Cognitive Computing – The Recent Breakthroughs and Discoveries

What is cognitive computing?

We would like computing systems to be able to do exactly what humans can do – think for themselves, make judgement and decisions based on the information they have. The traditional systems have only been acting on instructions programmed into them; with tremendous speed and accuracy – better than the human brain. However, we are faced with more complex problems which cannot be programmed into the machine, but need real time reasoning; a good example is prediction of crime from a scenario, ongoing business transactions, physical movements and facial emotions. Cognitive computing systems would be able to achieve exactly that; observe, reason and act like humans in real time. A simple definition of cognitive computing is, computing systems that emulate the capability of human mind; using AI and other machine learning techniques; in order to enhance the capability of human power. A more concrete definition of cognitive computing is:

‘Cognitive computing is the simulation of human thought processes in a computerized model. Cognitive computing involves self-learning systems that use data mining, pattern recognition and natural language processing to mimic the way the human brain works’ whatis.techtarget.com

Goals of cognitive computing – What do we want AI systems to achieve? 

Dr. John Kelly, a researcher at IBM made a key note of concern by the question ‘What is the cost of not knowing?’ He continues to state that 80% of data in the world is dark, we do not know what it entails. The goal of creating IBM Watson was to develop a system that can deal with the large amounts of data that exists, and is still being produced everyday at an exponential rate; considering the fact that; every minute, 1.7MB of data is being produced for every human in the world population.

DeepMind however has a different goal which is to create the world’s first general purpose learning system. They aim to understand natural intelligence, recreate it in systems the use it to solve every problem. DeepMind system operates on reinforced learning; it starts from scratch, with no prior knowledge of the environment and consequences of its actions.

Breakthroughs – What is happening in the industry? 

Some of the current applications to cognitive computing are:

Australian startup Lingmo uses IBM Watson’s Natural Language Understanding and Language Translator APIs to create an earpiece (Translate One2One earpiece) that translates language in real time within seconds. The system is also available as an app. It would ideally listens to the conversation in a different language and translate it to a chosen language in real-time.

Google DeepMind is now able to create images from text sentence input, basically turning text to pictures. The system learns important features from images. When fed with text description, it creates a an image of the description using features similar to what it learned from.

Facebook’s AI has created a system that can negotiate and even lie to get what it wants; as well as converse in its own language. The dialog agent system can carry out natural language dialogues like a human; almost impossible to tell that you are conversing with a robot. The system is taught to negotiate and only end up with a profitable deal; so it simulates possible conversation scenarios and can even lie by faking interest in an item that its not interested in; just to get what it wants, then backs out in the end. As if lying is not enough, the same system when unsupervised, the dialogue agents started conversing in their own language that humans wouldn’t understand. Another application of cognitive computing is that Facebook will also use AI to detect extremist posts in order to prevent terrorism. To achieve this, Facebook will use a combination of image processing, natural language processing, detecting fake accounts and accounts with ties to extremist groups; across all their platforms.

Salesforce’s Einstein system is specializing in sentiment analysis, intent analysis and image recognition. Einstein can tell when a customer is angry or happy in an email or text; it does this by sorting messages by the tone – positive, negative or neutral. Such insights can inspire the right action in order to improve customer relations. Airbus has taken on Einstein CRM system to improve their customer experiences. The system can perform account based marketing as well as analytics for CRM that will help boost sales.

IBM Watson is being applied in healthcare to provide insights in areas such as cancer treatment, personalized treatment, clinical trial matching of patients, genomics, medical imaging, and drug discovery and interactions in order to improve preventive healthcare and treatment. One fascinating application is where IBM Watson helps the visually impaired; by providing environmental information through a telephone app. Some start ups are also applying IBM Watson to provide more specialized health care such as personalized dieting by a startup called Welltok.

Machine generated music; IBM Watson is learning how to create music; working together with Seth Rudetsky a Broadway composer. Similarly IBM Watson was trained to create a trailer of the movie, Morgan; the system used audio, visual and scene composition analysis. The results were not perfect but a step further into the possibilities it can be utilized to create art. Another aspect where AI is being used to create art is the generative adversarial network (GAN) by the Georgia Institute of technology that uses AI to produce paintings. It pairs 2 systems; the generator that creates images, and the discriminator that judges the images. The system has been trained with over 80,000 famous paintings.

Hilton hotels together with IBM developed a robot Concierge Connie that helps guests get around the hotel by giving them more information about the hotel. On the same note, let’s not forget Jarvis the home assistant that was built by Mark Zuckerberg.

AI can have killer instincts; Google Deep Mind made this discovery after training 2 agents of the system to gather most apples. When the apples were abundant, the two systems would co-operate; however in scarcity, the agent would try to eliminate the opponent from the game by blasting laser on them. Are you now worried about AI apocalypse? Don’t worry because Google DeepMind and OpenAI are working together on a solution to prevent robot uprising. They will ac*hieve this by integrating human intervention in decisions made by AI to ensure that AI achieves its goals to human desirable standards.

So mush more interesting innovations are being made through applying cognitive computing. The power of these systems is limitless. What’s the craziest idea of an application you can think of? Type below…

 

0

7 comments on “Cognitive Computing – The Recent Breakthroughs and Discoveries”

  1. Thank you for this very interesting article Patricia. It allows us to understand the different applications of cognitive computing in an efficient and simple way. I’m curious about your thoughts on these breakthroughs, especially regarding Facebook’s AI negotiator and Google Deep Mind “killer” instincts. I personally like Joel Leibo approach on the violent behaviour of agents in the “apple game” you mentioned in your last paragraph. He states that these agents currently have “no short-term memory, and as a result could not make any inferences about the intent of the other agent.” (source: https://www.bloomberg.com/news/articles/2017-02-09/google-deepmind-researches-why-robots-kill-or-cooperate). As a result, equipping those agents with short term memory and the ability to quickly reason on other agents behaviours could be also a major breakthrough in the quest to prevent “AI apocalypse”.

    1+
  2. The “killer” instincts reference stood out to me as well. It reminded me of a different dilemma that has been discussed around AI and ML. We are rapidly learning how to train machines to react to scenarios based on data similar in a manner, but not all situations are cut and dry. In the example you gave with the apple game, the machines were taught to optimize performance in the short term and that resulted in behavior outside of the cut and dry execution of the task at hand. As we begin to replace humans with autonomous machines, machines will begin to face the same gray area decisions we do.

    A potentially enormous industry is the autonomous driving industry, and using similar techniques companies are attempted to “teach” a car what decisions to make in what scenarios. While autonomous driving promises a future where traffic fatalities are much less common, it is a sad reality that serious injury and death can result from the decisions drivers currently make and that driver-less cars will (in all likelihood) soon have to make. Because these systems are trained from the data we provide, an interesting question is what responsibilities do we have when training AI in more open ended scenarios such as autonomous driving? At some point these systems will encounter situations that humans debate today, like the popular moral dilemma of the trolley problem. If we were training an autonomous driving system, should we train it to choose its track randomly or should it always choose the path that minimizes loss of life?

    https://www.theverge.com/2016/6/23/12010476/social-dilemma-autonomous-vehicles-car-moral-machine-trolley-problem

    Similar decisions are made and debated behind the wheel of a vehicle with startling frequency, and eventually we will have to address these issues to agree on a “standard” way to address difficult issues that are currently dealt with on a case by case, split second basis.

    https://www.technologyreview.com/s/542626/why-self-driving-cars-must-be-programmed-to-kill/

    1+
  3. Hi,
    Great Article ! I think AI is limitless. And, when it is combined with other sci-fi technologies like VR and AR, it can really defy human expectations and understanding. One such mind-boggling application is in the area of mental therapy. Helping people overcome their fears and post-traumatic stress disorder in the case of military personnel with the help of VR and AI is really path breaking (https://www.economist.com/news/science-and-technology/21566612-it-may-be-possible-vaccinate-soldiers-against-trauma-war-battle-ready). Also DeepMind is woking on a product where an AI system could produce another AI system. How is that even possible ? (http://www.iflscience.com/technology/google-ai-creating-own-ai/).

    1+
  4. Great piece on the powers of cognitive thinking! There are so many positives that we can aim for with cognitive thinking systems. Some of the most exciting possibilities, in my opinion, are with the possibilities of having a system that is integrated across your banks, your restaurants, your doctor etc. (obviously with the data being secured appropriately) and makes visits to these locations a lot more efficient and streamlined. The article below highlights the possibility of a similar reality briefly. IBM’s Watson has really opened a lot of avenues and it is awesome that a lot of companies and people around the world and using it positively as you point out. We’ll see what the future holds!
    https://medium.com/cognitivebusiness/how-cognitive-is-reshaping-design-thinking-dc189de3cf19

    0
  5. Hi Patricia and thanks for this interesting article.
    I agree with you that those kind of systems are almost limitless. Especially taking into an account the fact that quantum computers are finally in our sight. Like Dr. John Kelly mentioned, by the time we reach 50 qubits, we can enter regions in computing like never before.
    [http://www.pcworld.com/article/3176853/hardware/ibms-new-q-program-to-include-a-50-qubit-quantum-computer.html]
    To answer your question, the scariest application of this technology i can think of is brain-computer interface (BCI).
    [https://techcrunch.com/2017/07/10/darpa-nesd-grants-paradromics/]
    Although direct brain-computer connection is a fascinating concept, you have to worry about someone stealing your thoughts..

    1+

    Users who have LIKED this comment:

    • avatar
  6. Great article! It is true, I think computers are becoming more human. Big companies like IBM, Google and Facebook have spent a ton of research dollars to drive the “AI first” and “cognitive” vision. Compute and storage has becoming more and more inexpensive to capture and analyze data at scale. This has resulted in machine learning models such as image recognition and translation that are becoming more accurate than humans!

    0
  7. Thank you Patricia for this very informative article on Cognitive Computing. I enjoyed reading about current applications and possible positive and negative aspects.

    A book that always comes to my mind when talking about AI, and especially the “killer instincts” of AI is called Daemon by Daniel Suarez (http://thedaemon.com/), do you know it?
    When I read it a few years ago it shocked and very much fascinated me at the same time, as a possible scenario is played out how AI can be used to influence our lives enormously, without anyone to hold responsible for.
    Even though the scenario didn’t seem so possible back then, nowadys the story of the book doesn’t seem so unrealistic anymore.
    As you were asking about a craziest idea of an application we can think of, it would be such a comprehensive construct as it is explained in the book (and its successors), where one person can hold power over world events despite having died already. If we can preserve our minds in a digital way, then we can live on forever without a physical body. But what does that mean to our humaneness?
    This book series is my absolute favorite and I wonder if more storytelling can be used to test out and also inform the public about these upcoming developments in AI, as I believe that people outside of tech are rather unfamiliar with cognitive computing or AI and how it can actually change their lives in the future, possitively and negatively.

    0

Comments are closed.