So. We've taught our bird a marker signal—the click, or a blink of a light, or both. We've used that signal to tell the bird what behavior we like—silence instead of screaming, for example. We've used it to shape new behavior, such as jumping to the fist or a perch. We have a bird now that doesn't scream, doesn't mantle, doesn't bate off the glove; a bird that accepts the hood; a bird that's as fit and strong as a wild bird, since, using Steve Layman's brilliant shaping programs, we can exercise it in the mews.
Because the bird understands the conditioned reinforcer, it is attentive to the falconer and comes to the lure reliably. Because the bird has learned several ways in which it can earn food, it feels in control of its world, so it is calm and ready to work. Are we pleased? Sure. Are we satisfied? Of course not!
"Tell me this," said the Old Falconer, rocking back on his heels a little. "I want my bird to come back to me when she's too far away; but I want her to come back AND to stay up in the air. If I give her your click signal for coming back, she's got to come down to me to get her reward, right? So how can I do that and still get her to stay overhead?" And he glared at me with "Gotcha!" in his eyes.
How are you doing it now? I wanted to ask. But I thought I knew. I would expect that some birds figure it out, and some never do. For most people, it would depend on lots and lots of trials, perseverance, a "good" bird, and a certain amount of blind luck. A skilled falconer might develop habits of his own, utilized during the first few hunts with each new bird, that serve to reinforce staying at altitude—matters of timing, such as when to swing the lure, when to stop; or when to beckon the bird back, and when to turn into a post and seem to ignore the bird. These nuances are bred of experience and intuition, the ability to "read" the bird—skills that are hard to verbalize and hard to pass on to the rest of us.
Operant conditioning gives us another way to look at the problem. As the pioneer in clicker training, Marian Breland Bailey, Ph.D., puts it, with operant conditioning you can train any animal to do any thing it is physically and mentally capable of doing. Is the bird physically capable of coming back without coming down? Certainly. Then all we need is a way to tell it what we want it to do.
To do this, we need to understand more about cues. A cue is different from a command. We give our dogs commands all the time: Come here. Lie down. Get in the truck. Shut up. A command is a signal to do something, but it is also a threat: Come here Or Else. Or else I will make you do it. Or else I will reprimand you. If the command is not heeded, we often escalate the threat aspect: we yell the command, for example. This often works with dogs. Sometimes it works with people. But we all know it NEVER works with birds.
That doesn't mean, however, that you can't tell a bird what to do; you just do it with cues, instead of commands. We don't normally teach cues methodically; animals pick them up. You decide to go to the store, so you make sure you've got your car keys. The dog hears the keys jingle, and runs to the front door. That sound has become the cue for a behavior: "Get to the door fast, and I might get to go for a ride." A cue is like a green light: it's a window of opportunity to do the right thing and get reinforced for it. Here's the cue: my fist in position. Step onto my glove, and you'll get a click and a treat. The cue need not be so obvious: a whistle or a gesture that's nowhere near the perch can tell the bird "Get on the perch and I will click you."
The big difference between a cue and a command is that while a command is an implied threat, a cue is a positive signal: a cue is an opportunity for reinforcement. When the cue is well-learned, it becomes a sort of "on" switch for action. The animal WANTS to do the behavior the cue indicates, because it has paid off handsomely now and then in the past. It doesn't stop to ponder, "Yes, but do I want a tidbit more than I want to sit on this telephone pole?" It just responds. "Oh, he said 'Come!' Good, here I come!" So, once you have established a cue, you don't need to escalate it, to get results. When the light turns green, we step on the gas immediately; two green lights, or a bigger green light, would not make us step on the gas any faster or better. When the phone rings, we answer it, even though sometimes the caller is a roofing salesman and sometimes we were doing something we really enjoyed more than phone calls when the phone rang. It's a cue, cues mean a chance to get something good, and we respond.
We shape the understanding of a cue exactly the way we shape a behavior: in small steps. First, you develop the behavior. Before the bird can learn to care about the cue, it needs to know that this particular behavior can earn reinforcement (to review the procedure of shaping behavior, see my previous article in Hawk Chalk for April, 1997).
Don't forget to use your marker signal—the clicker or the blinker (the same bird can understand both) to identify the behavior. Handing over the food may please the bird and make you feel good, but the food is not information; from the bird's point of view food might arrive by pure chance. The MARKER signal, however, is very exciting information. It tells the bird, "Hey! You just did something that Works!" To the bird, hearing or seeing the marker is sort of like flushing the rabbit: it's not the same as eating the rabbit, but in a way it is better. The swoop that flushed the prey is almost certainly something the bird will try again. So, the marker identifies the behavior: but it also tells the bird that it just did something important and useful. It makes the bird eager to learn more.
Let me describe how you might practice teaching your first deliberate cue. Pick some simple behavior, such as lifting a foot. Get the bird lifting its foot two or three times for a single click. Do this over several five minute sessions, once or twice a day, until the behavior is well-established and the bird starts lifting its foot when you come near. Then choose a hand signal as your cue: perhaps just pointing. Point at the foot, and click and treat the next lift. Put your hand down, let one try go by unclicked, point again, and click the next lift. Alternate between paying for cued behavior, and ignoring uncued behavior. Click even weak responses to the cue, and ignore even the strongest response when your hand is down. Work fast: five or six cues and five or six uncued responses in one or two minutes: then quit or do something else, something easier. Learning cues is hard work for a bird, at first; don't push it.
Sometimes, especially when teaching your first cue, you may see a flurry of off-cue behavior. It seems as if the bird is catching on; it's lifting its foot nicely on cue; but then it suddenly stops lifting its foot when you point, and instead lifts it madly when you put your hand down. This is called an extinction burst: "This behavior USED to work, now it doesn't work, dammit I'm going to MAKE it work!" Be patient; on the other side of the extinction burst, the bird will understand the cue at last.
You can help the bird learn a new cue in several ways. You can wait to give your cue until the bird has quit lifting its foot: then point. Then wait until the bird can stand still for two or three seconds, or five, or ten, before you point. Thus you are reinforcing "not lifting for longer and longer periods," with the signal to Lift. You are actually teaching the bird to wait for this cue.
Another helpful device might be to turn toward the bird when you point; reinforce the lifted foot; then turn away from the bird when you put your hand down. The bird may already know that you are more apt to click when you are looking at it than when you are looking away. And, finally, when you have established one or two cues that are well understood, and to which the bird responds reliably, you can go back and forth between the new cue and an old one. Lift your foot. Now jump to the perch. Lift your foot. Now come to my fist. Alternating behaviors and cues can sometimes build comprehension of a new cue without any extinction burst.
Will Shor, editor of Hawk Chalk, points out that if you are a newcomer to operant conditioning, you can develop your skills by practicing with cats. Cats learn very much the way a hawk or falcon does. They don't respond well to commands and threats (that's why people say they're untrainable) but they will work for clicks and treats, and they learn cues exactly in the same manner as a bird does. If you don't want to risk confusing a bird that's already working well for you, try this out with cats; then you'll have mastered the system when you get your next bird.
So, you now have a bird (or a cat) that lifts its foot on cue. What on earth does this have to do with a hunting bird coming back to the falconer without coming down?
When a bird has learned a few cues in hand, it has learned something more: it has learned to understand the nature of a cue. Here are some things you could say the bird has learned:
- A cue is the go-ahead for me to do something special, something specific.
- A cue is good news! I'd better follow up on it immediately.
- You have to do the right thing on the right cue, or it won't work.
- Hand movements can be cues.
- Cues are good. Watch for cues. Learn more cues.
NOW you're going to fly the bird. And all you need to teach it is two new cues: one for "come back" and one for "go up." Won't that be easy? You have a HUGE head start, since the bird already knows items 1-5, and you have already learned how to shape cues. If you made mistakes, while you learned, it didn't matter a bit. You had plenty of chances to correct them, because you were working safely at home, with trivial behaviors that you don't really need anyway, such as lifting the foot (or bowing, or nodding, or picking up sticks or...suggestions welcome. KP)
To teach flying cues, the bird has to be flying, so now you'll be working outdoors. You're going to teach the two cues separately, at first. And yes, you ARE going to reinforce the behavior with your blinker, and you ARE going to let the bird come down and get its treat, every time, just at first. Not for jillions of repetitions, just three or four times per hunt for the first few trips, taking advantage of whatever transpires that day.
Let's arbitrarily pick arm signals as our cues, and let's make them as different as we can. I'm going to choose, for "Come back," rotating my left arm in a sweeping smooth circles over my head, as if swinging the lure—a big, smooth movement; and, for "Climb," jerking my right arm backward from the elbow, over my head, as if tossing pebbles one by one into the sky. A sharp, short movement.
Now, when I see that the bird is climbing, I'm going to give my "Climb" cue and, at some point before she levels off, using my judgment, I'm going to click/blink her. The click means treat, so without further ado, down she comes, right? I don't have to call her in. Heck, have I ruined a nice climb? Not at all: I have reinforced climbing in the presence of the climb cue. Once. But once is a good start.
Remember, as we saw in the previous article, that you can reinforce also reinforce aerial behavior by flushing game. You don't have to blink and bring the bird down every time. But find any opportunities you can to cue and then blink and treat climbing, in the course of the day. Perhaps as you are getting ready to go home, for example, you can give your climb signal, blink the bird during the climb even if it is just going from ten feet over your head to fifteen feet, and treat on the return. The magnitude of the behavior—how high the bird gets—is not what you are worried about, at this stage. What you need to concentrate on is establishing a cue that means "Go up." You can teach that at any altitude.
When the bird sees your arm signal and reliably starts to climb, then you can shape for altitude, by withholding your blink until she's climbed a given amount; 50 ft, say, for one reinforcement; maybe 100 feet, another time; again, how fast to raise your standards is a judgment call, but your criterion now is not just "climb" but "climb further." When you have lengthened the response in this way, you can start using your signal in combination with the day's hunting. She's at 200 feet: you signal "Go up," and she does, and when she gets to 400 feet, bingo, blink her and flush those ducks. These birds are not dumb; they can learn from a single experience like that, especially one with a huge "jackpot" reinforcement such as fleeing mallards. Now you have a bird who, when you cue "Climb," will do it, and gladly. (We could stop right there, with the ability to send your bird higher at will, and some falconers would be pretty pleased.)
But we're also working on the "Come back" cue. Here again, just take advantage of any chances that crop up during the day, to give the cue and reinforce the correct response. If the bird is already conditioned to the lure, give your new signal first. Circle your arm, and when the bird heads your way and gets close, swing the lure. The lure is both a cue and a treat, so it isn't as clear as the arm signal; the lure is like free food. When you offer it, the food is already there, and the bird knows that, even if it hasn't tasted the pigeon leg yet. There are two disadvantages to this: the bird CAN weigh the prospect of dead meat on the lure against its present circumstances on top of the phone pole and decide "I don't want the lure." Secondly, swinging the lure may actually reinforce a behavior you don't want, such as sitting on the top of a phone pole for several hours. I would urge you to establish the arm signal, an arbitrary cue with its subtly irresistible powers (like the ringing phone) and using the lure AFTER your marker signal, as a primary reinforcer, rather than relying totally on the lure and baiting the bird in.
I'm not going to get into the problem, here, of the bird that habitually refuses to come back; very often that is a behavior that has accidentally been reinforced by the falconer's behavior, anyway. Let's just concentrate on using the natural events of the day as opportunities for reinforcement. It should be possible to give your new cue and blink the bird (and be sure to treat it) for heading towards you or getting back to you (when to blink? A judgment call) three or four times during the course of the day; perhaps more. At this point it doesn't matter that the bird is coming in AND coming down to get its treat. We are teaching it the come-in cue; nothing more.
Don't waste your cue: if you circle your arm, and the bird comes your way and then veers off, drop your arm and stand still. No cues = no treats, right? The bird likes cues. If you just keep cueing it because you HOPE the bird will "obey" it will learn that the cue means nothing. If you cue it as it heads in, but stop cueing at once if it heads some other way, you are attaching a small but real penalty to veering off. Thus, with your cue alone, you can shape behavior, by telling the bird what works and what doesn't! And what works, of course, is "Go toward the falconer when you see the circling arm."
And you maybe you STILL don't see what good all this cue work is doing. Here's the payoff. There will come a day—maybe the third hunt, maybe the fifth, maybe the tenth—when you feel comfortable about the cues; when you feel the bird understands that the circling arm means "Come back" and the tossing arm means "Go up;" when you feel confident that you will get a good response to each cue. And now the wind has come up, or the bird had a little visit from a local wild bird, and it's flying too far away. But you just got started, and you don't want to bring it down to you.
So you give both cues at once. Left arm circling, right arm tossing. And the bird comes back, and climbs at the same time. You don't even need to reinforce it.
But you might give yourself a pat on the back: nice training.
Author's note: The formal name for this procedure is "adduction." Behavioral psychologists have now developed ways of using adduction in the classroom. Marine mammal trainers have made use of it for years. You train Behavior 'A' and put it on cue, then train Behavior 'B' and put it on cue, then give both cues to get behavior 'C' on the first try, without any further shaping or reinforcement.
The falconer's "come back-stay up" problem lends itself beautifully to this elegant solution. Adduction takes some preparation, but depending on one's needs, the results should be worth the effort. Not to mention spectacular. If you are among those who try it out, I'd love to hear from you: karen [at] clickertraining [dot] com.