 I want to start this video off with a story. Once upon a time, there was a very good kingdom. The kingdom was beautiful, kind, just, and the people filled their days with art, learning, science, and joy. The king was so proud of his kingdom that he wanted a perfect map so that all the world scholars could study it, understand it. So he sought out and hired the finest cartographer in all the world, and gave him an unlimited budget to produce the perfect map of his kingdom. The cartographer labored for many years, surveying, measuring. After some time, he produced a map that, unrolled, covered the entire floor of the king's throne room. Yet the king was not satisfied. The hills and lakes and rivers were visible, but the cities and roads and houses had no detail. Only his scholars and scientists would not get a complete picture of the kingdom, would not completely understand it. It was missing too many details. So he sent the cartographer back to produce a larger, more detailed map. And six years later, the cartographer unveiled a map of exquisite detail that covered several city blocks in size. And the king was pleased that he could make out tiny houses and trees. But he still could not make out the leaves on the tree, or the carvings that the people put on their doorposts, or the lovely gardens with rocks carefully groomed into patterns. Without these details, the true beauty of the kingdom couldn't be understood. So he sent the cartographer back to his studio to produce a truly perfect map. Many decades later, as the old king was nearing the end of his life, he went to the cartographer to find that he was putting the finishing touches on his perfect map. What the old king saw was this. The mapmaker had made a perfect one-to-one replica in size and shape, in fact in every detail of the kingdom. The houses were the size of the houses they represented. The cities and roads and lakes were perfect recreations from the original surveys taken long ago. But the kingdom had changed in the decades since work began. Disease stalked the land, and famine and poverty. Fires had destroyed forests and homes, and villages abandoned. The good king, seeing that the map was so perfect, decreed that all the people of his kingdom should simply live in the map. And so they did, and they lived happily ever after, the end. There are many messages we can take from this story. Don't spend all your money on maps, for example, or get a written estimate from contractors before they start work. But the point I want to make is about models and maps and how we use them in science. My central idea is this. Every model is wrong. A perfect map is useless to help us understand the thing that it maps, so that only flawed and incomplete models are ever useful. Let's take a look at the human genome. You may recall that the Human Genome Project was completed in April of 2003. Except it really wasn't. The Human Genome Project never aimed to sequence all the DNA in human cells. Only the 93% that was worth sequencing, the U-chromatic sequences that contain genes. The other 7% near the middles and ends of chromosomes are structural and highly repetitive. So we still probably don't have a fully complete human genome assembled. And that's okay, because the bits we left out aren't likely to tell us anything. Even that incomplete map of U-chromatic DNA isn't your genome. It's not in fact anyone's genome. It's a generic consensus genome mixed and matched from various people from around the world who volunteered to donate samples. Originally, 5 people out of 21 candidates were selected for sequencing. As it happens, it doesn't much matter, 99.9% of human DNA is identical in sequence among all populations. Our incomplete genome sequence of no one in particular isn't a full representation. And that's because DNA isn't actually a series of A's, C's, T's, and G's. It's a real molecule with a real shape and real chemical properties. Some of your genome is chemically modified, and some of it is so tightly wrapped around structural proteins that it's inaccessible and inactive. Some of your DNA is damaged, and some of it is mobile and moving around. And you have different genomes in your different cells. The human genome, in short, is a pretty poor map of the real thing, and that's still okay. You may have seen this screen print, the treachery of images, by Rene Magritte. The text says in French, this is not a pipe. And of course it's not, because it's just a picture, a representation of a pipe. The word pipe is also not a pipe. It's merely a way of verbally communicating the idea of a pipe. You may have heard the phrase, the map is not the territory. The meaning of which is shown in the initial story I told. The only perfect model of nature is a one to one representation in every detail. Which is just the thing itself for a perfect replica of it. The question is then, what good is a map of a place that is indistinguishable from the place itself in size, scale, and detail? What good is a scientific model that reduces none of the complexity of the thing itself? All models in science are wrong by design. They are necessary reductions of the full characteristics of a thing. No genome map that is anything less than a perfect replica of the genome itself will ever fully model the properties of your genome. Every model we make of your genome will be wrong at some level. And any perfect model will be essentially useless as a tool. Scientific modeling requires a level of simplification, of wrongness. When we compare these simplified flawed models to real behavior in the natural world, we find the characteristics that can be reduced and the characteristics that cannot. Another good example of this phenomenon is in mechanics. The Newtonian model works well in the world we live in, living in a gravity well at medium energies and medium sizes and low velocities with easy to identify frames of reference. It's a deeply flawed system because it fails at high energies, small sizes, high velocities. But as an imperfect model, it makes useful predictions. In that way, even falsified models can still be predictive, can still serve a purpose. How can we apply this to medical research? Take animal models like the mouse or rat. They often fail to accurately predict the impact of an investigational drug or treatment in humans. How should we evaluate their usefulness in drug trials? The model is highly imperfect, but the only perfect model of human drug response are humans. Exposing humans to drugs with no idea of reaction raises ethical issues. We could use non-human primates, but we're going to be trading off accuracy of the model for ethical concerns. A better map would be a human surrogate without a nervous system. But even then, our model would be incomplete since we can't measure pain, neurological response, or psychological effects. In my own field of diagnostics, we often face the proxy issue. If I'm trying to determine if someone is infected by a virus, I look for signs of protein or nucleic acid, the DNA or RNA, in their blood that match the virus. Or in some cases I'm measuring antibodies to a specific virus. But this is a proxy measurement. I'm not actually detecting or enumerating virus. Not every HIV RNA molecule was part of an infectious virion. My model is flawed and inaccurate, and that's a feature, not a bug. It's a subject that some HIV denialists have seized on because it's a potential source of error. But it's like every other scientific model. It has built into it a simplification that makes it useful. We validate our proxy measurements against clinical outcomes. Proxy tests that are predictive of who gets sick are useful and can be compared to other tests to arrive at a test concordance or agreement between methods. This is equivalent to evaluating a map on its ability to guide you to your destination. Detail in a model can sometimes be a detriment. In conclusion, in order to understand something to make it useful to us, we have to simplify it with our model. Every useful model reduces the full behavior of the thing being modeled. And therefore every useful model is intentionally wrong. A fully correct model would be a useless replica of the thing being modeled. This is a problem we can't escape and one that scientists have come to accept and live with. Just remember, if you found this hard to understand or perplexing, you're not actually listening to my words. I'm not really here. You're listening to the ghost of my words, compressed into a series of ones and zeros, except they aren't really ones and zeros. They're magnetic fields or photons or electrons, except of course that photons don't really exist the way we think of them. You're not even really thinking about my words, just the model of my words triggering responses in your neurons. And the fact that this video is ending is all a construct of your limited model of time's arrow. Well, that's what I think. Thanks for watching.