What Shall I Do Next?
The neuroscience of action
It seems that the Clash were keen students of neuroscience. The band’s song “Should I Stay or Should I Go?” beautifully captures the action selection dilemma:
If I go, there will be trouble,
And if I stay it will be double,
So come on and let me know…
Should I stay or should I go?
Caught twixt two actions, our narrator cannot do both, cannot both go and stay at the same time. Our narrator must select one and one only.
To which their basal ganglia replies, “Go, you fool — you’ll be in half the trouble. How crap is your maths?”
The action selection dilemma is at the core of our everyday existence. There are many things we cannot do at the same time. For some actions, this is because there are only a limited number of ways we can use our muscles. We cannot hop, skip, and jump at the same time. No, we must select one, and only one. (And perhaps do them one after the other, in homage to the genius who decided that if you do them in the right order, and landed in some sand at the end, you can win a gold medal for it.)
For other actions, we cannot do more than one of them, because we have only one body. We cannot go the florists, the tattoo parlor, and the bank at the same time. Not unless a Flora-Drawa-Wonga franchise has opened in your neighborhood.
For yet other actions, we cannot do more than one of them because we have only one consciousness, one stream of awareness that we can control. So we cannot read and do math at the same time. Or contemplate the mysteries of existence and name every member of Iron Maiden.
So, almost every moment of every day, you face the action selection dilemma: What shall I do next? Of all the possible things I could do, what one thing shall I do?
Our best theory is that the basal ganglia solve this dilemma for you. Our theory says that this collection of brain parts, tucked under the cortex, listen in on the shouty, brash bits of cortex, each demanding that they, and they alone, be allowed to steer the good ship You. And having listened to these demands, the basal ganglia allow one demanding bit of cortex to have access to the engines — the control centres in the brainstem and spinal cord. And lo, the next action is selected.
We have a litany of reasons to believe the basal ganglia must do something vital for action.
For one, they receive inputs from every single bit of the cortex. From the back of the cortex, where reside the most primitive representations of what the eyes see, through the top, where reside the bits of motor cortex, all the way to the very front and the bits of cortex that deal with goals and how to obtain them. The basal ganglia even receive inputs from ancient cortex, the bits that deal with knowing where you are (the hippocampus) and how you feel about that (the amygdala). Why would anything need to know all that? If that bit of the brain was deciding what to do next, then it would need to know all that: what was happening in the outside world, what the body was doing right now, what the owner of that body was planning to do, where that owner was, and how that owner felt about it all.
For another, the basal ganglia’ outputs are on all the time—and inhibit whatever they contact. This is very weird. It is as though the output neurons are shouting “Nooooooooooooooooo” constantly and loudly at everything in earshot. Why spend all this energy? The action selection theory makes sense of this: This is the brake, the constant interrupter that prevents you from trying to do everything at once and crumpling to a heap on the floor.
These outputs occasionally pause, but just a few of the outputs at any one time. This pause removes the constant Nooooo from just a tiny portion of all the possible contacts of the output. As though the output is saying to the engines responsible for one specific action: Go on then, do your thing.
And what makes the outputs pause? Brief, strong input from the cortex. Put all together, this means that the basal ganglia can take in demands for action from different bits of cortex, compare them, and select only the strongest demand — allow only that demand to have access to the engine room of the brain.
Precisely how the basal ganglia turn the many demanding inputs from cortex into a single, brief pause is an open problem. The wiring within and between the different parts of the basal ganglia is complex, and our knowledge of it is evolving constantly — alongside the long list of Latinate names of its parts, cunningly designed to confuse generations of undergraduate students. Those wanting a deep dive can read more here and here.
How do the basal ganglia know which is the best demand from the cortex? Because these demands seem to have a value. Or, more accurately, a predicted value. The predicted value of how much reward you will get in the future if you did this action right now. It seems the strength of each demand from cortex — the amount of activity of those cortical neurons — tells the basal ganglia how much reward is predicted to happen in the future if their action is selected. So the basal ganglia can select the action with the highest predicted reward.
These values carry two terms and conditions. Condition one is that the value of an action does not mean that the reward will immediately follow that specific action. Rather, the value means that this action is a necessary step toward getting that reward. Much like if you head to the kitchen for breakfast in the dark depths of a winter morning, and the first thing you do is turn on the kitchen light. The action of turning on the light itself does not come with an immediate reward. Flicking that switch does not immediately shovel a pancake into your mouth. But it is a necessary, essential step in getting you closer to the goal of eating breakfast and so is highly predictive of future reward. So we can choose actions one at a time and still end up getting our final, valuable goal.
Condition two is that the values of actions are specific to context. In trying to make breakfast on a dark, dreary morning, flicking on the light is valuable. In trying to placate your angry boss because you just wiped the company’s database, repeatedly getting up to flick the light switch in the boss’ office is not valuable, not valuable at all. An action can be highly valuable in one context and idiotic in another. So the fact that the basal ganglia get information about where you are seems perfect for telling it whether you’re about to get breakfast or about to get fired. And it can appropriately select the most valuable actions.
Putting these ideas together, solving the action selection dilemma seems simple: The basal ganglia should just select the demand from cortex that is for the action with the best predicted value. Ah, but were it so simple.
You see, there is also the exploit-explore conundrum. This conundrum is basic to our everyday existence but barely registers in our awareness. The conundrum is that we cannot always simply choose the best action, because our estimate of the best action is simply that: an estimate. There are two problems: It could be wrong, or it could be right at the moment but become wrong at a later time because the world is constantly changing. So there is always the need to explore other actions, to find out if they are better or have become better while we weren’t watching.
But we can’t just incessantly explore either. If we did, we would never get much reward. If we never stick to one course of action, we never reap the benefits from repeatedly doing something we knew to be good. So there is also a need to exploit our current knowledge of the world, to use our predictions about the value of things. Thus the exploit-explore conundrum is ever present: We need to select an action, most likely the best one — but not always.
How do we know when to select the most valuable action and when not to? We don’t know. How does the brain create randomness? We don’t know. But every now and again, some part of your brain rolls some dice to select an action. It’s a fair bet that part of the brain is the basal ganglia.
There is a long list of mysteries about the basal ganglia that are solved by the theory that they do action selection.
A perplexing mystery is that we know the basal ganglia must have something to do with movement but cannot cause it. If you stick an electrode into a very specific group of neurons in the basal ganglia and tickle them with electricity, then your arm will move up and to the left. But if you destroy that same specific group of neurons, you can still move your arm up and to the left. The theory that the basal ganglia do action selection solves this mystery: They represent that action but do not cause it themselves — they allow it to happen.
(This is a clue that there is, of course, more than one system in the brain for selecting what to do next. Sometimes we need to select the next action as fast as the physical limits of our body will allow, like when you absentmindedly pick up a scalding oven tray without a mitt on, and you’re very grateful that your spinal reflex arc made your fingers drop the tray instantly rather than wait for permission from your cortex to do so. And sometimes we need to select an action fast when something predicts imminent danger, which uses the brain’s fear system.)
Another mystery is that the basal ganglia seem vital for survival. We think this because every animal with some sort of backbone also has something that looks very much like a basal ganglia in its brain, even if it doesn’t have much of a cortex to write home about. Birds have basal ganglia. Sharks have them. Even lampreys have them, and our last common ancestor with those ugly tubes of teeth was 360 million years ago (about 130 million years before the dinosaurs turned up). And the action selection theory makes sense of this too. Selecting the right action at the right time in the right place is essential to survival.
But perhaps the most resounding mystery we may solve is why damage or changes to the basal ganglia underpin so many brain disorders. Huntington’s disease, with its uncontrollable, random movements of the arms and legs. Parkinson’s disease, with its slowness of movement and inability to start moving. Tourette’s, with its tics and saying the wrong thing at the wrong time. Attention deficit disorder, with its inability to stay focused on one, appropriate action. Obsessive compulsive disorder, with its debilitating focus on one, inappropriate action. All have been linked to damage or changes to neurons in the basal ganglia. All have the potential to be explained as a malfunction of action selection. So when people ask me, “Why do you study the basal ganglia?” I point at that list and say, “Why would I study anything else?”