How do we integrate information?

Left or right? Apple or orange? Selma or Birdman? One way to make these decisions is precisely what intuition tell us it should be: we weigh up the pros and cons of each choice. Then, when we have sufficient evidence for one over the other then we go ahead and make that choice.

How this is represented in the brain is quite straightforward: the firing of neurons would go up or down as evidence for one choice or another becomes clear and, when the firing had reached some fixed threshold, when the neurons had fired enough, a decision would be made.

The difficulty has been in figuring out precisely where the information is being encoded; in determining which neurons were increasing their activity in line with the evidence. In fact, multiple regions seem to be participating in the process.

So let us say that you are a little rodent who hears sound from the left and the right; little click click clicks. And you need to decide which side the most clicks are coming from. Every click on one side gives you a smidgen of evidence that that side will have the most, while a click on the other side will make it less likely. You don’t know when the clicks will end – so you have to stay ready.

Now there are two interesting areas of the brain that we can look at: the frontal orienting fields (FOF) that probably guide how you will orient your little snout (to the left? to the right?), and the posterior parietal cortex (PPC), which integrates diverse information from throughout the brain. Here is what the activity of these neurons look like if you plot how fast the neurons are firing, separated out by ‘accumulated value’ (how much evidence you have for one side or another;  I will refer to this as left or right but is actually more like ipsilateral or contralateral):

PPC vs FOF

It looks like PPC, the cortical integrator, fires progressively faster the more evidence the animal has to go left, and progressively slower the more evidence it has to go right. In other words, it is exactly the evidence accumulator we had been hoping for. The orienting region (FOF) has a different pattern, though. Its firing is separated into two clusters: low if there is a lot of evidence to go left, and high if there is a lot of evidence to go right. In other words, it is prepared to make a decision any second, like a spring ready to be released.

It is interesting that this is implemented by sharpening how tightly tuned neurons in each region are for the decision, going from something like a linear response to something more like a step function:

value vs firing rate

This is consistent with an idea from Anne Churchland’s lab that the PPC is integrating information from diverse sources to provide evidence for many different decisions. This information can then be easily ‘read out’ by drawing a straight line to separate the left from the right, a task that is trivial for a nervous system to accomplish in one step – say, from PPC to FOF.

And yet – there are mysteries. You could test the idea that FOF provides the information for left or right by removing it or just silencing it. If it was standing ready to make a decision, you would only care about the most recent neural activity. Indeed, ablating the region or just silencing it for a couple hundred milliseconds has the same effect of biasing the decision to the left or right. But it is only a bias – the information for the decision is still in the system somewhere!

Even more baffling is that the FOF begins to respond about 100 milliseconds after hearing a click – but PPC doesn’t start responding until 200 millseconds after a click. So how is FOF getting the information? Is FOF actually sending the information to PPC?

Decisions are hard. It is not a “step 1. hear information, step 2. add up pro/cons, step 3. make decision” kind of process. A linear 1, 2, 3 would be too simple for the real world. There are many different areas of the brain getting information, processing it and adding their own unique twist, sending their evidence to other areas, and processing it again. Again: even simple decisions are hard.

References

Hanks, T., Kopec, C., Brunton, B., Duan, C., Erlich, J., & Brody, C. (2015). Distinct relationships of parietal and prefrontal cortices to evidence accumulation Nature DOI: 10.1038/nature14066

Brunton, B., Botvinick, M., & Brody, C. (2013). Rats and Humans Can Optimally Accumulate Evidence for Decision-Making Science, 340 (6128), 95-98 DOI: 10.1126/science.1233912

Mechanisms of collective decision-making in bees

Thomas Seeley gave a talk yesterday about how honeybees collectively decide on a new home, and how they use the evidence accumulation/drift-diffusion model to make decisions!  When bees are ready to start a new colony, they’ll find somewhere to hang out and swarm.  Foragers will then periodically wander off to try to find a good home: they like spacious holes high up on tall trees.  When a forager finds a home that it likes, it will report back to the swarm what its found in the form of a waggle dance.   Bees will shake their little bottoms as they walk in the direction of the potential home, and the longer they shake the further away their new home is.  But the job of the bee is more of that of a politician or proselytizer trying to get other bees to follow their lead to the new nest site that they found.  By dancing more and more, a scout bee will impress other watchful bees to go check that site out; generally, the more the scout likes the nest, the more time and energy its willing to invest in its wagglin’.

Waggle danceBut politics is rough and tumble.  Scout bees don’t only advertise nest sites they like, they’ll actively go find the other scouts advertising other nests sites and headbutt and buzz at them.  Needless to say, the more an Opposition bee gets headbutted, the less likely it is to continue advertising its own preferred site.

What this gives us, though, is a feedback loop where better nests cause more waggles and more inhibition of other nest sites which recruits even more scouts to check out the nest to do more waggles and more headbutts.  In this way, bees will essentially always find the best nest.

This resembles nothing more than the noisy evidence accumulation that is used to explain human and other animal behavior, but on a large scale; now it is not neurons or brain regions accumulating evidence, but a society.  Every time a scout brings back its opinion on the nest quality, it will recruit more bees (evidence accumulation in favor) as well as inhibit other scouts (evidence accumulation against).  When enough evidence has been accumulated, the swarm reaches a threshold and off they all go!  Interestingly, the threshold is always when 15 bees have reached the potential nest site.

Bees drift and diffuse

I had a couple of questions for Thomas Seeley which he unfortunately did not know the answer to.  I was curious whether the bees that waggle were the same ones that headbutt.  That is, are are bees just ferociously in favor of their personal site and will do what they can to promote it?  Or are some bees bullies and some bees charismatic politicians?  Each has a different set of implications.  I also wondered whether the threshold changes with swarm size.  Too many bees in a swarm and you can reach the tipping point too soon; too few and it can take too long.  Seely didn’t have evidence on that either and seemed to misunderstand the decision-making model, unfortunately (he kept explaing, “physicists say this is how it is done.”).

References

Griffin, S., Smith, M., & Seeley, T. (2012). Do honeybees use the directional information in round dances to find nearby food sources? Animal Behaviour, 83 (6), 1319-1324 DOI: 10.1016/j.anbehav.2012.03.003

Seeley, T., Visscher, P., Schlegel, T., Hogan, P., Franks, N., & Marshall, J. (2011). Stop Signals Provide Cross Inhibition in Collective Decision-Making by Honeybee Swarms Science, 335 (6064), 108-111 DOI: 10.1126/science.1210361

Decision Theory Journal Club: Our brains are perfect machines

A few of us have started a Decision Theory journal club where we plan on reading papers from a variety of fields that examine how decisions are made.  We have people from neuroscience, economics, and cognitive science participating (so far), including people participating through Google+ hangouts!, which will hopefully make lead to some productive discussions.  I’m a couple papers behind, but I hope to post summaries of what we have been reading.

Our first paper follows an idea that is common in the psychological literature concerning how someone gains evidence, the noisy evidence accumulator (diffusion to a boundary).  Let’s say you hear a loud noise and have to decide whether to look to your left or your right.  If the noise is almost directly behind you, it can be difficult to tell which way to look.  Both of your ears are going to be hearing something loud, and as the sound waves crash about the room it will make the sound even noisier: sometimes one ear will be louder than the other.  But one ear is usually louder than the other and when you’ve received enough evidence that one ear is hearing something louder than the other, your head will swivel and your decision is made.

We can do essentially the same thing with rats.  They can be put into a chamber where clicks will randomly come from speakers to their left and to their right, and if they turn in the direction with the most clicks, they get a reward.  Rats are fairly good at this – as are humans.  One interesting difference, though, is that when humans are certain, they will always go in the direction of the most clicks.  Rats, on the other hand, peak out at about 90% certainty; I guess they don’t trust the experiment as much as people do and want to explore their environment more!

But we’re interested in how this decision is made, so we can go back to our noisy evidence accumulator and see if that can explain how well the decision is made.  We can also through in all sorts of options: is the memory of the rats a bit forgetful? Is there all sorts of internal noise in the brain?  Is there noise in the environment?  And so on.  It turns out that the headline of the paper tells it all: rats and humans are optimal evidence accumulators.  There is no internal noise.  There is no forgetting.  Every bit of evidence that is given to the animals is in there, waiting to be used.

Fortunately, results from a different paper can explain to us what might be happening.  There are direct connections between the cortical auditory neurons and neurons in striatum – an area of the brain that receives dopamine and is involved in selecting the best action to take.  Activating these auditory neurons signals the striatum and makes the animal more likely to go in whichever direction the experimenter wants.  Inhibiting these neurons has the opposite effect.  It’s quite possible that the auditory input is interacting with this dopamine system to keep track of where an animal wants to go – and what decision it wants to make.

As for the optimality of the animals, well, that’s at least the headline, and it would be great if that were always true.  In actuality, there’s a large population of rats which show sub-optimal evidence accumulators.  Although they don’t discuss this in the paper, to me this is the most exciting result (although the lack of neural noise ranks up there, too.  Our brains are machines.).  Of course you’d expect that evolution would evolve animals that, well, make good decisions.  So why are there any animals that do show significant neural noise? Why is there such large variability in forgetfulness?  Although the majority of animals are almost perfect, many are not.  Hopefully in the future, we will be able to explain why it’s good to not always be perfect.

References
Brunton, B., Botvinick, M., & Brody, C. (2013). Rats and Humans Can Optimally Accumulate Evidence for Decision-Making Science, 340 (6128), 95-98 DOI: 10.1126/science.1233912

Znamenskiy, P., & Zador, A. (2013). Corticostriatal neurons in auditory cortex drive decisions during auditory discrimination Nature, 497 (7450), 482-485 DOI: 10.1038/nature12077

Photo from