Hello and welcome to Backchat. If the Nature podcast is a nice normal bell curve, then Backchat would catastrophically fail the normality test. This month, statistics, mathematics and intelligent algorithms. We'll be discussing misused P-values, getting the gossip on the latest success of Google's AI AlphaGo, and watching mathematicians win prizes, which is awkward.

I'm Kerry Smith, and I'm pleased to introduce my mathletes and stats superstars, Davidei.

Hello, I'm Davidei. I'm the Math Prize correspondent.

Lizzie Gibney also joins us.

Hello, I write about physical sciences and all different kinds of forms from here in London.

And special guest, Monia Baker, visiting the London office all the way from San Francisco.

Hello, I wrote about how scientists can make science more reliable.

Now, coming up this month, all of the glamorous topics, as we've hinted at already, starting with statistics,

specifically, P values, at their best rigorous evidence, at worst, wonky and

misleading and misinterpreted.

Monnier, as you mentioned, you're our kind of stats correspondent of late.

The problem here is that scientists are sometimes not using this kind of pillar of statistics

properly, right?

That's right.

They're trusting it way too much, and they think that it can say

whether or not their hypothesis is correct, which is not something that the P value can do. What the

P value can do is tell you how likely a certain set of values is to occur by chance, assuming that

there's no correlation between

various values and a variety of other assumptions are true.

Now, this month, it's become news because the American Statistical Association, which is a very

old and dignified body of concerned statisticians, have sort of weighed in on how people are

abusing the P value.

...