Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by RA_Fisher | Hacker News Reader

Parent Full thread

RA_Fisher·

I enjoyed the article but I see two overlooked aspects that significantly weaken the author's argument:

1. Statisticians are substitutes for data. You don't necessarily need new / more data if you have a statistician.

2. Data often contains a lot of redundant information. Big data may simply be duplicative.

View on HN

Could you elaborate more on the first point? I don't really get it

Humans can often learn new things from many fewer data points than current ML algorithms. Sometimes a single data point suffices.

Human statisticians can apply a variety of mathematical tools to fit different situations. ML systems tend to be more like one-trick ponies.

I’m not sure I could agree with that statistician can “learning new things” with smaller data points. Statisticans might come and see the pattern with “better” prior than ML models, that allow them to come up with better conclusion. However, given the same dataset, there is a maximum to thee information that can be extracted from the dataset. Ideally, Any human(think of human brain as a pattern recognizer) or appropirate statistical method would come up with the same information from the datast, given no prior

Amazingly there is no upper limit on information that can be extracted. :-) Also, by applying increasingly sophisticated techniques, statisticians can extract increasing amounts of information from the same data. Don't have data? Just add a statistician.

I’m not sure if you’re trolling or not, cus there is definitely an upper limit to how information can be encoded in data.

Reply on news.ycombinator.com