Do rats have values?

Neural correlates, computation and behavioural impact of decision confidence.
http://www.ncbi.nlm.nih.gov/pubmed/18690210

I would like to start off this post with a disclaimer: in no way am I claiming that rat cognition is on the same level with humans. I merely suggest that as we learn more and more about the rodent brain, the distance that separates humans and other animals become smaller and smaller. Last year came the suggestion that rats can laugh. Today there is the suggestion that rats assign complex values to decision making. None of these studies in isolation prove these phenomena, but evidence is mounting that at the genetic, cellular, and organ levels the tangible difference between rodents and man is shrinking. This makes it all the more shocking that humans, quite obviously, operate at a much higher cognitive level than rats; the use of tools, the building of complex structures, the creation of language and music, and so on. At least that is our perception; in fact, the candidate list for what 'makes us human' seems to shrink with each passing year. I and my colleagues in science are not trying to suggest that humans and rats think in exactly the same way. This is not the case; we are clearly different. The question that continues to elude us is, what lies at the heart of the distinction? Things that once seemed obvious (laughter, decision-making not based on binary instinct) are no longer so.

Article Summary
The hypothesis: Researchers claim to have discovered neurons in the rat brain that assign a confidence value to a decision - in other words, how confident am I that this choice I am making is correct? This type of brain function was previously thought to be the exclusive purview of primates.

The Setup: Researchers became interested in the following question: do rats have any capacity to assess the quality of their choices, prior to making them? It's a question of confidence, and here's an example: if I tell you to close your eyes, feed you some cheese and then ask you to identify which is cheddar and which is Swiss, here's how your thought process might unfold. First, you'd smell the cheese, perhaps feel it, and then taste it. Then, based on your prior knowledge of cheeses, you'd hazard a guess as to which is which. If you always order the cheese plate for dessert at your favorite restaurant, you'd be pretty confident in your guess. But if you're trying to lower your cholesterol and are laying off the cheese lately, you might feel a little uncertain. That's a poorly controlled study and wouldn't hold up in the lab, but you get my point. For millennia, scientists and philosophers have concluded that this thing called confidence is an exclusively human, and later at least an exclusively primate trait. Today's researchers want to show that this is not that case, and that rats can do it too. Here's what they did.

The Experiment: The researchers actually chose a (better controlled) version of the story I just fed you. First they poked 3 holes in a box; through the center hole, they puffed a combination of two odors mixed in different ratios. That's critical. Then, depending on the ratio of odors, they put a tasty piece of cheese (just kidding - they actually gave the rats water, which they love) in one of the two holes. If the amount of odor A in the mixture was higher, the water was always delivered to the left-hand hole. For odor B, the opposite was true. Take it from me - rats don't work for free.

The researchers had implanted little electrodes in the rats' brains so they could record the activity of the neurons in a region of the brain called the orbitofrontal cortex (or OFC, for short) during the experiment. The OFC is situated in the front of the brain, an area traditionally associated with decision-making, confidence, morality, and other heady stuff in humans. So, they stuck the rats (one at a time, mind you) in the box and fired up their little scent-puffer. They puffed in the 2 odors at different ratios and let the rats make their decision about which hole to run to in order to get water. They also varied the time until delivery of the water, so that in some later versions of the experiment, the rats could choose to give up and start over.

Researchers looked at the activity of those OFC neurons curing the course of the experiment. They saw two kinds of neurons: the first became more active when the mixture of odors was close to 50/50 and less active when the mixture was dominated (say, 80/20) by one odor. The activity of the second kind of neuron was exactly the opposite. The key to the activity patterns, however, was the timing. It has previously been reported that these two populations of neurons represent the response to obtaining or failing to obtain the reward - in other words, more activity would mean 'Yay, I got my drink!'. However, the researchers had good enough time resolution to observe that the neurons actually started to become more (or less) active nearly a second BEFORE the rat arrived at the hole to claim its thirst-quencher. This implies a greater or lesser sense of expectation, otherwise known as... confidence? They graphed the neuron activity (on the y axis) versus the odor mixture composition (from 100/0 to 0/100, on the x axis) and the graph had a unique looking curved V shape.

The mid-decision neuron activity pattern, when graphed, closely matched a mathematical model of confidence that simply combined the odor stimulus and distance of the stimulus from a 50/50 mixture. More complex models based on learning and memory did not match the neuronal activity. Finally, the researchers added a little twist to the experiment. The rats were allowed to give up early and restart the test if they (presumably) were not confident about their decision. In fact, the rats frequently gave up when the odor mixture was close to 50/50, but rarely when the odor mixture was dominated by one odor.

What does the pithy scientist think?
Disclaimer: what follows is merely opinion, possibly speculation, and occasionally hearsay. But it's the best part, darn it!

Let's explore the points that argue in favor of the hypothesis that rats possess 'confidence neurons':

1. Neurons in the OFC displayed different activities before a decision on which hole to run to for water was made. Some neurons became more active when the odor mixture was near 50/50; this activity may occur when the rat thinks it is making a wrong decision.
   
2. The confidence model predicts the exact and rather complex neuron activity observed while rats are making their choices. Models based on learning and memory alone do not produce the observed pattern.
   
3. Rats were much more likely to give up on a reward and restart the test if the odor mixture was near 50/50. This suggests that they are capable of uncertainty.
   

What about arguments against that hypothesis?

1. This is an odor test. Rodents live for odors. As one adviser once told me, you could take out 90% of a rat's brain and they'd still ace any odor test. That makes subtle differences like those observed in this study a little suspect.
2. The confidence model used in this study is based on very frew factors. Is that all there is to confidence? It seems - dare I say it - overly simplistic. Also, when graphed on the same axes, it did not perfectly match the actualy neuron activity data.
   

What further experiments would support this hypothesis?

1. I'd really like to see these neurons monitored while testing different sensory modalities - sight, touch, taste, and sound. Smell alone makes me nervous.
   
2. Other strains of rats, and mice should be tested for the same behavior. And maybe some more mammals while you're at it. Dolphins anyone?
3. The model seems a little simplistic. It needs to be expanded. For example, I'd like to understand why some neurons fire more when the rat is not confident in its choice, while others fire less. Why the two populations? There is the caveat, however, that the brain is typically a sort of 'push-pull' system, with some neurons pushing the animal to act one way and different neurons pulling the animal in a different way. The actual behavior of the animal is dictated by the neurons that win that push-pull battle (typically the ones that are more active, in a reductionist model).
   

Nevertheless, this is an intriguing study. If true (and it will require more test to prove), it actually provides us with two valuable insights. One, that primates are not alone in our possession of this magical 'confidence'. Two, it demonstrates how probability might be encoded as a pattern of neuron activity by our brains. As I stated in the introduction, this study suggests that another wall we have constructed between ourselves and the so-called animals that live around us could come crashing down. I won't enumerate all of the ways in which scientists and philosophers have in the past sought to separate us from other living creatures, but suffice it to say that number is shrinking at a faster and faster clip. From an evolutionary perspective, though, it comes as little surprise - the slow rate of evolutionary change nearly dictates that for most higher order functions that we typically ascribe to humans, there must be at least some rudimentary correlate back down the chain of evolution. Now, it is only a question of how far back we can look, and as this study suggests - we may have to look much further than we thought.