A lot is being made out these days for the re-emergence of AI. While some have expressed fears about machines taking over human functions, many use AI or machine learning to provide the cutting edge in terms of insight and to enhance our perception of the industry we are in and to maximize profit, user experience, customer satisfaction and so on.
Firstly, AI and machine learning are pretty much not the same thing.
The ideal of AI would be to create a similar or superior form of human intelligence that one would trust. A pure form of intelligence. Somewhat like how you would trust your mentor, spiritual guru, favorite expert or your grandma.
Now, that would be really surprising if it could happen. The thought that a machine created by humans would not have commercial motives (sponsored links) or motivated agenda (Foxnews vs MSNBC, anyone?) is unthinkable, however briefly it may work in the "pure intelligence" mode.
The evidence is that we are currently only looking at AI to augment our own intelligence. Our quantitative and qualitative understanding through AI and machine learning is from the self same models and data provided by us.
We ask a few questions and use machines to crunch out the answers. And how we do it is by feeding large amounts of data and correlate all the patterns to get some seemingly intelligent answers.
Now where is this data coming from?
Public sources of information - which clearly are not very reliable
Collections of data owned by various corporations, governments and other entities - which may be really expensive, confidential or inaccessible
Public sources of Information
As long as the data is from public sources, like Google we will have to look out for trolls, fake infomercial websites, separate the wheat from the chaff. There is an industry out there to make money from rumors, or false or highly speculative information. Example: Financial and commodity markets.
Proprietary data sources
If data is to be collected from various authentic sources, there are just too many and they cost money, if they could be bought at all. Even efforts like arXiv are not enough to gather authentic data.
Of course you can rely on people like Ed Snowden to arise periodically to leak some "authentic" information.
Authentic information is expensive or inaccessible
Hypothetically speaking, authentic data in gigabyte terms is probably more expensive than equivalent amount of gold in tons if accessible at all.
So, to build a real AI would take a magnitude more of capital and a fundamental change in how we reveal and exchange authentic information.
Till then we can apply ourselves to better machine learning algorithms to provide us localized insights using accessible and affordable data.