No, Brain Cells in a Lab Dish Didn't Play Pong

Given that the agents professing the creed of materialism very much act like clergy by dogmatically professing unproven belief tenets, it seems fair enough to refer to such agents as part of a Ministry of Materialism. The diagram below gives a crude sketch of such a power structure, which is far more complicated than the diagram suggests.  Key players in the power structure include the so-called skeptics mentioned at the bottom right, who do their best to suppress the reporting and studying of thousands of observations that conflict with the materialist worldview, while gaslighting, disparaging and defaming those who report or mention such observations. In psychology analysis of groupthink conformity, such agents are called "mindguards."

To properly understand this "Ministry of Materialism," you need to "follow the money," and the diagram below may help you do that. We get some clues here as to the parties that profit when dubious or misleading claims are passed around in "science news" stories. 

The world of so-called "science news" is like some seedy honkytonk used-car sales lot where pretty much "anything goes." The latest piece of baloney coming out of this "hall of mirrors" carnival-barker world is a claim that neurons in a lab (outside of any organism) have successfully played the primitive game called Pong. For example, an NPR story has this headline: "A dish full of brain cells has learned to play the computer game Pong." The claim is without any merit. 

I am old enough to remember the debut of the game Pong. On a TV show about five decades ago I was introduced to what was then a totally novel idea: that someone might play a game using his TV screen. Pong was about the most primitive video game imaginable. Two players each had controllers, allowing them to adjust the vertical position of a rectangle representing a paddle. An electronic ball would move back and forth on the screen. If a player moved the vertical position of his "paddle" rectangle to a suitable spot so that it blocked the moving ball, the ball would "bounce" by moving in the other direction, towards the other player's "paddle" rectangle. The idea would be to let the "bouncing ball" get past your "paddle" rectangle as infrequently as possible. 

The Pong study discussed in yesterday's news was a study involving some system called DishBrain. The name is doubly misleading, because it didn't involve a brain, and didn't involve a dish. Instead it involved neurons on a very small electronic chip. The study's results appear in a paper with the very misleading title "In vitro neurons learn and exhibit sentience when embodied in a simulated game-world." What the authors created was an extremely confusing "hall of mirrors" situation in which neurons and software were all mixed up with hardware, with the scientists throwing in chemicals and electricity. They placed rodent cells and human cells on some electronic hardware they called "high-density multielectrode arrays." The very confusing mashup included "custom software drivers .. developed to create low-latency closed-loop feedback systems that simulated exchange with an environment for BNNs through electrical stimulation."  With such a setup, it's very difficult to disentangle whether a response was produced by hardware, software or neurons. 

But there are some things that are very clear. The first is that the scientists followed Questionable Research Practices, as typically occurs whenever any cognitive neuroscience study claims impressive results. In the section of the paper entitled "Sample size and blinding protocols," we read this:

"No statistical methods were used to predetermine sample size. As all work was conducted within controlled environments uninfluenced by experimenter bias, experiments were not randomized, and investigators were not blinded to experimental condition."

Here the researchers confess that they failed to use statistical methods to calculate a required sample size, which is a very serious shortcoming. In the paper there are indications that sample sizes too small may have been used. We should not at all accept the lame excuse given for a failure to follow a blinding protocol, the excuse that "all work was conducted within controlled environments uninfluenced by experimenter bias." There would be abundant opportunities for experimenter bias in the collection and interpretation of data in an experiment like this, and in the construction of such "controlled environments."  No experiment such as this should be taken seriously unless the researchers declared and followed a rigorous blinding protocol, and the researchers did no such thing. 

The researchers did not actually produce the slightest bit of evidence of Pong playing by the neurons in their bizarre "mix the software and neurons and hardware" setup. Why is that? To actually play Pong requires muscle action. To play Pong you need to use muscles to adjust a controller. Such muscle action requires volition.  The setup of the experimenters involved zero muscle involvement, zero evidence of volition, and zero evidence of Pong being played by neurons.  

The version of Pong described in the paper wasn't even Pong, as it involved a single moving paddle rather than two (and a paddle much bigger than in the original Pong). The paper describes a "witch's brew" conglomeration of rodent cells, human cells, software, hardware, chemicals, and voltages. When software and hardware and neurons are all mixed up and entangled, in some "black box" complex system, it is easy to pass off software responses as neuron responses. We can be quite sure that the neurons were in no sense whatsoever playing Pong, because the neurons were not connected to any muscles. You can't play Pong without muscle action. What we have here seems to be neuron-entangled software/hardware/chemical action that is being misleadingly passed off as neuron action.

We read that voltage was applied to cells to tell them where the Pong ball was: "75 mV was chosen as the sensory stimulation voltage that would relate to where the ball was relative to the paddle as described in the main text to key electrodes." Then we have some passage letting us know that some objectionable algorithm was being used in which these voltage spikes were being treated as if they were motor activity (what happens when muscles move):

"Spikes are themselves optionally recorded in binary files, and regardless of recording are counted over a period of 10 milliseconds (200 samples), at which point the game environment is given the number of spikes detected in each of the configured electrodes in predefined motor regions as described below. These spike counts are interpreted as motor activity depending on which motor region the spikes occurred in, thereby moving the ‘paddle’ up or down in the virtual space."

So we have here some shady trick in which neurons are being zapped with voltage inputs corresponding to the moving Pong ball's position, and then readings of these voltages in these neurons (via spike counts) are being interpreted as muscle movements, even though the neurons were not actually connected to muscles. This isn't learning. It's just zapping some neurons, and then reading  voltage spikes that arose because of your zapping of neurons. Claiming that kind of thing is learning is  like stepping in some mud, and then claiming that the mud is learning something about the shape of your foot. 

Yes, you can jolt neurons with some voltages, and then read back  voltages arising from the electricity you supplied, using that to determine a Pong paddle position.  You can also supply a metal capacitor with a particular voltage, and then read back the amount of voltage that is in the capacitor. But neither operation does anything to show that neurons or capacitors are capable of learning how to do things.  Such operations merely show that neurons and capacitors can be set to "store a particular number" for a while. Similarly, you can heat water up to have some particular temperature, and then read that temperature with a thermometer. For a while, the water will be "storing" that temperature you gave it. But that does not mean the water is learning anything long-term. 

I can summarize this DishBrain experiment:

Evidence of muscle action: zero.

Evidence of volition: zero.

Evidence of neurons actually playing Pong: zero. 

As for the scientists claiming evidence of learning and sentience (in the title of their paper), the claim is as baloney as someone stepping in the wet snow, and saying this proves that snow is learning and sentient, on the grounds that the snow has learned the shape of your foot, and is now aware of your shoe shape. The Cambridge Dictionary defines "sentience" as "the quality of being able to experience feelings," and no such thing was going on here. 

What goes on in misleading experiments such as this is that at the heart of things is a trick. But the experimenters don't want you to see the simple trick that was being used, so they throw in all kinds of bells and whistles and distractions and complications to minimize the chance that you will see the simple trick that was used. Here the simple trick was mainly supplying neurons with some voltages corresponding to a ball position on a computer screen, and then reading the voltages of neurons to get some number corresponding to the voltage that was supplied, while then representing such voltage readings  as paddle movements (even though they were no such thing, because the neurons -- being unconnected to muscles --  did not move anything). The trick was all entangled with some complicated rigmarole so that the chance of you detecting the trick was minimized. Similarly, a magician doing some trick will surround his trick with various distractions that will minimize the chance of you detecting how the trick occurred.  We can call this kind of thing a "parlor trick experiment." 

It would be futile to defend this procedure by mentioning that there was a tiny physical difference between a "sensory area" of neurons and a "motor area" of neurons, because both areas existed on a chip so tiny (8 square millimeters, only about 3 millimeters wide) that electrical inputs into a "sensory area" of neurons would have been essentially equivalent to inputs into the "motor area" of neurons, particularly given neural connections between the areas and an underlying chip helping to cause electricity to spread around more quickly between the two tiny areas. 

I may note a nonsensical claim in the title of the paper ("In vitro neurons learn and exhibit sentience when embodied in a simulated game-world"), the claim that neurons can be "embodied" in a simulated game-world.  Neurons are incredibly organized and complex three-dimensional physical things, and cannot be "embodied" in any primitive minimalist 2D video game such as Pong.