Cognitive computing has revolutionised businesses and industries globally. There are now many cognitive computing services on offer, including IBM Watson Products, Microsoft Azure Cognitive Services, Google Cloud Platform, Amazon AWS, AppleCore ML, and more.
Although many startups and SMEs have yet to incorporate this powerful technology into their businesses, most, if not all, large organisations are streamlining and automating processors, improving accuracy and efficiency, and improving customer service and productivity with cognitive computing.
But what is it?
Well, let us take a deeper look at just one of the cognitive services, IBM Watson, to better understand this powerful new technology.
IBM Watson – Jeopardy
In 2010, IBM researchers reported on their three-year journey to create a cognitive computer system capable of competing and winning on the television game show Jeopardy.
In 2011, Watson competed against two of the best competitors that have ever been on the show, Ken Jennings and Brad Rutter.
Jeopardy is a TV show based on natural human language. As a result, this posed a real challenge for the developers of Watson and Doctor John Kelly, who led the project.
Firstly, what is Artificial Intelligence (AI)?
Before we go on, let us take a moment to consider what artificial intelligence is.
Dan W. Patterson defined AI as a discipline that is concerned with the study and creation of computer systems that exhibit some form of intelligence.
Pamela McCorduck, in her book ‘Machines Who Think’ (1979), said that AI began with an ancient wish to forge the gods.
In relation to IBM Watson, IBM prefers the term “augmented intelligence,” and this is because, firstly, artificial intelligence carries some negative connotations for humanity’s fears of the singularity.
The movies often depict this as “the day the robots take over the world.”
Furthermore, AI and IBM Watson simply extend our cognitive abilities and enhance our reasoning capabilities.
Therefore, augmented intelligence is likely a better description of IBM Watson and AI more generally.
Either way, AI is a very broad term. Here, in this diagram, you can see the different disciplines or subsets within artificial intelligence. The ones that we are concerned with for cognitive computing include machine learning or deep learning, NLP (natural language processing), and also vision and speech.
Collectively, these subsets are often referred to as “Cognitive Computing”.
Artificial Intelligence Fields of Research
Cognitive Computing
Cognitive computing is
“a term favoured by IBM, cognitive computing applies knowledge from cognitive science to build an architecture of multiple AI subsystems – including machine learning, natural language processing (NLP), vision, and human-computer interaction – to simulate human thought processes with the aim of making high level decisions in complex situations. According to IBM, the goal is to help humans make better decisions, not make decisions for them.”
Enterprise AI
In simple terms, IBM Watson is a tool to understand complex language based problems.
How Is Cognitive Computing Different From Traditional Computing?
More traditional computers (aka 4th generation computers) are great at mathematical logic and computation. They are fantastic at processing mathematical operations at an unbelievable speed, way beyond that of a human being’s capability.
However, tasks that humans find very easy, such as understanding natural language or visual or auditory information, are impossible for 4th generation computers.
This is where cognitive computing is required.
Machine Learning
So, let us look at the subset of AI called machine learning. There are many types of machine learning algorithms, including K-means clustering, logic regression, linear regression, support vectors and more.
For most people, though, it is not important (unless you are an AI engineer or scientist) to understand exactly how these machine learning algorithms work, but essentially, it is just very clever math!
However, the element of machine learning that IBM Watson is most concerned with, is artificial neural networks (ANN), also referred to as deep learning when the network has more than 3 layers.
In essence, ANNs are high-level machine learning algorithms that are inspired by the structure of our biological nervous systems.
So what makes this technology so different?
Well, neural networks can be described as stochastic algorithms rather than deterministic.
Up until fifth generation computing, all classical machines were deterministic. This meant that the output information from a machine would be fully determined by the parameter values and conditions set by the programmer or users.
This is why artificial neural networks are considered stochastic algorithms, as they have inherent randomness. Therefore, the same set of parameter values or conditions may lead to a variety of different outputs. These outputs are based on probabilities rather than predetermined instructions.
This is the fundamental basis of neural networks.
Hence, the neurons in the neural network are based on probabilities, and this is the fundamental difference between machine learning and classical computers of previous generations.
To explain further;
Each neuron, or point, is a mathematical function which simply holds a number, often between 0.00 and 1.00, but in reality, each number can be anything, i.e., negative infinity (−∞) to positive infinity (+∞).
This number inside the neuron is called its activation.
Through forward and backward propagation, which over time adjusts the weights and balances of each node, the machine’s output becomes more and more accurate, i.e., the machine learns!
Hence, this is why artificial neural networks are considered “machine learning.”
Natural Language Processing or NLP
Another vital element of cognitive computing is natural language processing, or NLP.
An analogy of a child learning a language is a good way to understand this concept.
First a child will learn common words, then progress to understanding grammar rules with increasing complexity, and then they will extend their glossary with synonyms or alternative words and meaning. In a similar way, cognitive computers with NLP will start with basic understanding and simple keyword search, eventually moving onto greater complexities, such as understanding irony.
However, this is obviously done at a much greater speed than human development due to the computational abilities of machine learning and supercomputers.
Why Do We Need Cognitive Computing Anyway?
Classical computers can only understand structured data, such as data stored within a relational database.
And since the 1970s, data has been stored in neat columns and rows within a relational database using the SQL language.
Edgar F. Codd created the relational model whilst working at IBM, and the model became the theoretical basis for relational databases and relational database management systems.
Therefore, classical computers, or computers before cognitive computing, could only understand structured data, such as within an SQL database. However, IBM Watson is able to understand unstructured data.
Unstructured Data and Big Data
At least 80% or more of the world’s data is unstructured, AKA “dark data,” and this means that humans can’t make sense of it, or at least, cannot draw meaningful analysis from it.
Understanding structured data is beyond the capabilities of classical computers, and it is the very reason that IBM Watson was created.
Stochastic optimisation addresses some of the challenges posed by the uncertainty inherent in big data.
Here’s some perspective;
Doctor John Kelly quoted these statistics at a talk he did in 2015:
- 2.5 quintillion bytes of data is created every day
- 90% of the world’s data has been created in just the last two years
- Every one minute 1.7 megabytes of data is created for every person on the planet. All 7.8 billion of us!
IBM Watson – The Hardware
There is much talk online about the clever algorithms that make IBM Watson and other cognitive computing services work. However, in truth, it is the developments in hardware, with greater storage, memory, and processing power, that have enabled the development of cognitive computing.
So, what is under the hood of IBM Watson?
- 2,880 3.55 GHz Power 7 processor cores
- 16 terabytes (TB) of RAM
- 90 IBM Power 750 servers, which are clustered together to work as one unit.
- Runs on Linux operating system
- Operates at 80 teraflops, which is 80 trillion operations per second
IBM senior consultant Tony Pearson estimated the Watson hardware cost at approximately $3 million.
The amount of processor power from IBM Watson is incredible. For purely illustrative purposes, we can compare it to some more well-known technology on the market.
For instance, the Xbox Series X is measured at 12 teraflops, which means it can handle up to 12 trillion operations per second, which is extremely fast compared to some of the first ever developed processors.
However, no doubt they will both be considered slow in the not so distant future with the rate of improvements in technology, not just with the exponential increase in Moore’s law but other advances in quantum technology and neuromorphic computing.
But still, if we compare those speeds with the ENIAC, which was a first generation computer and was the first ever programmable general-purpose digital computer, The ENIAC only had a basic machine cycle of only 5000 cycles per second! Therefore, you can see just how far technology has developed.
Power 7 Processor
So, what makes the power seven so powerful?
The Power 7 processor chip is only 567 millimetres squared, but it has 1.2 billion transistors. Comparatively, 3rd generation computers only have around 20 transistors per chip.
Furthermore, the Power Seven is an 8-core processor.
This allows for multiple instructions to be executed at any one time. The process of executing instructions within a traditional processor is done sequentially, meaning for an instruction to start, the instruction before it must end or go through the full fetch, decode, and execute cycle.
In contrast to this, Power Seven, with its eight cores, can process multiple instructions at once, the same as many other multi-core processors on the market today, just on a much greater scale.
In fact, most modern-day laptops or PCs are at least dual-core, allowing them to process multiple instructions at the same time.
The Power 7 processor has a clock speed of 3.55 GHz. So this means that the cycles happen at a very high speed.
Simultaneous Multithreading
The power seven chip is capable of four-way simultaneous multithreaded operation.
Simultaneous multithreading is capable of a far greater rate of processing due to the multiple threads executed simultaneously. This is unlike single-threaded processing, whereby the processor must wait for each thread to finish executing. Similarly, it does not waste time switching between threads, like with other multithreaded processors.
Simultaneous multithreading, or SMT, executes instructions simultaneously, allowing for far greater efficiency.
As you can see from the video above, a superscalar processor has a lot of wasted time with no execution, whereas the simultaneous MT, like the Power Seven, has a lot fewer empty boxes, meaning there is a lot less time wasted where an execution cycle can occur.
One other great feature in the power seven processor is known as “intelligent thread”.
This technology allows processing threads to vary in their execution dynamically, based on the current workload. Ultimately, this allows for an even greater capacity of processing power, as it allows more tasks to be completed in parallel.
As you can see, the processing performance of IBM Watson is very powerful.
Though, like all modern computers, a super efficient processor is useless without the memory to store the data it is processing, and IBM Watson has plenty of RAM!
In fact, 16 terabytes to be exact.
For comparison, an average Mac Book Air has only between 4 and 8 GB of RAM, and the 1951 EDVAC, which was a binary serial computer, had a memory in today’s terms of only 5.5 kilobytes!
750 Servers
All of this power is wrapped up in the power 750 servers.
These servers contain both the RAM and Power 7 processors. In total, IBM Watson has 90 IBM Power 750 servers, clustered together to work as one unit.
That is a lot of processing power and memory!
Linux Operating System
Watson also runs the operating system Linux, which is known for its performance.
It is very lightweight and focuses on speed and security rather than a user-friendly approach for systems like Microsoft Windows or Apple Mac OS, which are designed more for home or office use.
“In the latest top 500 list of the world’s most powerful supercomputers, 459 of the top 500 supercomputers were all running Linux.”
Steven Vaughan-Nichols (zdnet.com)
Watson for Smarter Businesses
So, how can businesses work with IBM Watson?
Well, at an estimated cost of $3 million, it would be a huge investment decision for many businesses to buy that hardware. As a result, this would make it unrealistic for startups or SMEs.
However, IBM made the decision to make Watson an open multi-cloud platform, whereby enterprise apps can connect via API’s.
This makes the commercial opportunities for IBM Watson plentiful.
Today;
“Watson AI is at work globally in farms, factories and offices, streamlining workflows, increasing productivity and freeing workers up for higher value tasks”.
IBM
It is also at work, in healthcare and in many other very important industries too.
The Future of AI and Watson
So, what does the future look like for computing?
Technology moves fast. Although cognitive computing and machine learning are unbelievably powerful, technology is still progressing. However, not necessarily in a way that supersedes it but further enhances the capabilities of machine learning and AI.
The most recent advancement in computing is quantum.
The real world is much more complex than binary data, and therefore, representing the real world in just ones and zeros has its limitations.
Whereas quantum computing isn’t just zeros and ones, it has another state that is neither zero nor one, but yet, it can also be zero and one at the same time. It is based on quantum physics, and research is ongoing as to how quantum computing can overcome the computational limitations of today’s computers.
- The Future of Personal Data and Digital ID - June 30, 2023
- Mobile Apps and Smartphone Technology: Challenges and Opportunities - December 17, 2022
- AI Cognitive Computing: An In-Depth Look At IBM Watson - August 15, 2022
