Do 800,000 human and rodent neurons in a dish in Melbourne, Australia, playing Pong (and doing it well) qualify as intelligent? Enter the neuro-nitpicking paradise if you must. The achievement is far more significant than the descriptors are likely to achieve. The researchers at Cortical Labs and Monash University have created a debate and a half.
The physical side of the story is purely and solely that a DIY neural net has learned to play Pong. Note the word “learned”. That’s the working dynamic here. Neurons do not have to play Pong. Some are even believed to be able to evaluate information.
According to the story, the neurons think they’re the bat, playing Pong. (“Think?”) The neurons themselves may come in various forms and have to perform a few functions. They have to be able to sense stimuli, respond, and coordinate themselves to play Pong. Neurons also act as coordinated switches relative to information and responses.
From this has come a truly turgid debate about whether the neurons are sentient. The logic here is about as good as the theory that the invention of making fire will one day produce dishwashers, but to do so, the sparks must be sentient and capable of designing dishwashers.
Umm… No.
So, are they “sentient”? Anyone’s guess. The guesswork isn’t helped by the ever-more-pedantic and inexcusably self-righteous distances between various definitions of sentience. Even the definition of the word sentience includes a few cop-outs. Sentience, defined purely as the ability to experience something, is a lousy definition of anything, let alone intelligence. What is doing the experiencing? Define that, and you get sentience, albeit in an equally half-baked definition.
Matters of defining sentience are not much helped by the statement that the neurons’ behavior was “compelled” by the researchers. Effectively, that means putting a collection of switches together and getting a light to turn off and on. Not sentience; just function. See this more in-depth discussion for a more rational perspective.)
The researchers may be short-changing themselves quite a lot here. (They’ve created a fully functional behavior pattern and made it work efficiently. This sort of thing is supposed to be the basis of artificial brains, and to a very large extent, it has to be the basis. The neurons have to be able to manage a vast range of tasks.
Playing Pong may not be everyone’s definition of sentience. OK. Fair enough. The essentials of sentience, however, are built in by necessity. To be a bat playing Pong, you need to exercise judgment and achieve an outcome. Sentient? Sort of, but add a bit to the description – You need logic to achieve a defined outcome.
That’s where pre-sentience, if not sentience, has to kick in. A random neural twitch could occasionally achieve the same thing, but not consistently. It’d be all over the shop. This is objective behavior, a very different ball game.
The next step up from this is multi-value logic, the Holy Something Or Other of neuroscience for the creation of artificial brains. Multi-value logic requires degrees of sentience to be objective and therefore effective.
Physiology? You don’t say? I mean, you really don’t.
In physiological terms, this research is probably documented to a very high degree. That little issue isn’t getting a lot of traction in the discussion about sentience, let alone future artificial brains. Why and how are these neurons doing something a bit out of character for a lot of neurons in a dish, for example?
Physiology dictates neural behavior. That’s unavoidable. This is not, however, a One Size Fits All neural network. It’s a custom collection of human and rodent neurons. It works. Why does it work the way it does? Does this neural network create a specific range of unique physiological capabilities? It’d be nice to know, wouldn’t it?
Can this outcome be duplicated by other neural networks using different neurons? This is where being pedantic is exactly the wrong option. Is there a fundamental, always-works, model for neural networks here? Looks like there could be.
…So let’s not go off the rails with not-very-useful debates about “sentience”. This network could at best only be part of an artificial brain scenario. It has delivered a clear achievement, no more.
From this achievement, and it is a significant achievement, we’ve somehow instantly gone to building non-existent sandcastles based on someone’s idea of sentience? Why bother? The debate is off-target, and far more to the point, objectively useless.
This research shows hard evidence of a specific range of capabilities of neurons in a specific network. That’s it. Blue sky should at least admit it’s blue sky. This work is too important to get sidetracked by meaningless abstractions.
As a writer, and without waking up, I could do better and get more commercial traction:
What will become of these poor neurons? Will they become disreputable Pong players in some enigmatic Melbourne university? Will they appear in Then and Now videos on YouTube? Will no one save them? Won’t someone think of the neurons?
Yes, it is a godawful pun. Let’s just get on with the actual research, shall we?
___________________________________________________
Disclaimer
The opinions expressed in this Op-Ed are those of the author. They do not purport to reflect the opinions or views of the Digital Journal or its members.
