Alicia, Four Cups discussed on Christopher Kimball's Milk Street Radio

Automatic TRANSCRIPT

And how how good was this cookie versus the mathematical average cookie? Oh, it was it was lacking. So the recipe that we ended up with had it was something like four cups of shortening. And then a whole bunch of Brown sugar, just a stupid work. Yeah. A lot. I don't remember the flower count. But not enough the ratio was really off. So what are the look like when it came out of the oven that it's spread all over the place or oh my God. Yes. It was just this big Brown nicely because there was so much Brown sugar. But it just spread it was absolutely enormous. It did cook kind of held its shape, but it tasted awful. I mean, it really tasted like shortening. Okay. So the mathematical average did pretty well. The predictive tests cookie was kind of a disaster. And then then you moved onto the third which is really interesting call the neural network cookie. How do you do that? So a neural network is piece of AI am artificial intelligence. And so it it's a bit of software or an algorithm that takes a whole bunch of texts that I put in and it learns the sequence of letters. So just by giving it a list of recipes it learns how to format a recipe it learns that it always starts with the list of ingredients. It has a rough idea about how long the instruction should be about. How long the ingredient list should be? So it's it's uncanny. But the reason why it's so often wrong is that it doesn't have any kind of concept of what cookie is it. Doesn't know what the end product is. And so it has no problem telling you to add sugars four times, it has no sense of ratios of the proportions that things ought to be in. It'll constantly. Leave out key ingredients like eggs, it's it's secretive specis. So so this is supposed to be deep learning. But obviously of the three approaches this may have been the worst. It was certainly the least like cookie, I think it's a very compelling technology. But the important thing is that it doesn't have any sense of what a cookie ought to be or what ingredients need to be in the cookie to give it that true kookiness. It just doesn't get that. It's completely out of its sense of understanding of the world. So us three techniques the mathematical average, the predictive text the neural network, which are really trying to do. I think if I can pull away back is there's an argument about the brain, right? And whether the brain is just a computer, and that it some point in time, everything the brain does can be replicated by a computer other people would say, the brain is more than a computer, it can make leaps beyond just computations through some sort of. Creative process or the soul, or whatever you wanna call it. So big blue with chess because chess so highly regulated as a process was able to beat a human. I wonder if what you're trying to do with food maybe food there is more art to food than there is art chess. And it's not purely computational is that possible. Yeah. I think it's both. I do think that computers can be creative. But the sense of appreciating. What's been created by a computer is something that only a human can provide? So I think the computer's endlessly creative in that can give me a thousand new chocolate chip recipes. But the sense of value is something that I impose as the human to say, this is, you know year. Sure, you can make me thousand you can make ten thousand chocolate chip cookie recipes you can do that in a minute. But only some of them are worth my time. A little bit better about my choice of profession at this point. L? Thank you so much. I I've I've learned a little bit about science and predictive technologies, and could you send me your recipe for the average shocker chip cookie. We'll try to straighten Alicia. Yeah. Thanks so much. That was computational scientists l O'Brien or article for the pudding is called baking the most average chocolate chip cookie..

Coming up next