Machine learning - who checks what they're taught?

In doing some research on Artificial Intelligence (AI) I came across an interesting article on the mathematical foundation for noise, bias and variance in neural networks, a foundation of machine learning. Whilst most of the mathematics in the article was way beyond my 1990 GCSE syllabus it got me thinking about oversight of AI developments.

Articles on AI, the rise of machines and dangers they may or may not pose in the future seem to be the flavour of the moment but is a Humans synth type scenario really a threat?

Given machine learning is essentially the use of algorithms for computational learning by using data sets surely human bias or deliberate misuse is a bigger cause for concern in the immediate future?

How do we know whether the data being used to train these decision making tools isn't biased towards a particular outcome or group of people when it shouldn't be or whether the algorithms used are correct?

This article by Jerry Fishenden raised the question of regulation in the digital economy, citing the recent VW emissions scandal as a wilful misuse of technology. Surely this is needed now more than ever, 

Similar points are raised in a TechCruch post but with an emphasis on corporate accountability.

These articles illustrate that it's not acceptable just to rely on what happens inside the black box of a device or piece of software because, "after all, computers just do what we ask them to don't they".

We need to go further though if we're relying on machines to learn. If they're taught with skewed lessons then we have no reason to believe the outcomes will be what's expected.

The measures needed cross regulators and corporates and importantly some form of ethical dimension needs to be added if we're expecting machines to function autonomously - even then it won't be fool proof.