Friday Weird Science: This bartending paper is not about you

Sep 27 2013 Published by under Friday Weird Science

I have a problem. It's a silly one to have.

I suffer complete and total bar invisibility. I am not exactly someone who fades into the background, but to a bartender, I might as well be an empty stool. I have gone as long as 45 minutes without having a single server come near me, even the ones I truly attempt to make eye contact with. Not only that, but I have gone 45 minutes without being served in a NEARLY EMPTY BAR. Sometimes I wonder if I just look wrong, or my body language is wrong. I try to imitate the people around me, hanging out at the bar, making eye contact. Nothing.

Bar invisibility. My dad says it's genetic, apparently he has it, too.

So you can only imagine how thrilled I was when I received a link to a video showing that SCIENCE has found how to best gain attention at a bar!

I was thrilled, gleeful! Finally I would know what I was doing wrong and bartenders would never ignore me again. I tweeted the link far and wide.

And then I received this reply from Thomas Williams.

And he's right! And much of the coverage is WRONG. Thank you to Thomas for pointing me in the right direction! He helpfully linked me to the BBC coverage (where the scientist corrected the interviewer), and then I got my hands on the paper. He pointed me, and now I can point you!

And the best part of this paper? It's not about PEOPLE AT ALL! It's not even really about how to get attention in bars! This paper? It's not about you. We just all wanted it to be.

Loth et al. "Automatic detection of service initiation signals used in bars" Frontiers in Psychology, 2013.

Who was this paper about? It was about a robot. Specifically, it's about getting a robot to successfully socially interact with humans. Not flirt with one or make deep philosophical conversation or anything difficult. Instead, it was just a basic human interaction.

But this becomes complicated because humans communicate a lot without ever saying anything. We are past masters at non-verbal communication. So when you want to train a robot to interact with a human, you have to teach it nonverbal signals.

Here we run into another snag. In many cases, we don't KNOW the nonverbal signals. We just perform them, learning from our peers as we go. So when you want to teach a robot nonverbal cues, you first have to figure out what those cues are. Which are necessary? How many are sufficient?

That was the question for this paper. The authors were training a robot to tend bar (bartenders of the world: don't panic. So far he just grabs and moves things, and can't make a Cosmo to save his robotic butt). In order to do this, they have to determine the nonverbal cues involved in getting the attention of a bartender.

So how DO most people get the attention of a bartender? To figure this out, the authors took video of a bunch of people in bars. They then looked at the frequency of various actions displayed by the customers. This is what they got.

Above you can see the frequency of actions that occurred. The two MOST COMMON actions were "body to bar" (walking up to the bar), and "looking at bartender" (making eye contact). Note, this was just the frequency of the interactions. Not whether or not they worked. These actions occurred in a sequence. Belly up to the bar, and make eye contact.

But now we have to figure out whether these either of these two cues are necessary or sufficient. Can you just belly up to the bar? Can you just make eye contact?

To do this, the authors grabbed undergrads and gave them photos. In each of the photos there were signals present.

You can see the signals above. For each of them the participants were asked to respond "serve" or "don't serve". At top left we have a "don't serve", he's at the bar, but he's not looking at the bartender. At top right we have another "don't serve", she's looking at the bartender, but standing BACK. Bottom left, we have a "serve", up to the bar, and looking. And bottom right we have an "accident", up to the bar, and looking...but not ordering.

The participants were predictable in their rates. The top two don't get served, the bottom two do, even though the bottom right is a mistake. The researchers concluded that the nonverbal cues are "come up to the bar", and "make eye contact". They could then code this into the robot in a nice sequence, and the robot could serve drinks! Sure the robot will get it wrong sometimes, but it's better to get it wrong by asking someone to order who doesn't than by ignoring someone who does.

But it's important to note: this is not the most effective way to get attention at a bar. It is merely the most common. The most frequent cues displayed. This may in fact mean that it is the most effective, because it's what bartenders are looking for, but the authors did not test for effectiveness. They were just looking at frequency. What it is that most people do when they want drinks. And they weren't doing it for the edification of humans. They were doing it for the robots.

So it looks like my bar invisibility will persist for now. 🙁 But maybe someday they will address effectiveness. And then we will know!

5 responses so far

Leave a Reply