I hope that our detailed look at the evolutionary genetics in the Levine lab. As well as our discussion of theory earlier has given us some background to develop good intuitions about the nature of experiments. But since we're doing philosophy of science, I want to get abstract a bit and think about how we should characterize experiments more generally. I've already explained the core of an experiment is an intervention, in order to think about the logic of experiments, we're going to need to make an idealization. We have to deliberately simplify something in order to understand it better. The idealization we are going to use comes from philosopher Jim Woodward, what he calls an ideal intervention. An ideal intervention is speaking metaphorically a surgical intervention. We have complete control of experimental variables and when we tend to change one of them, we only change that one and anything else that cause Ali depends on it. With the idea of an ideal intervention in mind, we can begin to think about the logic of experiments by returning once again to Mills methods. In fact, we can return to the unfortunate example of everyone getting sick from the tofu. In general tofu doesn't make people sick, in fact, it's an essential part of many people's diets. So something obviously went wrong in this case, and one thing we might try to do is go back to the particular situation and drive some clues. But another thing we might do is to create an experiment and to find out under what kinds of conditions illness causing bacteria like e-coli or listeria can grow on tofu. So let's imagine an experiment, in this experiment we might test tofu under five different conditions. Just taken out of the refrigerator, left on the counter for 30 minutes, left on the counter for two hours, left on the counter for 10 hours and cook to 75 degrees C. In principle, we could wait around to find tofu in each of these conditions, but a much more efficient way is to create the conditions we want to investigate. So we would take our tofu probably many pieces and subject them to the five different conditions, and then do a test for bacteria under each condition. So we can infer from this experiment that bacteria can grow at room temperature. And while these bacteria aren't an immediate danger to humans, we need to be very careful about leaving tofu at room temperature for very long. If this Were a real experiment and not just a simple example, we'd probably go into a lot more depth trying to determine how long it takes for bacteria to start growing. When the dangerous ones are present and what can be done to minimize health risks. But what I hope to have shown you, is that the logic of experiments is just like the logic of observations the application of Mills methods. Now let's recall what happened in the Lavin lab. >> What my lab does instead is we take versions of that gene from closely related species put it back into a drosophila melanogaster and ask what breaks. And we think that what breaks when we put the wrong species version of that gene into our model organism, that tells us what's the biology that evolution has shaped in the past 500,000 to 2 million years of evolution. And we're using this approach to study specifically the proteins that package chromosome ends or telomeres. So we are specifically interested in genes that encode proteins that glom on to the DNA. >> So we can see that from the simple example of tofu to the complex example in a modern genetics laboratory, much of the logic of experiments is the same. You might be wondering why I haven't mentioned statistics as I've discussed the logic of experiments, a lot of modern experiments, of course do involve statistics. And why do they? Well, some effects don't show up every time or a hard to detect or their causal factors that are hard to control. And in all of these cases, scientists need to use statistical methods to tease out the signal from the noise. This is a huge topic worthy of its own discussion its own course even, and we're not going to be able to discuss it other than to say that the basic logic remains the same even if we need to introduce the mathematics of statistics. Before moving on, we need to talk a little bit more about causation. I've spoken very loosely about how observational and experimental methods help scientists find out what cause something to happen. Food poisoning in our example or chromosome integrity and Professor Levine's lab. Mills methods are helpful guide to finding causes, but they do break down as a guide in certain circumstances. And one of the most well studied cases is situations where you have a common cause, so here's an example. A lot of the time when it rains or is about to rain I get a migraine, for some people the connection between migraines and storms is like clockwork and I'm sorry if you're one of these people it's totally awful. So let's say over the course of a month we made a mil style chart of migraines and storms, and we would see on this chart a link migraines and storms are highly correlated. But if we weren't careful we could end up inferring that migraines cause storms, just the way we inferred that illness causing bacteria grow at room temperature. In other words, correlations only tell us two things co-occur, they don't tell us what depends on what or even if there is a causal relationship at all between the two variables. Now everybody knows that migrants don't cause storms and storms don't cause migraines, and yet they're highly correlated. But what explains this? The explanation is that they have a common cause, sudden drops in atmospheric pressure cause both migraines and storms. In this kind of case, where there's a common cause that hasn't been explicitly considered, Mills methods will not uncover causes only correlations. And it's because of this that you may have learned that correlation does not equal causation. That statement is true although as we've seen correlation is a good guide to causation. So before moving on I want to raise one more issue about experiments, when most of us were in high school, we learned about what teachers called the scientific method. It was taught in a pretty formulaic way, do some background research, formulate a hypothesis, test the hypothesis with an experiment, and then conclude that the hypothesis was true or false. I hope our time together so far as convince you that this is a massively overly simplified picture of what scientific research looks like. But I do want to briefly focus on one particular over simplification in this formula. When we're taught the method, we're taught that the purpose of doing experiments is to always test hypotheses, but is this true? I think the answer is simple, no, lots of experiments to test hypotheses, but lots of other experiments as well as straight observations and simulations are exploratory or align with theories in more complex ways. Scientific results are often written up talking about the connection between theories or hypotheses on the one hand, and experiments on the other. But that doesn't mean that this is literally the method or procedure that has to be used in every case. In the next part, we're going to talk about simulations, something you may not have learned about at all in high school.