Human thinking is beyond imagination. Can a computer develop such ability to think and reason without human intervention? This is something programming experts at IBM Watson are trying to achieve. Their goal is to simulate human thought process in a computerized model. The result is cognitive computing – a combination of cognitive science and computer science. Cognitive computing models provide a realistic roadmap to achieve artificial intelligence.
“Cognitive computing represents self-learning systems that utilize machine learning models to mimic the way brain works.“ Eventually, this technology will facilitate the creation of automated IT models which are capable of solving problems without human assistance
Cognition comes from the human brain. So what’s the brain of cognitive systems?
Cognitive computing represents the third era of computing. In the first era, (19th century) Charles Babbage, also known as ‘father of the computer’ introduced the concept of a programmable computer. Used in the navigational calculation, his computer was designed to tabulate polynomial functions. The second era (1950) experienced digital programming computers such as ENIAC and ushered an era of modern computing and programmable systems. And now to cognitive computing which works on deep learning algorithms and big data analytics to provide insights. Thus the brain of a cognitive system is the neural network, fundamental concept behind deep learning. The neural network is a system of hardware and software mimicked after the central nervous system of humans, to estimate functions that depend on the huge amount of unknown inputs.
What are the features of a cognitive computing solution?
With the present state of cognitive computing, basic solution can play an excellent role of an assistant or virtual advisor. Siri, Google assistant, Cortana, and Alexa are good examples of personal assistants. Virtual advisor such as Dr. AI by HealthTap is a cognitive solution. It relies on individual patients’ medical profiles and knowledge gleaned from 105,000 physicians. It compiles a prioritized list of the symptoms and connects to a doctor if required. Now, experts are working on implementing cognitive solutions in enterprise systems. Some use cases are fraud detection using machine learning, predictive analytics solution, predicting oil spills in Oil and Gas production cycle etc.
The purpose of cognitive computing is the creation of computing frameworks that can solve complicated problems without constant human intervention. In order to implement cognitive computing in commercial and widespread applications, Cognitive Computing consortium has recommended the following features for the computing systems –
This is the first step in making a machine learning based cognitive system. The solutions should mimic the ability of human brain to learn and adapt from the surroundings. The systems can’t be programmed for an isolated task. It needs to be dynamic in data gathering, understanding goals, and requirements.
Similar to brain the cognitive solution must interact with all elements in the system – processor, devices, cloud services and user. Cognitive systems should interact bidirectionally. It should understand human input and provide relevant results using natural language processing and deep learning. Some intelligent chatbots such as Mitsuku have already achieved this feature.
3. Iterative and stateful
The system should “remember” previous interactions in a process and return information that is suitable for the specific application at that point in time. It should be able to define the problem by asking questions or finding an additional source. This feature needs a careful application of the data quality and validation methodologies in order to ensure that the system is always provided with enough information and that the data sources it operates on deliver reliable and up-to-date input.
They must understand, identify, and extract contextual elements such as meaning, syntax, time, location, appropriate domain, regulations, user’s profile, process, task, and goal. They may draw on multiple sources of information, including both structured and unstructured digital information, as well as sensory inputs (visual, gestural, auditory, or sensor-provided).
What is the scope of cognitive computing?
While computers have been faster at calculations and processing than humans for decades. But they have failed miserably to accomplish tasks that humans take for granted, like understanding the natural language or recognizing unique objects in an image. Thus cognitive technology makes such new class of problems computable. They can respond to complex situations characterized by ambiguity and have far-reaching impacts on our private lives, healthcare, business, etc.
According to a study by the IBM Institute for Business Value, “Your Cognitive Future,” scope of cognitive computing consists of engagement, decision, and discovery. These 3 capabilities are related to ways people think and demonstrate their cognitive abilities in everyday life.
The cognitive systems have vast repositories of structured and unstructured data. These have the ability to develop deep domain insights and provide expert assistance. The models build by these systems include the contextual relationships between various entities in a system’s world that enable it to form hypotheses and arguments. These can reconcile ambiguous and even self-contradictory data. Thus these systems are able to engage in deep dialogue with humans. The chatbot technology is a good example of engagement model. Many of the AI chatbots are pre-trained with domain knowledge for quick adoption in different business-specific applications.
A step ahead of engagement systems, these have decision-making capabilities. These systems are modeled using reinforcement learning. Decisions made by cognitive systems continually evolve based on new information, outcomes, and actions. Autonomous decision making depends on the ability to trace why the particular decision was made and change the confidence score of a systems response. A popular use case of this model is the use of IBM Watson in healthcare. The system can collate and analyze data of patient including his history and diagnosis. The solution bases recommendations on its ability to interpret the meaning and analyze queries in the context of complex medical data and natural language, including doctors’ notes, patient records, medical annotations and clinical feedback. As the solution learns, it becomes increasingly more accurate. Providing decision support capabilities and reducing paperwork allows clinicians to spend more time with patients.
Discovery is the most advanced scope of cognitive computing. Discovery involves finding insights and understanding vast amount of information. These models are built on deep learning and unsupervised machine learning. With ever-increasing volumes of data, there is a clear need for systems that help exploit information more effectively than humans could on their own. While still in the early stages, some discovery capabilities have already emerged, and the value propositions for future applications are compelling. Cognitive Information Management (CIM) shell at Louisiana State University (LSU) is one of the cognitive solutions. The distributed intelligent agents in the model collect streaming data, like text and video, to create an interactive sensing, inspection, and visualization system that provides real-time monitoring and analysis. The CIM Shell not only sends an alert but reconfigures on the fly in order to isolate a critical event and fix the failure.
Cognitive computing landscape
Present cognitive computing landscape is dominated by larger players – IBM, Microsoft, and Google. IBM being the pioneer of this technology has invested $26 billion dollars in big data and analytics and now spends close to one-third of its R&D budget in developing cognitive computing technology. Many other companies and organizations are developing products and services that are as good, if not better than Watson. IBM and Google have acquired some of the rivals and the market is moving towards consolidation. Let’s take a look at the prominent players in this market –
1. IBM Watson
Originally Watson is an IBM supercomputer that combines artificial intelligence (AI) and sophisticated analytical software for optimal performance as a “question answering” machine famously featured in show ‘Jeopardy’. Now it uses a set of transformational technologies such as natural language processing, image recognition, text analytics and virtual agents. IBM Watson leverages deep content analysis and evidence-based reasoning. Combined with massive probabilistic processing techniques, Watson can improve decision making, reduce cost and optimize outcomes.
2. Microsoft Cognitive Services
Microsoft cognitive services previously known as Project Oxford are a set of APIs, SDKs and cognitive services which the developers can use to make their applications more intelligent. With Cognitive Services, developers can easily add intelligent features – such as emotion and sentiment detection, vision and speech recognition, knowledge, search and language understanding – into their applications. We have made a chatbot ‘Specter’ using Microsoft Bot Framework to improve the efficiency of our marketing team.
3. Google DeepMind
DeepMind was acquired by Google in 2014 and considered to be a leading player in AI research. The team consists of many renowned experts in the field of deep neural networks, reinforcement learning, and systems neuroscience-inspired models. DeepMind became popular with AlphaGo, a narrow AI to play Go, a Chinese strategy board game for two players. AlphaGo became the first AI program to beat a professional human player in October 2015, on a full-sized board.
CognitiveScale founded by former members of IBM Watson team provides cognitive cloud software for enterprises. Cognitive Scale’s augmented intelligence platform delivers insights-as-a-service and accelerates the creation of cognitive applications in healthcare, retail, travel, and financial services. They help businesses make sense from ‘dark data’ – messy, disparate, first and third party data and drive actionable insights and continuous learning.
SparkCognition is an Austin-based startup formed in 2014. SparkCognition develops AI-Powered cyber-physical software for the safety, security, and reliability of IT, OT, and the IIoT. The technology is more inclined towards manufacturing. It is capable of harnessing real-time sensor data and learning from it continuously, allowing for more accurate risk mitigation and prevention policies to intervene and avert disasters.
Watson and DeepMind’s success has inspired other companies to develop cognitive platforms using open source tools. Other leading technology companies like Qualcomm and Intel are taking cautious steps to include cognitive solutions for specialized industries. Uber has established a research arm dedicated to AI and machine learning and acquired Geometric Intelligence and Otto. Otto is an autonomous truck and transportation startup and Geometric Intelligence is focused on generating insights from fewer data using machine learning. Gamalon has developed an AI technique using Bayesian Program Synthesis. It requires only a few pieces to train the system to achieve same levels of accuracy as neural networks.
Healthcare is the most popular sector to adopt cognitive solutions. Startups such as Lumiata and Enlitic have developed small and powerful analytic solutions that assist healthcare providers in diagnosis and prediction of disease conditions.Other companies in this market are Cisco cognitive threat analytics, CustomerMatrix, Digital Reasoning and Narrative Science.
Limitations of cognitive computing
Limited analysis of risk
The cognitive systems fail at analyzing the risk which is missing in the unstructured data. This includes socio-economic factors, culture, political environments, and people. For example, a predictive model discovers a location for oil exploration. But if the country is undergoing a change in government, the cognitive model should take this factor into consideration. Thus human intervention is necessary for complete risk analysis and final decision making.
Meticulous training process
Initially, the cognitive systems need training data to completely understand the process and improve. The laborious process of training cognitive systems is most likely the reason for its slow adoption. WellPoint’s financial management is facing a similar situation with IBM Watson. The process of training Watson for use by the insurer includes reviewing the text on every medical policy with IBM engineers. The nursing staff keeps feeding cases until the system completely understands a particular medical condition. Moreover, the complex and expensive process of using cognitive systems makes it even worse.
More intelligence augmentation rather than artificial intelligence
The scope of present cognitive technology is limited to engagement and decision. Cognitive computing systems are most effective as assistants which are more like intelligence augmentation instead of artificial intelligence. It supplements human thinking and analysis but depends on humans to take the critical decisions. Smart assistants and chatbots are good examples. Rather than enterprise-wide adoption, such specialized projects are an effective way for businesses to start using cognitive systems.
Cognitive computing is definitely the next step in computing started by automation. It sets a benchmark for computing systems to reach the level of the human brain. But it has some limitations which make AI difficult to apply in situations with a high level of uncertainty, rapid change or creative demands. The complexity of problem grows with the number of data sources. It is challenging to aggregate, integrate and analyze such unstructured data. A complex cognitive solution should have many technologies that coexist to give deep domain insights.
Thus, besides AI, ML and NLP, technologies such as NoSQL, Hadoop, Elasticsearch, Kafka, Spark etc should form a part of the cognitive system. This complete solution would be capable of handling dynamic real-time data and static historical data. The enterprises looking to adopt cognitive solutions should start with a specific business segment. These segments should have strong business rules to guide the algorithms, and large volumes of data to train the machines.