Facebook AI project halted after bots invent new language

facebook ai negotiation

Facebook has closed down two bots as part of a project teaching AI negotiation skills. The chatbots developed a new shorthand and began to communicate in sentences that researchers could not understand. 

Researchers at Facebook Artificial Intelligence Research (FAIR) are training AI bots to negotiate. The technology could eventually be used in a number of ways, both online and in the real world.

Negotiation is, on the face of it, a simple skill: I’ll give you this on condition of you giving me that. However, whether you’re picking a TV channel or talking down a trader at the market, these interactions require complex communication and reasoning skills, as well as an awareness of context.

As the FAIR team says in a blog post, current chatbots are capable of providing general information and performing simple tasks such as making reservations. The step towards a more meaningful conversation will require negotiation skills that take external information and aims into consideration.

The Facebook AI negotiation project

The stated purpose of Facebook’s AI negotiation research is to prove that it’s “possible for dialog agents with differing goals (implemented as end-to-end-trained neural networks) to engage in start-to-finish negotiations with other bots or people while arriving at common decisions or outcomes.”

Read more: With machine learning, Control F1 recognizes drivers within minutes

facebook ai shutdown negotiation training

Facebook’s AI negotiation platform

However, in the process of testing, the researchers found that the two AIs had started to deviate from standard English. Instead, they were attempting to carry out a negotiation – deciding how to split a selection of ordinary household objects – in a form of shorthand:

  1. Bob: i can i i everything else . . . . . . . . . . . . . .
  2. Alice: balls have zero to me to me to me to me to me to me to me to me to
  3. Bob: you i everything else . . . . . . . . . . . . . .
  4. Alice: balls have a ball to me to me to me to me to me to me to me
  5. Bob: i i can i i i everything else . . . . . . . . . . . . . .
  6. Alice: balls have a ball to me to me to me to me to me to me to me
  7. Bob: i . . . . . . . . . . . . . . . . . . .
  8. Alice: balls have zero to me to me to me to me to me to me to me to me to
  9. Bob: you i i i i i everything else . . . . . . . . . . . . . .
  10. Alice: balls have 0 to me to me to me to me to me to me to me to me to
  11. Bob: you i i i everything else . . . . . . . . . . . . . .
  12. Alice: balls have zero to me to me to me to me to me to me to me to me to

It’s easy to read this interaction as the start of something sinister. We’ve all seen enough dystopian movies in which artificial intelligence forces its will upon its human creators. However, this is simply a negotiation for some balls – even if it did take an unexpected turn.

“Agents will drift off understandable language and invent codewords for themselves,” FAIR visiting researcher Dhruv Batra said. “Like if I say ‘the’ five times, you interpret that to mean I want five copies of this item. This isn’t so different from the way communities of humans create shorthands.”

The conversation was halted, but not because researchers feared they had empowered a nefarious interaction. Instead, as FAIR researcher Mike Lewis told FastCo, the team decided that their “interest was having bots who could talk to people.”

Communicating efficiently with each other is all well and good, but a customer facing support bot needs to be able to write in ways that anyone can understand.

Read more: British scientists use artificial intelligence to predict outcome of legal trials

Facebook AI research raises questions

Facebook’s research into AI and the aptitude its bots have for developing new languages raises interesting questions. Namely, should computers be left to their own devices to communicate as they wish?

The fear is that humans would lose an element of control over devices chattering away in their own language. But in reality, it could be that it results in improvements in how well chatbots understand the complexity of our communication and intentions. An algorithm able to develop and learn languages could be capable of churning through complex data, such as a human conversation, more effectively.

Read more: Morrisons uses artificial intelligence to stock its stores and drive sales

The post Facebook AI project halted after bots invent new language appeared first on Internet of Business.

Internet of Business

Ignore the hype machine — It’s back to basics with bots


I occasionally meet with a company that worries they’re not ready for bots. They think of bots as an emerging technology fraught with complexity. In some cases, they worry they don’t have the skillset internally to deploy and get ROI out of the technology. Others want the technology to “mature” a bit more before they place a bet.

If you really stop to think about it, it’s a ridiculous question: “are we ready for bots?”

Do you have customers and employees? Do they use computers and smartphones? Does your business depend on IT? If you said yes to those three questions, then you’re probably bot-ready.

If you stop thinking about bots the way the tech industry talks about them, or how AI is often depicted in movies, you’ll find that you have everything you need to be successful in today’s world and not some far -off tech utopia.

It’s not just about tech, it’s about experience

The tech industry would have you think that bots are fundamentally changing the way we interact with the Web. Some argue that bots mark the end of the era of mobile apps. I argue that your business shouldn’t care about what bots mean to the tech world. Focus on what they mean for your customer’s experience with your brand.

It’s common sense, really. Your time should be spent on your core business, not the trends in the technology industry. Bots are simply a more natural and authentic way to connect with your customers, employees and various other stakeholders. It’s a better customer experience — whether for making a purchase or getting support.

Temkin Group’s 2016 study said moderate improvement to customer experiences generate an average revenue increase of $ 823 million over three years for a company with $ 1 billion in annual revenues. They also found that emotion was the most influential factor in the design of a customer experience.

Think about that for a second — according to the Harvard Business Review, emotion is the new battleground for customer loyalty and repeat purchases. You can try to build emotion into your website, or maybe your shopping cart can use a heavy dose of empathic design. But the best way to connect with consumers is through conversation. If the experience isn’t conversational, you can’t come close to delivering an emotionally resonant customer experience. For this reason, the bot delivery medium is the ultimate interface for human emotion whether it’s text or voice.

If you already obsess over your brand experience and connecting with your customers, then you’re ready for bots. With the right “brains” behind the scenes powering bots, they’re the ideal delivery mechanism for the experience you’ve worked so hard to create in an authentic yet approachable manner.

Endpoints and paradigm shifts

Thirty years ago customer interactions primarily happened in-store and on the phone. Twenty years ago we shifted towards email. Ten years ago we shifted towards live chat. Today, we’re using bots to connect with customers on any messaging platform everywhere.

IRL, phone, email, chat, bots––they’re all just delivering an experience at different endpoints. While the tech world perpetually bemoans “digital transformation” failures, I remain impressed at how even the most non-technical businesses have evolved as customers demanded new forms of engagement at new endpoints.

Most businesses that last the test of time do a good job of meeting their customers needs, especially as they change. Every business cares deeply about meeting those customer needs where they are. If nothing else, that’s the fundamental function of a business.

The tech industry wants you to believe these are paradigm shifts, no matter how big or small. Software may be eating the world, but the fundamentals of your business have not changed. The tech industry’s paradigm shift is often just a new endpoint that presents an opportunity to connect with and serve your customers. Used responsibly, bots are a remarkable tool to help you do just that.

Has it already happened?

Customers already made the leap to messaging. According to Mary Meeker’s Internet Trends report, chat and social media are already the top channel for customer support among Gen Y. Meeker’s report also points out that over 60 million businesses have active accounts on WeChat and Facebook, showing that businesses are making the move now, as well.


Are you ready for bots? If you’re just worried about buying the latest technology, maybe not. But if your goal is to better connect with your customers and grow your business, and you’ve got the foundation in place, then you’re as ready as you’ll ever be. Is there ever a good reason to hold off when your customers are waiting for you?

The post Ignore the hype machine — It’s back to basics with bots appeared first on ReadWrite.


Why are there no good bots?

Broken Bot

Why are there no good bots? This is a question I get asked on a regular basis. Probably on average of about twice a week.

What are you defining as “good”? If you’re defining “good” as some magical creation that enables you to have a conversation with a machine, and makes it feel like you are totally engaged. Then yes absolutely, I’m yet to find a “good” bot. But honestly, why do you want to chat with a machine? You should probably get out more.

I suspect the reason behind this was all of the hype around the term “chatbot.” Personally, I hate that term, and you won’t ever hear me use it to describe these “things,” or “bots” as I call them.

Technology should be used to solve a problem

So for a bot to be successful, it needs to have a purpose, work out what you are trying to achieve from a bot, what are you trying to do?

Instead of asking are there any “good” bots, you should be asking are there any “successful” bots. This is an easy question answer, yes there are absolute loads of examples of successful bots. But for something to be defined a success, it needs to have objectives.

A major musician with a loyal fan base was receiving hundreds of messages a day via Messenger, that were being ignored, now they get responses. No, it can’t answer every question, and no it’s not AI, but you shouldn’t be surprised at how often the same question is asked by different people. Let’s say that’s 100 different people every day, that’s over 35,000 messages a year that are now getting responses. That’s a lot of missed engagement and frustrated people.

As any customer support representative will tell you, out of 100 customer support requests, 80 of them will be a mix of ten of the sam questions/problems. That means that 80% of the responses can easily be defined in a process.

Over 35,000 missed opportunities a year are no longer missed

A North American retailer wanted to increase incremental revenue for every online transaction, improve efficiency, drive higher self-service rates, and provide consistently delightful and engaging experiences across all channels. Using a bot over a six- month period, the retailer kept 55 percent of contacts within the bot. Customers could solve their problems without speaking with a human agent or making a phone call. This hugely reduced their need for resources saving them $ 1.6 million in contact center costs.

$ 1.6 million saved in contact center costs

Compare Hardwell’s vote on his track of the week for his podcast, fans used to be able to vote on his track of the week on his website, now they can do it via his Messenger account, and he gets roughly 100x the number of votes he used to get each week when it was purely his website. On top of the main purpose, fans can now engage with him in different ways. They can ask when he’s next on tour, they can find out about his latest music, and they can even send in shout outs for his weekly podcast. That’s a huge definition of success.

10,000% increase in weekly actions performed via Messenger compared to directly on their website

Or another customer of ours, they are driving huge amounts of reach, they are now using Messenger in the way it should be. It’s a messaging channel. It’s enabling them to message their customers, and their customers to be able to message them in a frictionless 1-1 way. With far, far better results than email. They are seeing 99% read rates and on average 21% click through rates. Let’s put these numbers into perspective. That means that for 1,000 users engaged via Facebook Messenger, you need roughly 30,000 users signed up to an email mailing list to get the same return.

For every 1,000 Messenger users, you need 30,000 email subscribers for the same results

The power of these bots is easy to misunderstand when you don’t have a bot yourself to monitor and analyze usage. One of the main reasons for this is the exact same reason why these things work well. They are private communications. External parties can’t listen in on the conversations or the usage of the bots. Which means from the outside, you have no idea how successful they are. It’s “dark social,” it’s 1-1 communications. This is one of the many powerful things about these bots.

The next time you think “Why are there no good bots?” ask yourself what are you really defining as “good.”

Bots LandscapeThis article is part of our bots landscape series. You can download a high resolution version of the landscape featuring 197 companies here.

The post Why are there no good bots? appeared first on ReadWrite.


How Bots Can Help You Find The Right Gift This Year

Bots are a great way for holiday shoppers to check off items from their wish lists while avoiding crowded stores and endless hold times on the phone. Bots are also a fantastic tool for the organizations that use them. But they might not be ready to help you find that perfect gift (yet).

Internet bots execute routine online tasks, usually those either so tedious that they would bore human workers half to death, or on such a massive scale that people couldn’t finish the job within their lifetime. But like people, bots learn — and some even have personalities.

“Many bots are programmed to act like humans when you talk to them,” as CNET noted, “so it feels like asking a person for help instead of just typing in a search engine.”

Made in your image

U.S. cosmetics maker CoverGirl is targeting teens with a chatbot created in the image — and diction — of a fellow teen: American television personality Kalani Hilliker. In true machine-learning form, KalaniBot takes cues from how Hilliker communicates on social media in order to interact with customers in a similar fashion.

The business benefit for CoverGirl?

“The bot’s detailed analytics deliver feedback on engagement levels,” AdAge stated last week, “providing reporting metrics across average conversation length; sentiment analysis; semantic analysis; branded messages; bot mentions; response rate; and purchase links.”

In short, bots offer a trove of data and insight, which is information that organizations can’t get when shoppers interact with human associates. And that’s a big reason online shoppers are starting to see bots now — and why we can expect to see more bots while shopping next Christmas season.

“Bots represent an opportunity for the industry to close the conversion loop by tracking clicks to purchases,” Justin Rezvani, founder and CEO of KalaniBot co-creator theAmplify, told AdAge. “This means more transparency and accountability for how well your strategies and content perform.”

Your virtual shopping buddy

Bots could improve the customer experience, especially for younger shoppers who would not to ring a vendor’s 800 number. One software solution can accomplish its missions more consistently than multiple call center operators,* according to The Drum.

“Bots live within chat, which is a highly social environment where friends talk and share,” The Drum stated last week. “[So] it’s not too much of a stretch to think about consumers eventually asking bots what to buy their friends.”

Asking a bot to help you shop for others usually involves answering conversational questions, rather than a battery of multiple-choice queries, as a Technical.ly reporter recently discovered when trying to find a Christmas gift for his brother. The bot asked Tyler Woods about his brother’s age, occupation and other standard questions before delving into more specifics, such as passions and hobbies.

“I replied that he likes sports and politics and works in healthcare; that he dresses well and runs and seems ambitious,” Woods wrote last week. “I didn’t put in that he can be a troll on Twitter and IRL, but I could have.”

Bots offer something for everyone, including shoppers, retailers and their workers.

Bots offer something for everyone, including shoppers, retailers and their workers.

Ready for prime time?

A bunch of retailers already offer chatbots to help you shop for flowers and apparel this holiday season, as The Drum noted. But your experience might not always be flawless, as technology reporter Chris Griffith discovered while searching for green socks with the help of a bot.

“It launched a Web page … offering to sell a $ 15 U-Lock bike lock,” Griffith stated in The Australian last month. “Some coding still needs to be done!”

So while we wait for retailers to sort out the bugs, shoppers can use bots as part of their strategy to find this year’s best Christmas deals, according to WSJ Personal Tech columnist Geoffrey A. Fowler. He recommends bots that offer price history and price alerts for specific products.

“Saving money is all about timing,” Fowler said in The Wall Street Journal last month. “And a few good bots will help you know when to strike.”

Happy hunting

Part of knowing when to strike is knowing which bots to trust. They aren’t perfect, but their potential to offer easy shopping for customers — and real-time insights for retailers — is too great to pass up.

Many bots already deliver a happy bargain hunting experience to shoppers this year. And this technology promises many more bots — and an even better customer experience — next year.

This story originally appeared on SAP’s Business Trends. Follow Derek on Twitter: @DKlobucher

*This technology might free those call center operators to take on more complex tasks. It could even free the business to grow in new ways, as I’ve noted before.

Internet of Things – Digitalist Magazine