Is the Truth really out there? or Representing Uncertainty
I had a very interesting lecture today on Artificial Intelligence, I’d like to share a few ideas from that lecture that inspired me. They were all about uncertainty and started out with two terms that seem to be the foundation of uncertainty in all endeavors:
Either uncertainty caused by ignorance, or by laziness
Ignorance is coupled with not knowing because you simply don’t have a way to get the knowledge. The winning numbers of the lottery is one such event where you’re simply always going to be ignorant of the truth. And if you somehow end up certain about a future lottery, please do drop a line…
On the other side of this coin is laziness or in other words getting certainty is just to much work. An example I got in lecture was centered around pathology. Lets say you’d be able to reach certainty in a medical diagnosis but you’d need to perform 2000 tests. On the other hand you could be 90% certain with just 20 tests. Now those numbers are fictional of course but you get the idea.
Uncertainty doesn’t change anything about the truth values. Statements are always true or false, it’s the agent who isn’t certain.
Read that line again please and realise what it means: truth by it’s very nature is complete, we (agents) are the ones who distort it, doubt it and often even reject it. This -does not- change the truth however, just our perception of it.
So far it’s black and white, now a small sidestep into fuzzy logic. A logic that allows for an infinite number of grayish colors. Take the statement “This car is fast”, while it’s certainly true for an Austin Martin, it’s “more true” for a Koenigsegg. Gradual truth if you want to think about it that way. By themselves both statements are true but one seems more true than the other now doesn’t it?
It’s a very down and dirty simplification of the truth about uncertainty but at least it provides a fresh perspective on how we think. After all, artificial intelligence tries (among other things) to make a machine act like a human would in the same situation.
Handling uncertainty comes down to a few pointers:
- Don’t be lazy (Do the 2000 tests)
- Keep the cost-benefit trade-off in mind though
- We can’t solve the ignorance problem. Relax about that
Asking yourself questions about what you think is always a good practice I might add. Uncertainty is one of those subjects that really benefits from regular scrutiny. A few questions that might help you here are:
- How uncertain are you at any given point?
- How certain do you want to be at that point?
- How does the uncertainty increase and decrease?
Last but not least there is always the trade-off between Utility and probability: How strongly do we want/need some result (utility) and how certain are we to get it? (probability) A probability of 100% although very nice is very hard to come by. Is it always necessary to reach this value? I sure don’t hope so, imagine the costs that would bring, and is the benefit really that big? Also imagine the stress it would bring if we always wanted the truth and would not accept any amount of uncertainty.
It’s always good to want less uncertainty and seeing as the truth is out there we can choose to go look for it. Don’t be lazy, accept ignorance and above all don’t worry if you don’t have perfect certainty yet.
If it’s not ignorance that’s stopping you you’ll get the truth in the end if you are prepared to make the costs.
Happy hunting
