Be carefull about your objective

I’ve just been watching a really great lecture called The Long-Term Future of (Artificial) Intelligence by Stuart Russell and I think there’s a link to be made with the actions of governments across the world in reaction to Covid-19. Russell talks about the problem of not properly specifying, or misaligning, end goals when it comes to generally intelligent AI systems, something that has been widely discussed by Nick Bostrom and many others.

For example, if a super intelligent AIs only goal is to maximise paperclip production there’s a real risk of it doing so by destroying the whole of the planet if this is its one and only goal. If you fancy trying this out have a go at a game called Universal Paperclips, devised by Frank Lantz at NYU.

Such ideas go all the way back to Greek mythology and morality tales such as King Midas (be careful what you wish for). If an omnipotent power is focussed on one thing only and does not place this in context (or specify what else it wants, or does not want, to occur) there is a real danger of creating a big hole that’s difficult to escape from.

A link here is perhaps not being clear enough about what the goal with Covid-19 is, or else misaligning (or blindly not seeing) the objective in the context of broader, longer-term goals. We must think about priorities, but also the context in which such priorities exist.

If the aim is to defeat Covid-19 at all costs (unlikely as we’ve only ever totally eradicated two other viruses in human history – Smallpox and Rinderpest) then there will be substantial collateral damage. Maximising this outcome will inevitably impact other important things. Trade-offs might include substantial numbers of people dying from other perfectly preventable conditions (because they are either afraid or unable to access proper treatment), substantial economic damage, loss of jobs, loss of proper schooling, and potentially a mental health epidemic.

Perhaps, rather than trying to win a war against Covid-19 (or trying to stop anyone whatsoever from dying from it) the objective should have been to work out how to live with it, how to preserve a way of life for as many people as possible. If this were the aim then maybe countries should perhaps have vaccinated younger people first, not last, or vaccinated essential workers ahead of anyone else. And maybe older people that feel happy with isolation (and many, it should be remembered, don’t) should be able to gift their vaccination to someone else too.