Devesh Rajadhyax

4 hints to get started with #AI in your company by Devesh Rajadhyax @deveshrajadhyax

Most companies are working on Digital Transformation today, and Artificial Intelligence is a critical part of that transformation.

Two questions immediately present themselves-
1.    What is Digital Transformation and how it is different from the IT/ICT transformation that is happening since for than four decades?
2.    Why is AI a critical part of this transformation?

Let me take the first question.

Digital Transformation is actually a cognitive revolution. It is a more humanlike way of making sense of the world around. And this is our clue to the most important difference between IT and Digital-
IT systems are not humanlike. They don’t try to make sense of the world around them. They create a small world of their own and everyone follows the rules of that world. The input has to be given in the way they demand and output will be available in forms that they prefer. You better fall in line.
Digital, or Cognitive systems would want to fit into the world they find themselves. To understand this, see how we humans manage our operations. We take in all the signals from the real world. We see, hear, feel, smell and then speak, show, move and push at the world around us. That’s what Digital systems aim to do. They want to process all the available inputs and then interact in a natural way. That’s why they hope to solve problems much closer to our life.’
Now let’s turn to the second question, why AI is so critical in this transformation.
I think the above distinction must have given you an intuitive answer. Because we are aiming for a humanlike way of interacting with the world, we need AI to do the processing. We also need sensors (IoT) and motors (Robotics), displays (VR) and so on. You can see how well the various Digital Technologies fit in this paradigm.
But even before AI, there is a big big factor that drives this whole transformation, and that factor is Data.
Since we are talking about real world data, we are talking huge huge volumes in forms that we never processed before. Vision, audio, waveforms, text, handwriting and many more types of data need to be captured (again IoT), stored and processed (Cloud, Big Data) and managed (Blockchain). But if you are the one responsible for planning and implementing systems, this central importance of data means a fundamental change for you.
When you plan classical IT systems, you start with the objectives of the company. Accordingly, you define the requirements. Analysis and design are further carried out on the requirements. Data is an outcome of this process.
Fig 1: Planning for classical IT systems
In Digital systems, however, data drives the system planning. This is again somewhat like like humans. We cannot demand more data from the world. We work out our affairs in such a way that we can manage with whatever data the world gives us. And we have become very good at extracting as much meaning from that data as possible.
So the new paradigm will be:
Fig 2: Planning for Digital systems
As we can see, we are now thrown in a more uncertain and complex world. The existing data indicates possibilities of what can be done. These possibilities have to be mapped on the organizational objectives to decide the Digital Transformation plan.
There are more complexities in the Digital paradigm that shown in the diagram. But for the purpose of this article I would like to present a simple view, leaving the complexities for a future and longer article.
This immediately leads us to the four most important points that can help us to get started with AI. This framework does not consider AI in isolation, but the whole Digital Transformation.
The four hints are:
1.    The data that you have: Identify all the data that your company owns. The data can be put into three major buckets:
  • Structured Data: This is the easiest data to identify. It will be found in all the IT systems that you have implemented so far. Also look for the countless excel sheets that your employees have created.
  • Unstructured data: This will be typically text that has no fixed format. Emails, proposals, invoices, challans, vouchers and so on. Hint- even if you might have some of this data in your IT system, there may be ‘left-over’ data. For example, while some part of an invoice is entered in your AP system, there may be some that is not entered. This left-over might contain interesting possibilities.
  • Dark data: This is that part of your data that you never really thought about. In fact, it may not be being captured today. Photos that can take, vibrations that you can record, video that you can capture – these possibilities are endless. No wonder that IBM put the volume of dark data at 80% of all data.
2.    Objectives of your company: There is a whole load of literature on how to identify objectives of your organization and I am definitely not qualified to comment on it, but broadly they will fall in two buckets:
  • Aspirations: what you would like to happen – improvements, new initiatives, lead over competitors and so on.
  • Pain points: what you would like to remove – delays, leakages, inefficiencies.
There will be various areas of your business that will have their own objectives – customer experience, process efficiency, employee satisfaction, innovation, leadership and so on. I will write no more – you are the best judge.
3.    Applications (or Possibilities): The applications of data are of course endless, but since we are developing a framework, let’s again put them in the biggest buckets:
  • Automation: Essentially replacing human efforts by machine. This not only saves cost, but improves accuracy and speed. In most cases, human effort can be diverted to higher cognitive tasks, giving further advantage to the organization.
  • Analytics: Again lots of material is available on this topic, generally categorized in four types:
                    i.    Descriptive
                    ii.    Prescriptive
                    iii.    Diagnostic
                   iv.    Predictive
The most valuable use of analytics is in decision support, for which it has to be combined with Knowledge (see below).
  • Knowledge: The third and often ignored use of data, is also the most difficult to achieve. Combined with automation and analytics this can give rise to spectacular applications. The currently popular use case of chatbots is an example of the Knowledge possibility.
4.    Existing systems: The fourth hint for getting started with AI is studying your existing IT systems. Most of the above possibilities give best results when interfaced with one of the existing system.
I will now try and put these four things in the framework together with help of a simple example.
Let’s say you have an invoice management module in your existing Accounts Payable (AP) system. The scanned copies of incoming invoices are entered in the system by your employees. These scanned documents are the unstructured data.
The possibility this data presents is that of process automation, along with many others. Now, your company has a defined objective of improving process efficiency, and automation of invoice management fits well with that. Since you already have a AP system in place, the forth criterion is also met.
It seems that automation of invoice management is the right problem statement for your company to get started with AI. There will be a few startups and experienced companies who will be able to help you to get started.
The article was first published on the author’s linkedin pulse page, its been re-published here with the author’s permission. 
Author
Devesh Rajadhyax

Founder and CEO, Cere Labs, AI, Machine Learning, Deep Learning

#AI in Healthcare by @deveshrajadhyax


There are some subjects that invoke sharp and contrasting emotions in the society. In present day India, the GST tops the list of such things that are considered boon by some and curse by others. New technology usually does this to people. The steam engine, the telephone and the computer all have been greeted both as the savior and the nemesis of the mankind. 


Using Artificial Intelligence in Healthcare is one such subject. If you have to believe what the media says, AI is going to transform healthcare in the near future. In fact, the services of the doctor may not be needed very often, if at all. AI will do everything in medicine including diagnosis, treatment and even finding out new drugs.

But not everyone is so optimistic or even welcoming. The idea of machines taking care of our health is creepy to some. Others claim that AI can never replace a doctor, at least not in a foreseeable future. Medicine is too complex for machines to figure out.

(You must have noted that I am using Healthcare and Medicine interchangeably. The reason being that this is what most people do. Healthcare is the practice of medicine and as such is wider than it. I refer to healthcare as including medicine.)

The reality, like in the case of GST and most other things, will be somewhere in the middle. The purpose of this article is to find that balanced view. In effect, what I will be saying is:

“while the replacement of the doctor is a faraway dream, there are a number of things that AI can do in medicine even today. This can turn out to be valuable help for doctors, patients and other stakeholders”.


Let me first present a short introduction to AI.

AI, like Philosophy, is a very hard term to define. AI is not really one technology. It is a collection of techniques. Strictly speaking, AI is actually an ambition. The ambition of machines to imitate human capabilities.  
   
But this definition does not take us very far. Since human capabilities are many, ranging from walking to writing poems, imitating any of these capabilities can be called AI. So for our purposes, we will define AI as the pursuit of those capabilities that are strong points of human beings.

As an example, consider language. Reading an article and understanding its gist is a simple matter for us humans. For machines to achieve this capability will be quite something. If that happens, machines can go through a number of articles for us and feed us with the just the little bits that we need.

A whole lot of mathematical and computational techniques have been developed by researchers in the last sixty years to achieve this goal. Deep Learning, Machine Learning and NLP are some names given to a bunch of such techniques. In the last few years, AI has risen to prominence mainly due to three reasons – availability of data, increase in computing power and discovery of new methods. 

Armed with that introduction, let’s try and put down the areas where AI can make a difference in healthcare. While we do that, we can also try to answer the ‘replace the doctor’ question.

Diagnosis: Diagnosis is the hardest part of medicine. There is no definite pathway to diagnosing a patient. A lot depends on the experience and intuition of the doctor, in that way, it is more of an art than science. As of now, it is difficult to see AI taking over this role. However, there are many areas where AI is already making a difference:

  • Conditions in which diagnosis is dependent on analysis of a signal over time, such as an ecg or an eeg. Machine Learning combined with signal processing can achieve good results here. Arrhythmia or irregular heartbeat is an example of such a condition that AI can detect well.
  • Diagnosing some disorders involves referring to a lot of data such as past and present reports, images and history. Gatro-intestinal disorders are notoriously difficult to diagnose and require a lot of information to refer. AI can make a big difference here by sifting through the pile of data and presenting important facts to the doctor.
  • In radiology, the volume of cases is huge and the radiologist needs to look at every image to come to a conclusion. Some investigations like MRI produce a large number of images for each patient. This makes the doctor’s time a bottleneck in handling the ever growing number of patients. Deep Learning has shown great promise in being able to classify medical images. For example, it can separate images that indicate normal functioning from those that have some abnormality. This will enable the radiologist to focus on the abnormal cases first. This method will also be a boon for the remote places where a radiologist is not available.
  • AI has provided a new method for laboratory investigations. This may mean that in the future most lab tests including pathology will be done with basic instruments at a very low cost. In a disease like HIV/AIDS, being able to determine the viral load in a quick and inexpensive way can be a very big benefit to the patients. 


Treatment: The biggest contribution AI can make to treatment of patient is in the area of drug discovery. Currently, discovering a new drug costs more than 2.5 billion dollars and takes more than a decade. The pharmaceutical industry is desperately searching for new ways to reduce the cost and time. AI may be one of the solutions to this problem. Machine Learning and Deep Learning are being used in various stages of drug discovery, such as identifying candidate molecules and studying the expected response of the new drug.

In our fight with cancer, AI may be an important weapon. Personalized Oncology is rapidly getting attention from the medical community as the way forward in battling with the cancer scourge. To describe in brief, cancer is not one disease – the cancer of every patient is different. If the individuality of cancer is decoded, a personal treatment path can be planned for every patient. AI will become a key part of this process.

AI is already playing a role in treatment by making robots that perform surgeries. This contribution will grow in the time to come with the robot costs falling and capabilities growing. This will reduce the strain on surgeons and they will be able to perform far more surgeries in the same time. 

Care: Care during the illness and recovery is as important as the right diagnosis and treatment. Along with IoT, AI will transform patient care. Everything from medicine intake to prescribed activity will be monitored by these systems. Monitoring includes two components – sensing and analysis. While the sensing part is done by the IoT devices, analysis is provided by AI. 

Prevention: Prevention is definitely preferred to hospitalization and AI is going to play a major role in this. It will involve both personal and public health. Personal health is monitored by the wearables and other simple devices. The AI systems will process this data to look for possible indications of disorders so that they can be fixed inexpensively. 

Public health will be monitored in the same way but from data that is coming from various healthcare institutes. This enormous data will forewarn us about various health risks such as outbreaks of diseases. It will enable the state to take measures to avoid the calamities.

In short:
To summarize, AI will really be a transformational technology for healthcare. It will make healthcare cheaper and faster and enable it to reach more number of people. AI will reduce the strain on doctors and nurses. However, for the future that we can see, AI will serve more as an assistant to the doctors, rather than being their replacement. 

Author
Devesh Rajadhyax

Founder and CEO, Cere Labs, AI, Machine Learning, Deep Learning

Four ways in which #AI can help humankind @deveshrajadhyax


Artificial Intelligence is receiving more than its fair share of public attention. On one side there are promises of miracles, while on the other side there are warnings of doomsday. What is probably missing is a simple listing of clear benefits. This is article is an attempt to create such a list.


Artificial Intelligence is more of an ambition than a technology. The ambition is to imitate human capabilities. Since human capabilities range from walking to solving mathematical problems, AI also encompasses systems of various types – ranging from the humble calculator to Google’s DeepMind.

In this article, I am majorly referring to the AI systems that try to achieve the cognitive abilities of human beings. Cognitive abilities refer to the processes of our mind such as understanding, reasoning, planning and selecting the right action. Understanding a question and supplying the right answer from our memory is an example of cognitive ability, the one that AI systems called ‘chatbots’ try to imitate.

Cognitive systems are currently in their initial phase of development. Once they come close to human beings in their competence, they can prove useful to humankind in a number of ways. Here are some:

1. Better utilization of resources: In our current world, it requires a human being to use resources. For example, a car needs a driver. Platforms like Uber have made it possible to share your car when you don’t need it, but it still requires a skilled human being. Talk to your Ola or Uber driver and you will realize that they are already working at the limit of their capacity. A cognitive system driving vehicle will use them much more efficiently. You will actually need much fewer vehicles than you need today (and probably a lot lesser parking!). This is true of most other resources.

2. A fairer society: Human beings have many faults in their thinking. In another article I have highlighted this faults, called biases. These biases have their roots in the evolution, so the AI systems will (hopefully) not have them. ( Pl see http://blog.cerelabs.com/2017/06/will-ai-evolve-to-be-as-bad-as-humans.html). This will make decision making at all level fairer for the people. To take an example, typecasting is a very strong bias that we suffer from. This affects decisions taken by, say an interviewer. We can hope to see much fairer selection processes in the future. For an interesting example of how statistics can help to break biased notions, see the movie or read the book ‘Moneyball’.

3. Repositioning of human efforts: World over, a large number of people are engaged in time consuming tasks that require moderate cognitive ability. Take for instance cooking. A big part of a woman’s day in India is spent in preparing food. Cognitive systems such as robot chefs can easily take over these jobs, freeing up a lot of time that can be invested in more valuable responsibilities like education of the children. In the industries, as the simpler tasks are done by cognitive systems, humans can move up the value chain, pushing the efficiency of the enterprise higher.

4. Improved care: Currently, care of patients, elderly and disabled is primarily a human responsibility. Many times, this compromises the quality of care as people cannot take out so much time from their daily activities. Cognitive systems can make the life of those in need of care much better. They can talk to the elderly and carry the disabled to places otherwise difficult to reach. The systems can keep an eye on chronic patients, not just reminding but making sure their regimen is adhered to.

This is of course just a small contribution to an ever growing list of benefits. While we keep our eyes open to the warnings given by the likes of Stephen Hawking and Elon Musk, we should continue in our efforts to harness the power of AI for these benefits.

Author
Devesh Rajadhyax

Founder and CEO, Cere Labs, AI, Machine Learning, Deep Learning
Scroll to Top
Connect
1
👋 Hello
Hello!! 👋 Manish here, Thanks for visiting The Healthcare IT Experts Blog !! How can i help you?