Skip to main content



Effects of Optimization of Preference Models on Innovation in Society

Nowadays with ML and analytics we optimize for metrics positively correlated with peoples preference models. Does this sort of trap us in a bubble of our preferences, where we only see content we’d be interested in or enjoy?

What kind of impact does this have on innovation and the creation of new ideas? If we look towards academia we see staunch central focus on a specific idea or problem for years to end with countless renowned scientists coming through with renowned breakthroughs yet in industry we see just as many life altering creations from a diversity of disciplines and backgrounds.
Which is better for society? Companies may try to predict new content you’d prefer but would they really risk the large revenue for low possible return?
We can think of information cascades as viral content or mass adoption of a product or technology, information consumption and production through diffusion of ideas as networks of agents connecting in transactions, and the optimization towards user preference models as matching markets.

This seems ok if we think of this as just a reflection of our preferences, but if one considers the collective knowledge pool is it fine to avoid ideas we don’t like?

What exactly constitutes our network? Is it just our friends, our families, our neighbors, our local community, the world?

Does society or the general collective benefit more when a diverse group with many different backgrounds outputs something or when someone with a very strong background in one specific field contributes something?

In finance they define synergy as an increase in monetary value, of a company after it merges with or acquires another company. In economics they discuss the way that consumer producer exchanges interacting in a market with influences such as restrictions(price floor/roof) quotas or subsidies/taxes. In this case we should formalize the problem to companies using metrics as regulating these matching markets and formulate appropriate equations. Through earnings reports, approximate usage etc you can benchmark the value provided to companies through these interactions and essentially analyze which kind of exchange might result in higher output, you can look at specific cases of companies that were founded in industry and analyze case study style their teams and backgrounds and the companies growth vs academic projects in universities and industry scientists.

Then if there is some relationship, then you can look at a younger age group to see if it still holds for university students. Compare success of undergraduate research vs project teams or hackathon teams for another dataset to look at based on people with significantly less experience and knowledge and see which does better based on your metrics.

It would be interesting to see which of these results in more consistent high value outputs.

Comments

Leave a Reply

You must be logged in to post a comment.