We have a new, high-definition video camcorder, which will allow bigger and better video footage for The Ugly Guppy performance and for The Bubble Blog. Feast your eyes:

I upped the hoop behavior's difficulty to a chain: Follow my finger -> stop and check in at front of tank -> receive iPod cue as reinforcement for check-in -> perform 3 passes through the hoop in response to this one flash of the cue.

And she did it! Almost every time! Awesome!

I have also come to a better understanding of Mango's view of The Rules of Engagement. She does a few reps, then she wants a "coffee break" to putter around the gravel, just in case something yummy fell down there. I'm not fighting her on this. Substrate foraging is deeply innate in goldfish. If I tried to get rid of this behavior during training, everything else might break down. This is an example of what Bob and Marion Bailey called "The Misbehavior of Organisms": some behaviors are so instinctive that they can't be trained out of the animal. So I let Mango have her 5-10 second coffee breaks, then I ask her to check in and re-engage in training. What I might eventually try is actually timing the coffee break, then gradually shortening its duration.

But for now, I'm just happy with my little fishie's golden performance!

Mango is now reliably doing a triple pass through the hoop, cued by a blue circle projected on an iPod. During past training sessions I have presented the cue at the beginning of each hoop pass, to help Mango associate the blue circle with hoop swimming. Now she has improved to one cue flash per triple pass through the hoop. This video demonstrates her progress.

You Minnesotans will get the reference, but others might not. When we see something adorable here in the Frozen North, we say "Oh, fer cute!"

I've been working to add cues to Mango's hoop behavior and Ricky's piano playing/spins. Initially I thought a flash from a colored penlight would work well as a cue, with different colors to cue different behaviors. That's not a bad system, but I thought it could be improved. Since I use a flash from a white penlight as the "click" to mark rewardable behavior, sometimes I feel the fish become confused by a second flash, even though it has color. While the confusion could probably be overcome with many, many training sessions and repetitions, I decided to search for a better cueing system. What I wanted was the potential to create dozens of distinct cues for different behaviors. And I wanted the cue to be very different from the click.

I have read astracts from several studies where goldfish are taught shape and color discrimination; the test fish could tell a red square from a blue circle, for example. A system of colored shapes for cues would be infinitely versatile. But how to get my fish to notice such shapes? And how to be sure the fish can see the cues?

iPod to the rescue! I am now using my iPod to flash a JPEG of a colored shape for the cue. At first I tried doing this on my Powerbook, thinking that a large picture would be more salient. But that was too unwieldy; I couldn't cover/uncover the screen fast enough to time cueing correctly. The iPod works beautifully. It's small and handheld, thus easy to flash quickly. The shape picture is backlit, which makes it more noticeable to the fish. (I do have to be careful not to get the iPod wet.)

I tried the iPod cueing system with Ricky, and it really didn't work. He's fixated on receiving his cue on the piano keyboard or (for Spin) on the back wall of his tank. His responsiveness to training is variable now, too, because he's still suffering from bacterial enteritis. He generally does pretty well with a blue or red penlight flash, though. So for now I'll stick with what he knows.

But Mango Smoothie is really catching on to the iPod system. My criteria for rewarding the hoop behavior is currently 3+ passes through the hoop per rep. To start, I let her make one pass through. Then just as she begins each of the remaining hoop passes, I flash an iPod pic of a blue circle. (I only flash the iPod when Mango is facing front, otherwise she can't see it.) She does look at the iPod. We're not quite at the stage of flashing the iPod BEFORE the hoop pass, but I do believe Mango is already associating that iPod blue circle with hoopage. I expect in another week or so we'll have a bona fide iPod cue.

I suppose next she'll demand a little cell phone. ;-)

Video of iPod training coming up...

My ultimate goal is to have Mango swim through, about-face and swim back through again, then repeat that sequence 2-3 times before expecting a reward. In this video sequence I first reinforce a couple of single hoop passes, then raise the criterion to two passes. Mango offers the about-face, and I snatch the opportunity! We still have a ways to go to cement the about-face pass and move on to several passes per reward, but we're on the way!

When training fish, I have always held the feeding wand in my lap until it's time to reinforce (reward) a behavior. I also keep my hand on the white penlight all the time except for loading the feeding wand. But I had the feeling that Mango was watching my body language really closely and using something I was doing as the "treat coming" signal, rather than the white light flash.

Realizing this reminded me of my Chicken Camp experience. One of the hens I worked with was very hesitant to walk across a wood bar between two platforms. This was part of an obstacle course we were training our chickens to traverse. Lucy would trot right through the first part of the course and walk halfway across the bar. But instead of going all the way to the end of the bar, she came to a dead stop in the middle and just stared at me. I had really tried to be a quiet trainer and not let my body language give her the false impression that I was about to feed her (thus ending the behavior). But our Teaching Assistant, after watching my frustrating the-cluck-stops-here scenario, said "Are you aware that your right shoulder muscle twitches slightly when that hen gets to the middle of the bar?" And she was right. At that point I was THINKING of the reinforcement (feed in a cup) that I was going to present when the hen made it across the bar. Lucy observed this minute muscle twitch, thought "Oh! Treat coming! Done working!" and just stopped. Once I started freezing like a stone statue during training, the hen made it through the course. Wow.

My goldfish may not be the detail savants that chickens are, but they do watch me closely. I think they are very capable of receiving a false cue from a twitch of my hand or a change of facial expression. So now I, once again, freeze like a stone statue during training. I particularly don't make eye contact until the fish has done the required behavior. I cultivate stillness. I try to focus on the moment and not anticipate what I'll do five seconds in the fture. After only one session of this stricter control of my body language, Mango's hoop trick is exponentially faster, cleaner, and more reliable.

So let go of desire, Grasshopper. Observe the Zen of fish training!

For the first three sessions she was convinced that targeting (touching) the hoop was the goal. Targeting will probably always be a default behavior for her. This is partially, perhaps, because she targets the feeding wand every time I feed her. So she automatically gets more practice with "target an object" than she does with any other behavior. I have also observed, with dogs and other species, that the first really solid behavior they learn sometimes becomes default; they often offer it when they're not sure what the trainer wants. The first trained behavior an animal learns probably makes a big impression; that's the behavior that elicits the first "Aha! I get it!" moment, when the animal figures out the system: behavior -> click -> treat. This is a big moment in the training life of an animal, so it's not surprising that it becomes deeply entrenched in the animal's memory.

In spite of Mango's fondness for targeting the feeding wand, I've mostly been shaping her to swim through her hoop instead of luring her all the way through with the wand. She's a wiggly, impatient fish. If I rely too much on the wand as a lure, she gets very caught up in swimming up and down the length of the wand trying to get the treat out. This feeding frenzy sometimes seems to obliterate her ability to associate the hoop with the treat. In operant conditioning, the animal's behavior needs to be a clear predictor of the reinforcement (treat) to follow. If the feeding wand itself becomes the strongest predictor of reward, the training sequence falls apart, and I end up reinforcing the wand shimmy instead of the hoop swim.

To correct this problem, I stopped using the wand as a continuous lure (pulling it from one side of the hoop through to the other). Instead I clicked for her head touching the hoop, then quickly put the wand in the tank and fed her in front of the hoop. This prompted her to swim through, since swimming through was the fastest way to get to the treat. In this scenario, the wand is technically still a lure. But it's a more passive lure that "feeds for position" at the end of the behavior instead of encouraging fixation on the wand through the whole behavioral sequence.

The brilliant advantage of the feeding wand is that it can be such an alluring target. Lots of behaviors are based on targeting. Every animal is different, too; the wand was  perfect for initially luring Ricky into a spin. He's more relaxed about training, so the presence of the wand wasn't much of a problem. And, again, the wand makes it possible to quickly and accurately "feed for position." That alone makes it a spectacular training tool.

So how is Mango doing now with her hoop training? Quite well. In her fourth session she began swimming through the hoop during most reps. She's not fluent in the behavior yet, so not ready to have a cue attached. But I expect to reach that point soon.

I'm also still asking for a few repetitions of Check In (stop at front of tank and look for cue light) at the beginning of each training session. Eventually I hope this will be her default behavior. Right now I don't reinforce Check In during reps aimed at hoop swimming. Once Mango is fluent in Hoop and the cue has been added, I will put Check In and Hoop together as a behavioral chain. Then the light cue for Hoop will serve as the reinforcement for Check In.

Video of Mango's hoop trick coming soon!