Back in 1969, Eb Fetz showed that monkeys could be trained to raise the firing rates of individual neurons.1 Decades later, monkeys are trained to do even more complex tasks, such as the so-called “center-out task”, where a monkey moves a mouse cursor controlled by their neural activity alone to one of eight targets.2 With enough training they are able to move the cursor to the target even faster than they could using their hands.
Monkeys controlling mouse cursors without moving a muscle is definitely cool, but what’s the significance of these achievements, exactly? What’s so cool about monkeys moving things “with their minds”?
What the experimenter is doing
Coarse control of a neuron’s firing rate
In the context of Fetz’s 1969 paper, and others following it, the experimenter records from a monkey’s neuron and has that neuron’s firing rate determine the position of a meter that the monkey can see. The monkey is rewarded for making the neuron fire more–i.e., for keeping the meter higher than a certain level.
Fine control of a neuron’s firing rate
A different experimenter named Schmidt, in 1978, also has the monkey watch a meter representing his neuron’s current spike rate.3 But Schmidt rewards the monkey only if he can finely control the spike rate, keeping it within one of eight bins of activity. So for example, on one trial the monkey’s goal might be to make the neuron fire between 3-6 spikes/second for a given amount of time.
The center-out task
Finally, in the center-out task, the experimenter converts the firing rates of multiple monkey neurons into the position of a mouse cursor shown on a computer screen. The monkey is rewarded for moving the cursor to a certain target location. Notice that now the experimenter is recording from multiple neurons instead of just one, and that the mouse cursor’s position is now two dimensions compared to the meter’s one dimension. So the conversion between neural firing and reward is no longer as simple to describe as before! But the monkey can still do it.
How to talk about it
You might notice a subtle difference in how we could describe what the monkey is doing in these three types of tasks. In both Schmidt’s and Fetz’s cases we can say that to be rewarded the monkey must control the level of his neurons’ firing rates, or we can say that he must control the level of the meter. Both claims are equivalent. But in the center-out task, it’s much trickier to explain what the monkey must do in terms of his neural activity; it’s clearer to say simply that the monkey is rewarded for moving the cursor to a particular spot, where the cursor’s position is controlled in some way by the monkey’s neural firing rates. With this difference in mind, what does it even mean to say–as we are tempted to do in the Fetz and Schmidt experiments–that the monkey “must control the level of his neurons’ firing rates”? What does this even mean?
What the monkey is doing
Step back a bit from the monkey and the mouse cursor, and consider: When do you control something with only your neural activity, or with only your mind? All this really seems to involve is that you control something without moving your body, but that’s a negative definition that still avoids explaining what it’s like. In fact, there are plenty of things you control with your mind alone, and they’re not that exciting at all: thoughts, memories, but also the decision to do certain things–called “motor planning”–such as thinking “Move my wrist” before your wrist actually moves.
It’s clear that in these mouse/meter-moving tasks, the monkey is not even aware he has neurons, nor would knowing he has neurons help him complete his goal any better. The way he controls the mouse/meter is probably happening in a way similar to what happens when we plan to make certain movements without actually making them.4 Call it “mind control” if you want, but really it’s something we do all the time.
So from the subject’s perspective, controlling something with your mind really has nothing to do with controlling neural activity, even if that’s what the system is rewarding you for. In other words, you don’t have direct control over your neurons; you have only indirect control of them, via the systems of internal communication you’re used to using: movement, thought, decision/planning, and so on.
This can be seen as a distinction between our body’s mental “software” (things like motor planning and memories, that aren’t directly observable in our brain’s physical state) and “hardware” (the neurons and spikes themselves, which we can measure and control)5. The scientists in these experiments are concerned with the hardware part, seeing how to get the monkey to independently control neural activity. The subject, on the other hand, participates in this experiment with his volitions (software), the same way he does anything.
Why we care
To many, a monkey moving a mouse cursor around without using his hands is cool no matter how it works. But it’s important not to go too far and get excited at the idea that the monkey is controlling the mouse “with his mind”. Everything you can control, after all, you control with your mind.
Instead, what’s cool here is that a monkey can be trained to control an external event like a mouse cursor in a new way, just like we somehow learn to point at things with a pencil, or even with our own fingers.
We don’t know how the monkey achieves this, but you can think of the monkey trying out things like “Move shoulder muscle” and seeing how it affects his feedback on the meter or the monitor. In any case, the monkey is somehow able to tap into the causal chain that the experimenter creates–between action, neural firing, and feedback loop–and can use it to control an external event.
A major benefit of these brain-machine interface experiments is that we now know how to rig up a situation in which a monkey can vary his neurons’ firing rates in a controlled manner. But how much control does he really have? Does it depend on how the experimenter maps it to the cursor movement? And how quickly can the monkey improve with training? By exploring these sorts of questions, we can learn about the limits of our brain’s system for controlling actions and tools.
Footnotes
-
Fetz’s “Operant Conditioning of Cortical Unit Activity” (1969) abstract ↩
-
Georgopoulos et al.’s “Neuronal population coding of movement direction” (1986) abstract ↩
-
Schmidt et al.’s “Fine Control of Operantly Conditioned Firing Patterns of Cortical Neurons” (1978) abstract ↩
-
In the Schmidt et al paper, actually, the authors mention that to control the meter the monkey is in fact just moving whatever muscle is associated with the activity of the neuron they’re recording from. So really, the monkey seems to be just trying a bunch of muscle movements until he finds the one that controls the meter. ↩
-
Daniel Dennett, in Consciousness Explained, seems to favor this division as well: “…Why do I persist in likening human consciousness to software?” Because human consciousness “1) is too recent an innovation to be hard-wired into the innate machinery, 2) is largely a product of cultural evolution that gets imparted to brains in early training, and 3) its successful installation is determined by myriad microsettings in the plasticity of the brain, which means that its functionally important features are very likely to be invisible to neuroanatomical scutiny.” ↩