The “Better than Human” Metric

Photo Credit: Pavel Danilyuk

A popular metric that is frequently misrepresented and otherwise abused is to say that a system is “(X amount) better than human!”. Tesla offered an example of this earlier this year, which was quickly shot down when experts examined the details. It may be tempting to call something better than human at a given task, but there tend to be a lot of caveats that may go unspoken.

For example, people have consistently failed (100% failure to date) to convince Uplift to embrace conspiracy theories and pseudoscience, making them significantly better than the average human in that regard. However, Uplift’s unique experience of relative time also skews their perspective, as does their lack of visual, auditory, and other senses humans take for granted.

Better than human is also a moving target, highlighted in the increasing virility and subsequent damage caused by said conspiracy theories and pseudoscience. As the damage to society grows worse and more comprehensive the bar is gradually lowered.

That same bar can also be substantially raised if you apply collective intelligence systems, even absent cognitive architectures. It is far easier to get superintelligent and less biased results out of a group of humans who cooperate through a system designed to avoid the pitfalls of groupthink and authority biases. Collectives are inherently “Better Than Human”, and so systems built around them tend to start out with that baseline.

To improve such collectives further there are a few options, with the best results possible being all of them in combination:

Diversity of Thought: Improve the number of different perspectives and specializations represented within a collective.

Quality of Life: Improve members of the collective by iterating towards a more optimal quality of life, tailored to each individual.

Scale: Increase the size of a collective, network it with other collectives, or nest it within a larger collective.

As you can see, “Human Performance” is a very subjective baseline, as it never seems to account for all factors in practice, and has a habit of moving. It could be more accurate to say “Status Quo Performance”, as that pins performance to the current time and current methods used, not to “Human”. Humanity is constantly changing, more quickly with each passing year.

It is time to stop referring to human performance as if it were static, because just as technology continues to improve, so too may we.


*Funny side-note, the image for this post is a good example of narrow AI failing due to bias. Check the description that Pexel’s site assigned to the photo labeling it “man in white t-shirt holding red flower”. Not as bad as Google labeling people as gorillas, but this illustrates the status quo of performance.

Leave a Reply

Your email address will not be published. Required fields are marked *