Elon Musk’s AI learned to play Minecraft

3 mins read
Elon Musk's AI learned to play Minecraft

OpenAI’s new AI learned to play Minecraft by watching 70,000 hours of YouTube videos. But experts warn.

Elon Musk's AI learned to play Minecraft

OpenAI, the artificial intelligence company where Elon Musk sits on the board of directors, has been surprising people for the past years. The company, which previously launched two artificial intelligences named GPT-3 and Dall-E 2, has now announced that it has developed a new neural network that can play Minecraft.

Artificial intelligence started playing Minecraft by watching YouTube videos

In a new blog post by OpenAI, it was stated that the neural network training was completed with a successful Minecraft experiment. According to the company, the algorithm dropped into Minecraft learned basic skills such as chopping trees, collecting wood and setting tables. It also performed functions such as swimming, hunting and cooking, depending on the situation the character was in.

After 70,000 hours of training, the neural network watched the experiences of other players through photos and videos. As a result, it was found to be successful in performing complex skills from hunting animals to consuming their meat. According to the researchers, this is the first time in history that an artificial intelligence has gone beyond “drag and drop” and caught up to the level of other players.

Researchers at OpenAI said the neural network also fulfills long-term experiences. For example, it successfully produced the completed diamond pickaxe, with 24,000 actions performed by players in about 20 minutes.

OpenAI utilized a process called Video PreTraining (VPT) in training the neural network. After 20,000 hours of watching videos, the AI was finally able to perform similar experiences as humans.

Even though we are only experimenting in Minecraft, the game (mouse and keyboard) is very open-ended and produces very general results. We therefore believe that our results bode well for other similar domains, for example, computer use

While the neural network has learned to play Minecraft with VPT, the researchers point out some risks. Security experts restrict access to algorithms like GPT-3 and DALL-E, in part to limit misuse.

What do you think about this issue? Don’t forget to share your opinions with us in the comments!


The ancient idea tries to provide the most accurate information to its readers in all the content it publishes.

Fatal error: Uncaught TypeError: fclose(): Argument #1 ($stream) must be of type resource, bool given in /home/fikrikadim/public_html/wp-content/plugins/wp-super-cache/wp-cache-phase2.php:2381 Stack trace: #0 /home/fikrikadim/public_html/wp-content/plugins/wp-super-cache/wp-cache-phase2.php(2381): fclose(false) #1 /home/fikrikadim/public_html/wp-content/plugins/wp-super-cache/wp-cache-phase2.php(2141): wp_cache_get_ob('<!DOCTYPE html>...') #2 [internal function]: wp_cache_ob_callback('<!DOCTYPE html>...', 9) #3 /home/fikrikadim/public_html/wp-includes/functions.php(5420): ob_end_flush() #4 /home/fikrikadim/public_html/wp-includes/class-wp-hook.php(324): wp_ob_end_flush_all('') #5 /home/fikrikadim/public_html/wp-includes/class-wp-hook.php(348): WP_Hook->apply_filters('', Array) #6 /home/fikrikadim/public_html/wp-includes/plugin.php(517): WP_Hook->do_action(Array) #7 /home/fikrikadim/public_html/wp-includes/load.php(1270): do_action('shutdown') #8 [internal function]: shutdown_action_hook() #9 {main} thrown in /home/fikrikadim/public_html/wp-content/plugins/wp-super-cache/wp-cache-phase2.php on line 2381