TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

'Eureka machine' works out laws of nature

16 点作者 echair大约 16 年前

3 条评论

ovi256大约 16 年前
I can see how a machine like this can help in situations where we have lots of data and plenty of alternative hypotheses to test, like for example the work the Adam robot is doing in molecular biology and genomics : it sifts through dozens of known genes to find the one that is responsible for a certain enzyme. This is very keen to a large search in a huge search-space. However, it still needs fitness criteria to be defined by humans - in Adam's case, how does it know the gene is responsible for the enzyme ? The experiment to assert that was designed by humans.<p>Furthermore, there are sciences like neurobiology where we have lots of disparate data and due to increasing sensor resolution, we are getting reams more. What is needed are theories to explain it, and thus shed lights on the underlying mechanisms. This kind of unifying theory is much more complex than newtonian mechanics - with no offense, I hope, to the great Sir Isaac Newton.<p>Furthermore, the Eureka machine was already studying the right experimet which generated the right data - from which Newton's laws were obvious. Sir Isaac Newton had to decide, among many other things, which was the right phenomenon to consider, and from which angle - and all this against huge cultural biases against "simplistic" and "mechanical" laws that might restrain God's power.
sh1mmer大约 16 年前
It's very interesting, but I dislike how the article refers to a "thinking machine". The computer is finding a best match algorithm from a search space in an experiment designed by people.
paraschopra大约 16 年前
"When fed information on the mass of the apple and its velocity as it falls, the machine would be able to work out the equation."<p>Isn't it plain regression or other prediction tasks. More important that finding out relations is the ability to look at the right places for data. If you feed data to any prediction algorithm (ANN, GP, Regression, etc.), it is sure to come up with an equation.