Brainy bots

Relax! Chatbots didn't take over at FB, they just needed to reset priorities

chatbot-ai The AI software is working efficiently to the priorities that the bots were set—that is, 'perform with maximum efficiency' | File

The brainy bots need to be reprogrammed so that they don't get too uppity

Imagine a father teaching his kid how to drive a tricycle, and a few months later, the smart kid starts driving the father's Ferrari, and even figures out driving modes the father had little idea about.

Take another example. A few years back, the plot of one of Rajanikanth's blockbusters revolved around a robot going haywire just by following human orders. If programmed right, the robot could, well, kill its creator. This plot had quite a deeper meaning, one which we have somehow overlooked. If our Ferrari kid was far brighter than the previous generation, the robots are far ahead.

Chatbots, one of the favourite buzzwords in technology for 2017, are an interesting amalgam of human conversation and artificial intelligence. Simply put, you can 'programme' a bot to do what you like and the bot will keep learning from experiences. The chatbot is a combo of two compelling technologies: artificial intelligence (AI) and simple human conversations. Mix them up, stir well—and you can serve a potent new tool that mobile users are lapping up wherever they find it. For starters, you don't need to download a fat app.

Amazon's Alexa, Apple's Siri , Samsung Bixby and Google Assistant are all chatbots.

Sounds cool? Or so, everyone thought. From ordering pizzas to emergency evacuations to customer care services, bots supposedly had the ability to perform these activities as efficiently as humans. Facebook is working on an advanced chatbot for Messenger that tries to do better—going beyond rules and set scenarios to displaying initiatives—this shouldn't be a problem.

However, the problem is that these bots somehow figured out a new language to converse among themselves—a language that was unknown to humans. What's worse is that humans initiated this language but somehow the bots evolved the language much beyond human comprehension. Besides being a complete facepalm moment for humans, experts are wondering if there are limits to what they can do with AI.

Researchers at the Facebook AI Research Lab (FAIR), including Indian-origin Devi Parikh and Dhruv Batra of Georgia Institute of Technology, are an elite group of people who were trying to 'teach' the bots negotiation and conversational skills. One fine morning, they realised that the bots had become efficient negotiators and communicators, but between them they stopped speaking in English. Instead, they had devised their own language that was beyond anyone's comprehension. Worse, the researchers believed that "things could soon go out of hand”.

“The new language that the bots used was more efficient for communication between the bots. Agents will sometimes drift off understandable language and invent codewords for themselves. if I say 'the' five times, you interpret that to mean I want five copies of this item,” Batra clarifies. (see: bots transcript)

fb-chatbots

Parikh and Batra have co-authored a paper 'Deal or No Deal? End-to-End Learning for Negotiation Dialogues' on how they are training the chatbots at Facebook to do complex negotiations. The paper was presented at a Chinese AI conference on June 16.

However, the researchers assure that the AI software is not taking over. Rather, it is working efficiently to the priorities that the bots were set—that is, ' perform with maximum efficiency'. They were rewarded for it.

But they were not rewarded for sticking to plain English. So, they worked to maximum efficiency and took short cuts which involved letter sequences that were not plain English. If the bots were programmed to accord priority to 'converse' in English, they will do so.

So relax! The machines are not taking over, contrary to what popular media claim. At Facebook, they just need to be reprogrammed so that they don't get too uppity.

On one hand, experts see huge potential in developing chatbots, and on the other, there are massive threats lurking around, pretty much like Hollywood's Independence Day, when aliens (almost) took over the earth. According to Durgesh Kaushik, former Facebook marketeer and currently, the CEO of video-sharing app Wishfie, this is not the first time that AI Bots have developed their own language. “In fact, the AI Google uses for its translate tool also created its own language, which Google allowed to continue. Personally, I find it quite fascinating and hope that we can leverage AI to understand human behaviour better,” he says.

Kaushik adds: "Currently, these bots communicate only using text; it will be even more interesting when AI bots start to communicate verbally (which is closer to the way human language has evolved). However, we need to have a very tight control on how AI evolves in future, never giving the control away! At Wishfie, we plan to leverage AI to moderate live debates on the platform. It'll be interesting if those bots develop biases like today's TV anchors!"

Pretty much like most things in the world, seems like chatbots too are good, as long as we don't use them in excess or give too much control away.  

This browser settings will not support to add bookmarks programmatically. Please press Ctrl+D or change settings to bookmark this page.

Related Reading