The Limits of Machine Learning in the Physical World

“As a man who has devoted his whole life to the most clearheaded science, to the study of matter, I can tell you as a result of my research about the atoms this much: There is no matter as such! All matter originates and exists only by virtue of a force which brings the particles of an atom to vibration and holds this most minute solar system of the atom together. . . . We must assume behind this force the existence of a conscious and intelligent Mind. This Mind is the matrix of all matter.”
– Max Planck

If we envision a day when machines will equal, and eventually surpass, human intelligence, we must consider what this thing called “intelligence” really is and where it comes from. Some careful logic reveals a profound insight that raises serious questions about whether and how true intelligence could ever come from anything we humans create.

First, let’s decide whether intelligence even exists. Isn’t the answer obvious? Yes, I think it is, but hang on to that thought.

Did you know there’s a perfect, giant hexagon on Saturn’s north pole? Check it out. Does the regularity of this shape mean that Saturn was created by intelligence? Maybe, maybe not. After all, regular patterns do emerge in nature – in crystals, for example. So, although this planetary feature is strange, scientific explanations as to how it could emerge without conscious intention have been presented.

What about a skyscraper? Does the existence of a tall building with all the interior and exterior design elements and intricate infrastructure imply that it didn’t emerge by accident? It didn’t just appear as a result of mindless, random movement of people and materials, right? Of course not. That’s ridiculous. So, are you sure intelligence – that is, real choice supported by acquired knowledge and understanding – is involved in an endeavor like that? I am, too.

But if you believe that what we call “intelligence” comes from a physical object – like a brain – then you must believe that intelligence doesn’t really exist, either to create beautiful geometry in the solar system or to design and construct a modern skyscraper. You must believe that everything you see around you, no matter how complex or seemingly intentional, just sort of happened. It’s all just the result of lifeless molecules bouncing around, energetically flowing down an entropic hill without aim or purpose.

How can this be? Think about it: Have you ever heard of a molecule deciding where it wants to go? Have you ever heard of electricity flowing with conscious intent? Building up from there, then, can you imagine a neuron deciding whether it wants to fire? Of course not. Neurons fire based on inputs. So, if no neuron has ever decided to fire through conscious choice, then how can any living being, no matter how sophisticated its brain, ever make any decision at all? Yet, it certainly looks like we do exactly that.

Decisions may be real, they just can’t come from brains because brains are physical things and physical things don’t make decisions.

You might think that at the macro level properties have emerged, through a long process of evolution, that don’t exist at the individual molecule or neuron level, somehow producing freewill and such. How does that work, exactly? It doesn’t make any sense. If you shake a bag of marbles for a trillion years, it won’t suddenly become intelligent one day.

“Complexity theory” may attempt to discover how at least the appearance of intelligence emerges in complex systems, but real intelligence in any system still rests on the assumption that choice is occurring somewhere. Where? Nowhere, that’s where. At least no place physical.

Even if you go beyond classical physics and consider theories that involve quantum mechanics, such as the proposed quantum mechanical properties of cellular structures called “microtubules” and their possible role in consciousness, that doesn’t introduce intention arising from anything physical. It just gets you a little closer to the source, with a brain as the conduit from that source, not the source itself.

OK, so we’ve decided that intelligence does exist but that it can’t possibly come from physical brains. If this is true, how can anything we might call “real intelligence” ever be created in a machine? It can’t, unless we can figure out a way to replicate or connect machines to the true source of intelligence, whatever it is.

Machine learning can do some amazing things, and there will be dramatic advances in the automation of apparent intelligence in the coming years. And, who knows, since artificial intelligence is subject to the same mysterious non-thing that produces real intelligence, the Blue Fairy may one day transform our Pinocchio creations into real beings, with all the trappings of consciousness and will, whatever the source.