Hey Alexa…..Bring Me A Beer

Howard "Bart" Freidman
Rule the Robots
Published in
5 min readJun 26, 2019

“Uzing Microzoft Aritifical Intelligence, vee vill be able to detect new flavorz in a zplit zecond….”

Like IBM’s flaming logo e-business ad in the dot com bubble, Microsoft's “AI unlocks the art of science” ad is a flare signalling the start of the race to mainstream AI. It took three years for the AI-powered beer angle to percolate from TechCrunch to TBS — “AI Beer” hit the scene in 2016 with IntelligentX, a UK startup selling a monthly beer box subscription. IntelligentX solicits subscriber taste feedback — AI analyzes the data to drive iterative recipe modifications. With “beer fingerprinting,” Carlsberg is taking a different approach — at least initially — a sort of human genome project for beer.

Where there’s AI…there’s beer.

A human-powered riff on the same ideas — incrementally modifying beer taste toward consumer preference — is responsible for the bland, generic flavor profile of mass produced American brews. A GWU professor, who led Miller’s 1980’s project to combat Budweiser’s ever-increasing market share, gave me a first-hand account. After profiling Budweiser’s taste via chemistry and focus groups, the number two brewer found that beer drinkers readily identified Miller High Life by its distinctive sweetness. In response, they spent five years dialing it down, along with other incremental recipe changes. In the end, Miller’s flagship beer tasted just like Bud — which was precisely their goal.

Hopefully AI can do better.

Elsewhere on Microsoft’s AI front, they deactivated their Zo chatbot*, the successor to their Tay bot, itself shut down for spewing out inflammatory tweets. When I tested Zo last year, like Tay and other general purpose chatbots, it wasn’t very good. I revisited Zo in April to see how it improved, but before I could complete my 6-month checkup, Zo was gone.

With no mo Zo, I shifted to Mitsuku, said to be the worlds most convincing chatbot, having won the Loebner Prize Turing Test four times running. Asked what “she” can do, Mitsuku replied: “If you tell me a date, I can tell you what happened on that date in history and also what was the UK’s number one hit single on any day since 1952. If you tell me when your birthday is, I can tell you some facts about your birthday.”

Missed it by that much….

Or not…..

Chatbots like Mitsuku are prone to embarrassing mistakes, so they’re yet to find widespread use. On the other hand, now that speech recognition is stunningly good, and seeing as talking is easier than typing, Digital Assistants (DAs) like Siri and Alexa have become ubiquitous. Unlike chatbots, DAs don’t need to maintain conversational context, a task humans find simple but computers vexingly hard. Still, after years of trial and (often comical) error, chatbots are pretty common too, although for limited applications. Because of comprehension issues like Mitsuku’s, thus-far, chatbots are replacing web contact and information-request forms. In this role, they’ve proven to improve conversion, albeit only about as much as more user-friendly, multi-part forms.

People expected more. Gartner in particular.

In 2011, Gartner predicted: “By 2020, customers will manage 85% of their relationship with the enterprise without interacting with a human.” Five years later, with that a seeming pipe-dream, they doubled down: “By 2020, the average person will have more conversations with bots than with their spouse.” Given their wildly ambitious predictions, it’s not surprising that Garner has bots toppling over the Peak of Inflated Expectations and into the Trough of Disillusionment.

Regardless of expectations, chatbots are already quite capable of fooling humans in mundane roles like answering TV ad call-ins. Each week, thousands of people hopeful of winning a fortune in fabulous prizes talk to Vocinity CCAs (Cognitive Conversational Agents). Listening to recordings, it’s pretty apparent they either don’t know, or don’t care.

Even so, for mass adoption, the role of conversational AI, including chatbots, remains TBD because of evolving behavior.

The best — maybe the only? — real, direct measure of “innovation” is change in human behavior.” Slack CEO, Stewart Butterfield

Enterprise chatbots like Vocinity CCAs handle conversations typical of a low-skill call center agent more efficiently and effectively, with less effort for the caller than web-based self-service. The question is whether that matters. Humans naturally communicate via speech, but we changed long ago to accommodate computers. Now that computers finally understand speech, will we want to change back? Behavior has changed so thoroughly, it’s hard to even imagine conversations that once took place millions of times a day. Before the web, before IVR’s, and before Domino’s Pizza, did people loathe calling the corner pizzeria?

Since those days are long gone, we’ll never know. We do know, at least for now, conversational commerce is still a tiny niche. With a smidgen of setup, yelling “Hey Alexa….order me a pepperoni pizza” will, in fact, result in a pizza showing up at your door. So far though, Alexa isn’t ordering many.

In the Enterprise, DAs are catching on, slowly beginning to suppress the tyranny of tiny tasks imposed on most of us since word processors began filling5 in the secretarial pool. As of now, the tasks entrusted to DAs are more steno pool than executive assistant.

That’s changing as I type, which we’ll dive into next time.

* What exactly constitutes artificial intelligence? According to the Chinese Room argument, chatbots are “weak AI,” which some say isn’t artificial intelligence at all.

Whatever……

--

--

Howard "Bart" Freidman
Rule the Robots

Revenue accelerator: distributes growth hockey stick. Futurist & pastist. Loved by both Rick and Morty.