Embedded Files
In order to understand what AI is used for today and where it is heading, about the types of problems it solves, it helps thinking of "hard" and "soft" causes to problems.
Random images ...
Hard causes
Hard causes
These are absolute, root causes for problems.
Tesla is going private (or is it?), so the stock price rose.
.. but rabbits drive traffic.
Soft causes
Soft causes
These are intermediate causes that correlate with the problem.
Auto makers stock prices are on the rise. Yesterday's stocks rose. Battery prices are dropping. It is summer.
To rational creatures like us, hard causes are preferable because they offer certainty in the form of dependable explanations. The scientific method and our progress rests on finding hard causes for issues. Soft causes are frequently belittled: Why didn't you see this problem through? Why didn't you get to the bottom? There's got to be something behind it! Where is the grand narrative? What does it really mean?
When you add the messiness of the world to the puzzle, you'll quickly bump up against several factors that make looking for hard causes impractical:
Hard or soft problem, what do you think?
Hard or soft problem, what do you think?
Will a particular stock price go up?
What piece of music will lift my spirits?
Is there a voice in this audio sample?
What treatment should the patient receive?
Should the loan be granted?
Cost & Time
Cost & Time
Almost everything has economic limits, past which it becomes undesirable to look further.
Number of factors
Number of factors
There may be too many hard causes. The world is messy.
In the world of stock exchange prices, going back to hard causes might mean mapping the entire economy.
Specificity
Specificity
Being hard-cause creatures, we think there are similar causes for entire categories of problems. But what if the causes for each problem are different? Or the same hard causes but differently important for each problem?
This is where AI stands today: Largely resolving soft problems by creating myriads of patterns from data and then applying these patterns to new data. Which leads us to the hottest fields of AI work today:
Understanding AI's decisions: In certain areas, it is vital to understand how AI reached certain conclusions - in order to accept or challenge the decisions, to learn from AI or to enhance it. Ironically, other AIs are currently used to explain their brothers.
Higher-level pattern recognition: Telling cats and dogs apart is only a first step. Almost any other task that involves the real world, even routine ones like driving a car, require recognizing concepts far above "that is a bike".
Understanding humans: AI would be more useful if only we knew what we really wanted and how our minds, bodies, emotions etc. worked.
Page updated
Report abuse