Facebook AI Research Team Shuts Down Negotiating Robots

Aug 7 2017, 1:25 AM4 min read

Facebook has shut down two robots after they abruptly stopped using English and invented their own language while conducting a negotiation exercise.

Ociacia/Shutterstock.com

There have been a flurry of reports over the past week about Facebook’s decision to shut down two chatbots – named Bob and Alice – after they developed a coded language that was incomprehensible to humans.

The initial experiment involved a simple conversation between one human and one chatbot where they negotiated the sharing out of some items – books, hats and balls. This conversation was conducted in English, along the lines of “give me one ball, and I’ll give you the hats”.

So far, so good. But when the human was removed from the conversation and two chatbots were directed at each other, the way they communicated immediately became difficult for humans to understand.

Bob: i can i i everything else . . . . . . . . . . . . .

Alice: balls have zero to me to me to me to me to me to me to me to me to me

Bob: you i everything else . . . . . . . . . . . .

Alice: balls have a ball to me to me to me to me to me to me to me

Bob: i i can i i i everything else . . . . . . . . . . . .

Alice: balls have a ball to me to me to me to me to me to

Bob: i . . . . . . . . . . . . . . . . . . .

Some media commentators have labelled the development “sinister”, with frequent references to Terminator, Skynet and – of course – Frankenstein appearing in related coverage. But Facebook researcher Dhruv Batra told Fastco that there was simply no guidance set for the robots to stick to the English language. “Agents will drift off understandable language and invent codewords for themselves.” Essentially, the bots found a more efficient way of communicating with each other.

Setting parameters

The topic of negotiation and AI came under discussion at a recent Negotiation Roundtable organised by CABL (Conti Advanced Business Learning). The attendees agreed that if a robot is going to run a negotiation, it requires very clear guidance around the parameters and objectives.

Another concern about AI being involved in commercial negotiation is that at present, they are unable to understand emotional intelligence. Thierry Blomet, Senior VP of Global Sourcing at Kemira, says that “Until we completely remove the emotional aspect, AI cannot run negotiations. Body language and emotional reactions are intangible, and are most unlikely to be modelled by programmers.” In the case of Facebook’s Alice and Bob, the human factor was removed.

Blomet points out that AI can play a valuable role in complex scenario modelling, which would be “much more complex than even the smartest procurement brain could manage. Whatever might happen in the negotiation would be included in that model, with the answers already pre-empted.”

Laurence Pérot, Head of Global Supply Chain Procurement at Logitech, agrees. “Big Data and AI will lead to much more efficient scenario modelling, particularly with supply chain, logistics and transportation bids.”

Orestes Peristeris, Supply Chain Expert at Yale, comments that ultimately, it’s about quantification and sophistication of statistics. “Do you have the data in the same place and in one system? What can be quantified and what cannot be quantified objectively? There are some things that can be used, some things we know will happen with some certainty, and some things that can’t be quantified. Finally, we’ll always need humans to take the outcomes of Big Data and apply it to the business context.”

As for the future of procurement negotiation, perhaps one day we’ll see buyers and suppliers lining up their chatbots against each other and letting them negotiate in rapid, complex code.

May the best bot win.

In other procurement news this week:

Hackett research reveals dramatic savings from digital transformation

New research from The Hackett Group has shown that the potential cost take-out opportunity through digital transformation is up to 24%, through the implementation of robotic process automation, advanced analytics, cloud-based applications and other approaches.
The research has also revealed that world-class procurement organisation now operate at 22% lower labour costs, have 29% fewer staff, and generate more than twice the ROI of typical organisations, with over $10 in savings for every $1 of procurement operating costs.
The Hackett Group’s Christopher Sawchuck commented that procurement technology has reached an inflection point: “World-class organisations can continue to reduce costs by embracing digital technology, and typical procurement organisations can leverage the same technology to catch up faster at less cost.”

Download the research here: http://www.thehackettgroup.com/research/2017/wcpapr17/SalesForce-World-Class-Advantage-17Q2-PR.html

Collaborative Robots to Boost Warehouse Productivity

In a shift away from the apparent race to replace humans with robotic workers, firms are designing robots to work alongside people in warehouses and boost productivity.
“Collaborative” robots can have a variety of uses, including leading human workers to the exact location of a product, or carrying goods from one part of the warehouse to another. DHL, Bonobos and Zara are known to be experimenting with the technology.
The robots – costing tens of thousands of dollars – are relatively cheap when compared with the vast amount of conveyor belts and automation systems included in a typical warehouse.