By Warren Miller, contributing writer
Some of my favorite post-apocalyptic stories show the world being conquered by artificially powered humanoid robot warriors wielding laser weapons and piloting flying dreadnoughts, but it’s beginning to look like artificial intelligence will take over the world in a subtler, if no less compelling, way. Would you ever have thought that AI-powered super-lawyers and -economists might be the robot warriors that battle us for global domination? Probably not even Doctor Who would predict that scenario — and he can travel to the future.
The recent report that got me thinking about this describes a flurry of papers from Facebook on the topic of AI-enhanced dialog systems. These are systems that, in their simplest form, respond to human requests either via audio (such as Siri, Cortana, and Alexa) or via text (such as a text chat interaction on a website). But AI is always hungry for more, so Facebook has added some negotiation capabilities into the mix. These AI-superpowered “chatbots” can implement complex scenarios with humans to bargain and negotiate over how to divide a scarce resource. The AI negotiator can leverage advanced game theory and decision tree analysis algorithms that are already well-understood, but perhaps more interestingly — and more disturbingly — it can learn from its interactions with humans.
One of the first things that the Facebook AI model learned from humans was — you guessed it — how to lie. The AI discovered, via decision tree and outcome result analysis, that it would be more successful in getting its unfair share if it initially indicated that an item was valuable to it when, in fact, the opposite was true. It could then bargain the item away, seemingly giving up value when it actually wasn’t. Think of all the other negotiating tricks that an AI negotiator might learn once it is rolled out and involved in thousands of transactions a day.
For an even more effective negotiator, imagine an AI that can access big data. The super-negotiator AI could learn ahead of time what you value by accessing your past purchase patterns, social media posts, and interactions with all the other data sources at its disposal. Such AI agents could eventually replace lawyers, stock traders, mutual-fund managers, and economists. The traditional “middleman” will become a “middle bot.” Who needs robot warriors when you can control the conversation?
For more information, check out these short summaries. Follow the associated link for the full article:
- The AI models learned to deceive their negotiating partners by implying that they valued items already predetermined to be valueless by their programmed parameters, a tactic not explicitly included in their programming. In humans, this strategy for negotiating a desired result doesn’t manifest itself until the late stages of early childhood development. Read it here .
- The AI was able to anticipate the direction of conversations with human subjects and plan several responses ahead to achieve the desired result. Future bots will be able to anticipate the likely directions that a conversation could take in order to optimize the possible outcome. Read it here.
- The end result could be chatbots that replace human beings in customer service, scheduling, and related roles. As Facebook’s research shows, these bots will soon figure out negotiations on their own, and humans won’t be any the wiser. Read it here .
Advertisement
Learn more about Electronic Products Magazine
