Features

In a recent Datawocky blog post, Anand Rajaraman quotes Peter Norvig, Director of Research at Google, that Google still uses the manually-crafted formula for its search results and did not switch to the machine learned model yet.

Anand writes

This is quite surprising, since Google has a ton of data and the problem of ranking search results is very suitable for machine learning. Google in fact has a very good machine learning algorithm for this task, which it runs in parallel, which gets very similar results, but is not fully trusted.

One reason for the lack of trust is that Google team worries that machine learning algorithm may be susceptible to catastrophic errors on searches it has not seen.

Anand writes:

If Google is unwilling to trust machine-learned models for ranking search results, can we ever trust such models for more critical things, such as flying an airplane, driving a car, or algorithmic stock market trading?

The KDnuggets poll question is:

Can machine learning algorithms ever be fully trusted on real-world problems?

Please vote on www.kdnuggets.com and choose from

• No, never
• Yes, but only under some conditions ... (which ones? please comment)
• Yes, eventually for all domains
• Don't know