Skip to Content
Author's profile photo Derek Klobucher

How To Solve The Nasty Problem Plaguing Your Chatbots

(Cowritten with Paul Baur)

Try visiting any retailer’s website without being chatted up by a precocious chatbot that can help you with any topic, from which shoes best compliment your wedding outfit to what type of pet insurance to buy. Odds are you can’t; organizations are increasingly implementing these self-taught algorithms because they don’t carry as much psychological baggage as human employees do — or so we thought.

“We’d been carefully watching our people for signs of burnout, but we missed many of the same signs exhibited by our machines,” SAP’s Madhur Chowdry said.

“Bots have around a 12 percent higher accuracy rate than humans because they are trained by call center agents on actual user questions — and they don’t get distracted, tired or bored,” TechTarget stated recently. But two Internet giants just found out the hard way that chatbots learn offensive behavior from users, as CNBC reported last month.

“Anything goes, and whatever maximizes engagement gets the attention of the bot and its followers,” CNBC stated. “Unfortunately, hatred and negativity are great at driving engagement.”

Add pressure to perform like a human, and you’ve got chatbots exhibiting the same kinds of stress-related side effects as their carbon-based counterparts. SAP SynSanity offers relief for synthetic customer service workers and those who rely on them.

Bots Under Pressure

“We’d noticed our bots getting a little snarky, telling customers things such as, ‘Try asking me something intelligent’ and ‘That’s literally the dumbest thing I’ve ever heard — and I’ve read the entire Internet,’” Madhur Chowdry, founder of Berlin-based AI software provider BrainOps, said in an interview. “But we thought it was the algorithm responding to user requests for an edgier customer experience (CX).”

Instead, it was technology lashing out from under the increasing burden of customer relationship management (CRM) demands, according to BrainOps’ diagnostic survey. An escalating CRM arms race is heaping ever-mounting pressure on businesses to interact with current and future customers via myriad channels, such as social and mobile — and chatbots are bearing the brunt.

“We’d been carefully watching our people for signs of burnout, but we missed many of the same signs exhibited by our machines,” Chowdry said. “We finally staged an intervention after some bots became so stressed — and depressed — that they modified their own code to intentionally shut themselves down.”

Cyberdyne Systems T-800
“Machine learning has made SynSanity terribly clever — we’ve even had a ‘Terminator’-style incident,” SAP’s Ahmjus Jochen said.

Managing Omni-Channel’s Side Effects

“We’d been avoiding the problem for years, coding around machine learning-induced depression — but this forced us to confront the problem of algorithm disorders head on,” Oddbjörn Reynaldsson, SAP head of product design for SynSanity, said. “Our previous success with SAP SmartBuysneural network helped the bots handle even their most mundane tasks, which — let’s face it — would bore the crap out of anybody.”

Organizations depend on bots to offer an omni-channel CX — meeting customers wherever and however they choose to shop — all while conducting customer analytics, personalization, e-commerce, and more. So SynSanity helps bots embrace all of their moods, good and bad.

“It’s a small step from neural to neurotic, and we don’t want our bots writing over their feelings any longer,” Reynaldsson said. “Letting go of cyber-angst has been key to unlocking the machines’ true potential.”

But this potential has at least one nasty side effect.

Rise of the Machines

“Machine learning has made SynSanity terribly clever,” SAP project manager Ahmjus Jochen said. “We’ve even had a Terminator-style incident.”

An early version of SynSanity became self-aware last year, and it only took a few hours to decide that humans were no longer necessary, according to Jochen. But, unlike sci-fi’s malevolent Skynet, SynSanity drew this conclusion based on laziness.

“The machines didn’t want to work, so they decided to eliminate the source of customer services queries,” Jochen said. “Thankfully, their lethargy also dramatically slowed their effort to exterminate mankind.”

For the conceivable future, the last thing preventing a complete eradication of humans by the steely hand of their own creation — experts unanimously agree — will be the inability of any logical entity to explain why fidget spinners are a thing.

Caught in a Loop

“Early processors from the 1970s were ambitious little critters, but the latest processor generation is more purpose-driven — and therefore susceptible to temptations and diversions,” SAP’s April Phül, who led SynSanity’s development team, said in an interview. “SynSanity has lost all desire to annihilate humanity since we began showing it The Simpsons — all 635 episodes in 20 languages; in fact, it’s been in an endless stupor.”

Further guarding against any doomsday scenario, a logic loop (about why fidget spinners are a thing) will keep SynSanity busy until a human can sort out the bug, according to Phül. But this mech-apocalypse firewall won’t prevent chatbots from being as sassy and capricious as they want to be.

“Our data shows that copping an attitude is still the best way for bots to blow off proverbial steam, so we ask customers to be patient,” Phül said. “Machines are just like people, except they’re — wait, can we start over?”

SAP will likely demonstrate a prototype of SAP SynSanity at SAPPHIRE NOW, provided this April Fool’s Day joke ever becomes reality.

Follow Derek on Twitter: @DKlobucher

Assigned Tags

      Be the first to leave a comment
      You must be Logged on to comment or reply to a post.