An article written by Mitch De Felice from CIO – “Big Data and Machine Learning – is the glass half empty?” which appeared also on Bloomberg for Enterprise have made some great points on why big data is having a difficult time delivering on the promise of rich discoveries. The author opened the topic of Artificial Intelligence by saying –
Today, the focus is on machine learning and statistical algorithms.
To understand the significance of today’s Ai focus it is important to know what Ai was. As the author pointed out- AI is different today from 20 years ago. He said:
AI 20 years ago was focused on what is known as logic based AI or Knowledge Representation (KR).
However, back in the early 90s, the tools and frameworks to make KR successful had never materialised until recently. A plain wiki definition on KR is the following –
Knowledge representation and reasoning (KR) is the field of artificial intelligence (AI) dedicated to representing information about the world in a form that a computer system can utilize to solve complex tasks such as diagnosing a medical condition or having a dialog in a natural language.
Examples of knowledge representation formalisms include semantic nets, systems architecture, Frames, Rules, and ontologies. Examples of automated reasoning engines include inference engines, theorem provers, and classifiers.
Knowing what Ai was 20 years ago, the next question becomes – what Ai is now? Do we have the tools and frameworks to make KR successful today?
Big data and machine learning could provide some tools and frameworks to make KR successful – at least in part. The author illustrated the capabilities of Ai from data usage perspective – highlighting the role of big data and machine learning in making KR successful –
From a practitioner’s perspective, most machine learning we do are supervised machine learning which has existed, according to the author, since the 1950s. 🙂
You might wonder – what exactly is Supervised / Unsupervised Machine Learning? A quick internet search returned in a PDF tutorial by Zoubin Ghahramani from Gatsby Computational Neuroscience Unit, University College London, UK, a total of
Four Types of machine learning
- Supervised learning – Consider a machine (or living organism) which receives some sequence of inputs x1, x2, x3, . . ., where xt is the sensory input at time t. This input, which we will often call the data, could correspond to an image on the retina, the pixels in a camera, or a sound waveform. It could also correspond to less obviously sensory data, for example the words in a news story, or the list of items in a supermarket shopping basket. In Supervised learning, the machine1 is also given a sequence of desired outputs y1, y2, . . . , and the goal of the machine is to learn to produce the correct output given a new input. This output could be a class label (in classification) or a real number (in regression).
- Reinforcement learning – the machine interacts with its environment by producing actions a1, a2, . . .. These actions affect the state of the environment, which in turn results in the machine receiving some scalar rewards (or punishments) r1, r2, . . .. The goal of the machine is to learn to act in a way that maximises the future rewards it receives (or minimises the punishments) over its lifetime. Reinforcement learning is closely related to the fields of decision theory (in statistics and management science), and control theory (in engineering). The fundamental problems studied in these fields are often formally equivalent, and the solutions are the same, although different aspects of problem and solution are usually emphasised.
- The third kind of machine learning is closely related to game theory and generalises reinforcement learning. Here again the machine gets inputs, produces actions, and receives rewards. However, the environment the machine interacts with is not some static world, but rather it can contain other machines which can also sense, act, receive rewards, and learn. Thus the goal of the machine is to act so as to maximise rewards in light of the other machines’ current and future actions. Although there is a great deal of work in game theory for simple systems, the dynamic case with multiple adapting machines remains an active and challenging area of research.
- Unsupervised learning – the machine simply receives inputs x1, x2, . . ., but obtains neither supervised target outputs, nor rewards from its environment. It may seem somewhat mysterious to imagine what the machine could possibly learn given that it doesn’t get any feedback from its environment. However, it is possible to develop of formal framework for unsupervised learning based on the notion that the machine’s goal is to build representations of the input that can be used for decision making, predicting future inputs, efficiently communicating the inputs to another machine, etc. In a sense, unsupervised learning can be thought of as finding patterns in the data above and beyond what would be considered pure unstructured noise. Two very simple classic examples of unsupervised learning are clustering and dimensionality reduction.
If you are interested to go deeper on unsupervised learning, please check out the tutorial linked above.
Jumping back to the CIO/Bloomberg article, the greatest challenge facing Ai today is probably the Cognitive Computing that links Knowledge Representation and Machine Learning. The lack of understanding of concepts by machines is hindering their ability to perform Logic Reasoning and thus Cognitive Computing.
On the machine learning side, there is a limited “knowledge” that unsupervised learning (eg. clustering and dimensionality reduction) can deliver – as the author has put it –
The best big data can offer is the ability to classify, falling short on being able to provide any contextualization capabilities, which is the real value to the business.
So the outlook of handling unstructured data explosion and developing successful Ai solution looks daunting
The reality is that technology is shifting faster than leadership can understand what threats it poses. Large bureaucratic companies that simply adopt the big data paradigm will struggle to keep up with explosion of unstructured data and data streams that has caused the role and value of information to drastically change.
One thing I can predict here is – by the time when cognitive computing fully materialise, the only thing humans could do against machines might just be – Pulling the Plug. 🙂