My daughter just turned 9. A few months ago, she asked me to uninstall the TikTok app from her mobile phone. TikTok started showing her videos of animal abuse, probably because its recommendation algorithm thought it would increase her viewing time.
It doesn’t sound very good, but other social networks (Facebook, Twitter, LinkedIn, etc.) also do this: they show us what they think will increase our viewing time. When you think about it, even content recommendation systems (such as Tabula and Outbrain) show us content we are more likely to click on, even if it doesn’t serve us well.
Recommendations are driven by a wide family of algorithms, generally called “machine learning algorithms”. Social networks aim to gain our attention and increase viewing time. Since attention and viewing time, equal more ads, and more ads are more money. This is an over-simplification of the situation, but not entirely wrong.
Sometimes recommendations can go wrong, like showing a young user abusive videos, in many other ways, e.g., Twitter keeps showing me dating stories of single women, and I’m not sure why.
Machine learning algorithms are customized in a process called “model training” and follow the guidelines of a mathematical model with an objective function. The objective function is a mathematical representation of the goal. Most commonly, maximizing prediction accuracy (but might also be maximizing retention, viewing time, or other KPIs).
ChatGPT and other generative AI models are also a part of this family. They, too, have an objective function. They have been trained by teams of labelers/responders who answer questions. The algorithms maximize the “fitness of the answer” or the score of possible answers and make a choice. This makes it prone to other issues (like the Chat being 100% certain in completely nonsensical answers simply because they seem right).
When the objective function is money, and the span is virtually unlimited, the social network is designed by machine learning algorithms to make us spend our time within the network. Sometimes this is positive and useful, but a lot of the time, this has a lot of negative effects (e.g., influence on young minds, waste of time, polarization within the population, and more).
Social networks and machine learning algorithms can be used for good: for contact tracing in epidemiology, education, and more. But this depends on the objective that was engineered into the algorithm and also depends on the essence of the network.
We are social beings in nature, and we have different kinds of objective functions. When we are the ones behind the wheel (instead of machine learning algorithms), what are we optimizing? what is our objective function? (why are you here?)
Most likely – you’re interested in meeting new people, having a good time with new or old friends, making new friends, maybe getting some good business opportunities, and having a good meal. Maybe a combination of these.
Why am I telling you this?
- Apply critical thinking when machine learning algorithms are behind the wheel.
- We should all be aware that strong statistical manipulation is underway everywhere we look and is expected to grow even further.
- Take the time to get to know real people in new environments (and make sure your kids do the same!)
- Educate and explain to the younger generation how to use the available resources, how to apply critical thinking, and how to step out of the virtual world once in a while.