Sneaky robots taught to deceive each other

I’ve long had an interest in the field of Artificial Intelligence (AI)… the area of computing science dedicated to making machines act, well, intelligently. In fact, I specialised at university in AI and harboured thoughts of becoming an AI researcher afterwards. Things took a different direction, but I’m still fascinated by the study of AI and found this post from Gizmag really interesting: Robots taught to deceive.

The robots in question were programmed to play hide and seek, with the “seeker” tracking its prey down by taking note of markers that it knocked over while looking for a hiding place. The thing is, the “hider” knows this, so it would knock markers over that made it look like it was going to hide in one place, but then actually hide somewhere else. Sneaky stuff.

This brings back memories of projects at university. I ended up writing a system that helped companies automatically source supplies from various locations, but briefly toyed with the idea of writing a sort of virtual wargame, with two “robots” trying to outwit each other in a simulated environment. Whichever triumphed would pass its knowledge on to the next generation of robot, hopefully making each successive generation better at the game. This deceptive behaviour is just the sort of thing that would have been handy there!

What are the applications for deceptive AIs? The ones that immediately come to mind are military in nature: using deception to infiltrate an area for surveillance or attack, and avoiding capture. The guys at Gizmag also talk about the robots having to “deceive” panicky humans in a search and rescue situation. I guess it could also be used in a ┬áleisure capacity… anyone for paintball with autonomous drones in the mix?

This idea of deceptive behaviour is very, very clever and makes me wish I’d kept more up to speed on developments in the AI community. If you want to read the research paper that came out of this study, a 22-page document by Alan Wagner and Ronald Arkin, you can read it here.

What do you think of robots being taught to be “deceptive”? Clever? Scary? Does it have any useful applications, or is it just scientists being smart for the sake of it? I’d love to know your thoughts on this story – why not share them in the comments?

Join us on Facebook

Facebook icon

Declare your geekdom for the world to see... well, the part of the world that's on Facebook anyway.

Visit our Facebook page to keep up to date with the latest Geek-Speak posts right in your Facebook stream, as well as hearing about discounts and offers before they're posted on the site.

What are you waiting for? Head on over and "like" us.

Facebook icon used under CC license


  1. I have no idea how that happened, but I’ve just noticed the comments were disabled for this post. Sorted now.

Comment are closed: Sorry, comments are closed on this article. We automatically close comments on older articles to try and cut down on the amount of spam comments being submitted to the site.

If you want to tell us what you think about this article, why not visit our Facebook Page or Subreddit and leave a comment there instead?