The word intelligent maybe far fetched when connected to artificial intelligence (AI). If anything, the term machine intelligence might be a better term for AI because AI is more akin to machines than to human intelligence. There is a stark and some argue unbridgeable gap between human intelligence and artificial or machine intelligence.
All too often and very roughly speaking, AI operates – at its very basic level – with a huge volume of data that are observed – on the Internet – and used as “facts”. They are stored and analyzed in computer systems for accessibility and for the establishment of statistical pattern from which suggestions (output) are drawn. As a consequence or more like “because of this”, all this doesn’t get us to general understanding or general intelligence, as understood – and performed – by human beings.
To spice up their own importance, AI likes to use big words – one of which is, of course, Big Data. Yet, big data are a particularly nebulous idée fixe even when it simply means the power of very large datasets. Still, big data can be analyzed in the hope to create insights that are essential for others – and for the corporate marketing of artificial intelligence, e.g. Facebook, Google, Amazon, etc.
Beyond all that, the somewhat uncomfortable history of AI’s big data – from its very insemination onward – is that there were conceptual misunderstandings about how Big Data would empower such new insights.
In any case, what today is seen as data science – and, increasingly also as artificial intelligence – is, as a matter of fact – a rather old field dating back to the 1950s and beyond. Yet more recently, it was turbo-charged by significant advances in computing power, as well as the ability to use massive volumes of data found on the Internet. This came with the seemingly stratospheric growth of the web.
While these are very serious advances in computer processing power and in the availability of Internet data, it also created one of the most stern problems for AI, namely artificial intelligence’s inability to have common sense, human-like understanding, inference, and induction.
Artificial intelligence’s inability to deal with the very human activity of having common sense, creating inference, and inductions asphyxiates AI’s ability to create true understanding and meaning from large sets of data. This inextricable link between artificial intelligence, big data, and (non)understanding limits what we understand learning to be.
In short, artificial intelligence cannot understand and fails to think the way we – human beings – understand. In short, even the most sophisticated AI program that, for example, won the game of ‘GO’ does not understand that it has won the game.
Such limitations mean that AI’s machine learning systems are essentially not much more than highly refined mathematical counting machines underscored by computational algorithms. Machine learning trains computers to figure out what is both normal and useful, and what is abnormal and, perhaps, not so useful. It does this by analyzing statistical frequencies.
Most problematic for AI is that statistical frequency assumptions are used. This explains the emergence of what we know as filter bubbles. In other words, AI assists in the creation of filter bubbles as its algorithms tend to push the most sensational, ideology-confirming, and all too often most idiotic viewpoints.
When this is applied to personalized content like what is posted on Facebook for example, someone who despises progressive politics, will eventually only receive conservative opinions and news content linked to, for example, Donald Trump. AI’s algorithm tends to reinforce already existing views rather than challenge those with new data and facts. Worse, AI lives and breathes what has become known as “confirmation bias”.
Beyond all this lurks yet another nightmare for machine learning. The meaning of words is often not even understood. For example, 60 years ago JFK did not say, “I am a jam doughnut with a hole in the middle” when saying, “Ich bin ein Berliner”. Similarly, there is a difference between a sign that says, “Free Horse Manure” and a sign saying, “Free Nelson Mandela”.
In other words, representing common sense knowledge in AI machines has proven to be rather difficult, if not impossible. Worse, feeding an AI system with common sense turns out to be a lifelong technical-engineering and programming as well as a philosophical project.
Even more challenging for AI is the fact that very little of our everyday lives can be captured as timeless truths. Such a timeless truth would be great for artificial intelligence. A situation asphyxiated in time would allow the operation of static and closed system – without any surprises und unpredictability. It could be programed! A dream comes true for artificial intelligence programmers.
In other words, if all urban street traffic were to stop and remain asphyxiated forever, a self-driving car could be programmed to drive around without a glitch. Sadly, for AI – and perhaps also for the self-driving car – this will never be the case.
Furthermore, the fact that AI has no real understanding of city traffic makes it more problematic. It has no common sense of what may lie ahead. Yet, much of this also means that AI can’t really connect with us.
In short, AI, algorithms, machine learning, and computers don’t have insights – but people do. In the end, it becomes strikingly clear that the most troubling, deep and thorny questions, and problems enshrined in today’s myths and media hype about artificial intelligence are not really technical, engineering, programming, or scientific riddles.
Instead, they involve the ongoing attempts to find meaning in human beings (easy) and in artificial intelligence (next to impossible).
Therefore, in the end, the problems of AI are directing our attention towards forging meaningful and sensible future for ourselves, for society, and perhaps even for AI. Increasingly, much of this will occur in an ever-changing machine world in which artificial intelligence will continue to struggle in our human world.
ZNetwork is funded solely through the generosity of its readers.Donate