Instrumental conversion: unintended side-effects of AI

Much has been said about the existential threat that artificial intelligence may pose to humans. Most experts claim that this would need machines to be self-aware and that self-awareness is still a long way off.

However, it’s plausible that intelligent algorithms unintentionally harm humankind, in pursuit otherwise harmless final goals. That hypothetical concept even has a name. It is called instrumental conversion. This is how it works.

If we give sufficiently smart algorithms a task and few restrictions on how to complete this task, it may find work-arounds that end up being harmful. There would be no evil intent, but the harm to humans would be a side-effect of the steps it takes to finalize a task.

In other words: such intelligent algorithms, designed to complete certain tasks, without given sufficient boundaries, may become an existential risk to humans, via the instrumental goals it pursues to reach that harmless final goal.

What is an instrumental goal?

Money is a good example of an instrumental goal. Most people don’t want to have money for the sake of owning money, but would still pursue money as it would enable them to achieve other goals.

So you would not work hard for the money itself, but for what it would enable you to do (for instance, owning your own house, taking time off to write book, etcetera). Money is instrumental to achieve your final goals.

Instrumental conversion

Oops, missing a gear.

Self-aware killer machines

Self-awareness is relevant because without self-awareness machines themselves would not decide to go to battle with humankind.

Men may direct machines to kill humans, but, but at present, even very smart machines need tasks and instructions how to achieve those tasks. They just follow orders, so to speak.

Nobody really knows if machines will ever gain self-awareness, let alone how long that would take.

What we do know is that only a few decades ago, few people knew how to use a desktop PC and even fewer owned one. You could write articles on them and play simple computer games. Internet access for the masses was not on the horizon. The level of computing power that is compacted in our smartphones was inconceivable to most experts.

Yet, here we are. A few decades later these little machines tell us how to navigate, they contain our agenda and can even measure our heart rates. They can recognize our face and ask us what we need.

So, the mere fact that we cannot imagine machines to gain self-awareness in our lifetime or within a decade, does not mean it will not happen.