AI bots develop their own non-human language
While developing negotiating chatbot agents, Facebook researchers found that the bots spontaneously developed their own non-human language as they improved their techniques, highlighting how little we still know about how artificial intelligences learn.
At one point, the researchers write, they had to tweak one of their models because otherwise the bot-to-bot conversation “led to divergence from human language as the agents developed their own language for negotiating.”
In other words, the model that allowed two bots to have a conversation—and use machine learning to constantly iterate strategies for that conversation along the way—led to those bots communicating in their own non-human language.
The larger point of the report is that bots can be pretty decent negotiators—they even use strategies like feigning interest in something valueless, so that it can later appear to “compromise” by conceding it.
Already, there’s a good deal of guesswork involved in machine learning research, which often involves feeding a neural net a huge pile of data then examining the output to try to understand how the machine thinks. But the fact that machines will make up their own non-human ways of conversing is an astonishing reminder of just how little we know, even when people are the ones designing these systems.
Discover, Learn and Evaluate AI Companies and Solutions
The latest updates from AI companies in your industry
Get a weekly newsletter with the latest posts directly from the AI companies. Follow companies to tailor your feed.
Experience the leading AI demand forecasting platform.
Get your image labeled efficiently and cost-effectively.
SPELL - HYPERPARAMETER SEARCHES
Optimize hyperparameters to improve the accuracy of your model with the Spell Hyper command
SPELL - JUPYTER NOTEBOOK SERVER - GPU POWERED
Collaborative Jupyter Notebook or JupyterLab workspace server with powerful GPUs
SPELL - DEEP LEARNING PLATFORM
The fastest and most powerful end-to-end platform for machine learning and deep learning.