when playing a cooperative game together, posing challenges for “teaming intelligence,” study shows.

When it comes to games such as chess or Go, artificial intelligence (AI) programs have far surpassed the best players in the world. These “superhuman” AIs are unmatched competitors, but perhaps harder than competing against humans is collaborating with them. Can the same technology get along with people?

In a new study, MIT Lincoln Laboratory researchers sought to find out how well humans could play the cooperative card game Hanabi with an advanced AI model trained to excel at playing with teammates it has never met before. In single-blind experiments, participants played two series of the game: one with the AI agent as their teammate, and the other with a rule-based agent, a bot manually programmed to play in a predefined way.

The results surprised the researchers. Not only were the scores no better with the AI teammate than with the rule-based agent, but humans consistently hated playing with their AI teammate. They found it to be unpredictable, unreliable, and untrustworthy, and felt negatively even when the team scored well. A paper detailing this study has been accepted to the 2021 Conference on Neural Information Processing Systems (NeurIPS).

2030’ish: on demand books, movies, TV shows, video games, etc…

---

In the ever-increasing list of things that machine learning AI can do in our modern world, there’s now a program that will code (or at least, try to code) whatever you tell it to in plain English. Want some flashy banner text that changes color every few seconds? Tell that to OpenAI Codex and it will code it for you in seconds. The OpenAI Codex beta, currently only available through an online waiting list, is a simple web tool with three windows: one to type in commands, one that shows the code generated by those commands, and one that shows what the code does. You could theoretically use Codex for all sorts of tasks in over a dozen coding languages, but the coolest use I’ve seen is coding simple Javascript videogames with just a handful of natural language instructions. Check out the video below from YouTuber Joy of Curiosity to see it in action.