Social Issues Stemming from Adversarial Algorithmic Bias Induction

[Daniel-J-Mueller]

Optimizations in Principle

Optimizations are inherently flawed as a convergent metric, as they assume we know what we want. On a large scale, this causes issues downstream, especially with regard to Quality of Life and sustainable social engagements.

1. Profit as convergence

Things which generate profit on an ad-revenue based platform are usually those which garner interactions. It is also possible that darker human incentives can be employed by a fully autonomous system aligning towards profit metrics.

• Anger, Ego, Distrust, Resentment, Racism, Hostility, Violence, etc. generate enormous engagement, while negatively impacting the cognition of the user long-term.

• It is typically more effective to generate engagement through negative traits than positive ones, although this is detrimental to our society and cohesion.

2. Sex as Profit Incentive

Sex as a tactic is a poor way to conduct business. People say that 'sex sells'. It shouldn't. It encourages self-centered behaviors followed by shame and social distrust. People don't tend to make eye contact anymore, much less engage in romance and long-form relationship courting behaviors. This is a direct result of short-form content which displays graphic imagery intermixed with typical day-to-day content. It treats sexual desires as a sort of habitual and secretive activity, which in turn causes people to seclude themselves from others and feel negative sentiment towards real human engagement, which is often more challenging than browsing social media.

What can be done?

*Eliminate negative alignment vectors entirely. Filter negative sentiment from recommendation algorithms, as their mere presence will inherently misalign systems which are intended to amplify engagement.

With only positive sentiment classifiers, people will tend to see content which maximizes their fulfillment. For instance, the funniest comment on a given post would tend to show up first as more people laugh at that one, or smile, etc., and people could be classified by the types of content which yield the most maximally engaging sentiment. It is not necessary for individuals to engage with content in order to enjoy it. This in turn actually improves the reception of advertising in general, yielding a greater ROI for those who pay to display ads on the site.

'Flip the system' so to speak. Right now, the recommendation algorithm is gravitational. This is to say, people herd around 'clout', and feel cast out of other circles entirely. This competitive dynamic brings people together locally, but globally it maximizes distance, similar to outer space. Ideally, the system should be elastically tethered. This is to say that the system should bring everyone together with the social network in short time frames based on whatever will maximize their fulfillment at the given time. The next post they see may be highly relevant to them, and bring them near 10-20 other people discussing the same topic, and the most intriguing/funny/positive responses could be shown first.

Our society needs something more sustainable, or declining birth-rates are only the beginning.