A good thing? A bad thing? A nonexistent thing?
Let me tell you what I mean by "jackpot." I mean exactly what the casinos mean: a surprisingly big reinforcer, delivered contingently. The key is in the word, contingent. To reinforce a particular behavior, a jackpot has to appear, and be perceived by your learner, while the learner is doing that particular thing you want. Not afterwards.
If you click, and then deliver the treat afterwards, an especially large, numerous, or wonderful treat is no different from any other treat, in terms of its ability to reinforce behavior. Good treats, big treats, may make the clicker even more desirable and worth working for; but those differentiations don't specifically strengthen one behavior over another. It's the click that does that. Good treats or big treats may make the learner more interested in the training and work in a general way; but they're not specifically informative. The association is Pavlovian. Click means treat is coming. If the treat is sometimes a kibble and sometimes chicken, sometimes small and sometimes huge, that's fine, it keeps your clicker nice and strong; but it doesn't tell the animal anything different about the behavior. So that's not what I mean by a jackpot.
People often assume that a jackpot is any unusually large reward. An obedience competitor tells me, "I use jackpots; we go through the ring work, and then back at the crate I have six or eight wonderful treats waiting, and he knows that ahead of time." Well, maybe he does, and maybe he doesn't; hard to say what a dog "knows." And what would you say those treats reinforce: the work in the ring? Getting back to the crate, would be my guess. That's not a jackpot. That's just a windfall, an unusually large treat, associated with nothing in particular. Nice, butâ€¦.
A parrot owner tells me, "I use jackpots. When he's done something especially good I make a big fuss and give him six pieces of fruit, one at a time, instead of the usual one piece." Same deal. What does that tell the parrot? "Sometimes my person gives me more fruit than other times." Nice, butâ€¦.<shrug>
Several people have posted that they tested the "jackpot" theory, usually with a setup such as this: You divide treats up into small and large amounts, and then reward one behavior with the click followed by small amounts, and another behavior with the click followed by large amounts. Usually there's no discernable difference between the results. Both behaviors are maintained at whatever level you're clicking for. Ergo, jackpots don't work?
No, because that's not a jackpot. To me, a jackpot only a jackpot if it's delivered in such a way that it functions as an event marker, identifying the wonderful act as it is happening, and also functions as an unusual primary reinforcer, making that behavior more likely to happen again.
A real jackpot should be such a surprise, and happen so suddenly, that it is actually startling. Think about the casino slot machine jackpot. Ever won one? You're playing away, shoving in quarters, pulling the handle, getting close to a win now and then, or getting two or three quarters back which you feed into the machine again, and just when you're thinking of quitting and getting something to eat, you pull the handle, watch the things spinning around, and WHAMMO! Bells ring, the machine lights up and flashes (those are clicks, of course) and with a rattle and roar a hundred quarters pour out of the machine. What a surprise!
What were you doing at the time? Watching the wheels in the machine spinning around the way you made them do by pulling the handle. What behavior was reinforced? Watching the wheels roll around while hoping to win.
To repeat this experience you have to get more quarters, and pull the handle again to make the wheels spin so you can watch them in hopes of another jackpot. It's a little behavior chain. To maintain the chain, only the last behavior, watching the wheels, needs to be reinforced, and then only very sporadically (you're on a long variable ratio schedule, natch, which can be addicting). Watch the people in a casino intently staring at the screens of the slot machines they are "playing," hour after hour after hourâ€¦
OK, now imagine if what happens is, you get the wheels spinning, and they come to a halt, and there is a sign saying, you win. But you get your winnings non-contingently. When you go back to your room, you discover 100 quarters on your bed.
The amount is the same. But what is it actually reinforcing? Maybe you feel you'd like to come back to this hotel again. Maybe you'll drop in on your room more often. But you may not feel at all like going right back and making the slot machine work some more. You might decide that spending the money in the gift shop down the street would be more fun. Believe me, casinos never make that mistake. Their jackpots are not only extremely noticeable (to you and everyone else) they are absolutely contingent on what the casino would like you to keep doing: Putting your money in slot machines so you can watch those wheels go around.
So, in working with an animal, I would never use larger quantities in the hopes of maintaining behavior that's already trained. I would only use an unusually large quantity as a true jackpot, that is, delivered all at once and contingently, to both mark and strongly reinforce the first occurrence of a rare behavior, or the first achieving of a difficult move.
If I wanted to give a dolphin a jackpot the first time it swam through a really scary hoop, I'd blow the whistle and upend the fish bucket simultaneously: it's raining fish! I have given my dogs jackpots for coming when called from long distances for the first time.
In Don't Shoot the Dog (pp. 11-13, revised edition) I wrote about jackpots, giving what I still think is a good example of the jackpot: the tactic of the horse trainer I mentioned on page 11. He worked with American Saddlebreds. These horses are taught a somewhat unnatural four-beat gait called the "rack." They are bred to do it, but they also may have to be helped by the rider at first. When working with a young horse it was this trainer's habit, when the colt "first hit a lick of the rack," to jump off him immediately, strip off the saddle and bridle, and turn him loose in the ring. Now that is a jackpot. It's negative reinforcement, freeing the colt from the demands of the rider, but reinforcement, all the same; and the fact that it was a completely new and thus memorable event makes it memorable. I think it facilitates also the remembering of the brand-new movement that made it happen.
Remember, though, that no one is always right. We all keep learning things all the time, revising and adding to what we knew before. That's one of the joys of science and a valuable phenomenon in the clicker training world.
Looking back at the jackpot section in Don't Shoot the Dog, now, twenty years after it was written and six years after the revision supervised by Murray Newman, I think that I failed to differentiate between jackpots as I see them, and another tool altogether: the non-contingent reward.
A non-contingent reward is also something you get by surprise, but it is not associated with any particular behavior. One example in the book was the two free fish we gave to a discouraged dolphin, which perked her up and set her to trying to earn reinforcement again. Another example in the book was the ticket for ten free riding lessons that my parents bought me when, at sixteen, I was behaving poorly for weeks on end. It instantly corrected my bad mood. I included these as jackpots, but they were not; they were both examples of a non-contingent reward. The most powerful use of a non-contingent reward is to counteract the effects of an extinction curve; I know the dolphin in question was undergoing extinction of a bunch of operant behaviors; probably this sulky teenager was, too. Getting the news that good things are still available revived the efforts to seek reinforcement again.
Like the jackpot, a non-contingent reward is a tool to use rarely. And, like a jackpot, if it is going to work, you only need to do it once.