> I think most would call the original pagerank algorithm unbiased.
I disagree: the point of the original Pagerank algorithm was its bias. In particular, its use of link weights gave the algorithm a strong bias for (a proxy for) relevance.
An "unbiased" search algorithm would need to return results based on simple criteria, readily explained and presented to the searcher, such as "date first crawled" or "number of keyword appearances."
You are trivializing every bit of research and advancement gone into improving search engines in the last 20 years.
Please attempt a simple implementation with the methods you describe and see how it performs over even a non trivial amount of data and you will see the difference in results.
And I'm not even considering malicious results gaming the whole system.
> You are trivializing every bit of research and advancement gone into improving search engines in the last 20 years.
I disagree. If anything, I think I give this research more credit, because I appreciate it as imposing a normative view on what makes search results "good."
This is more than just "how do we sort a list containing billions of keys" or "how do we grab data from a distributed database;" it's also "how do we understand relevance or quality based on machine-interpretable signals?" The latter is a judgment call.
This is a conflation between the meanings of the word bias as “a weight” and “an ideological stance”. The thing that people take issue with is generally the latter, rather than the former.
It is a conflation, but I argue that there's no bright line between the two. The ideology behind pagerank is that relevant results are better and that backlink popularity measures relevance. Another sorting objective is that "more engaging results" are better.
Each is ultimately an expression of will, since there is no objective ranking of what makes something relevant or engaging. Everyone will agree on the order of names in a phone book, but nobody will agree on the "relevance" order of search results.
You see this every day in news-adjacent search topics, where popularity, relevance, and factual correctness can all give different "proper" orderings.
Ultimately, people dislike stances that give disagreeable results, but they think benign stances to be invisible. I don't believe a coherent, legal argument can work this way.
I agree with this line of argument in the sense that a choice of loss function is still a choice with implications that one cannot escape. All choices imply the expression of will and the exercise of power.
I think the essential disagreement isn’t so much about the choice of bias as it is about the exercise of power, specifically on who’s behalf do you exercise power? Do you do it for your users or for the greater good? Or for your customers? Or maybe your investors? Or maybe your ideals?
I disagree: the point of the original Pagerank algorithm was its bias. In particular, its use of link weights gave the algorithm a strong bias for (a proxy for) relevance.
An "unbiased" search algorithm would need to return results based on simple criteria, readily explained and presented to the searcher, such as "date first crawled" or "number of keyword appearances."