Introducing cognitive automation to your organization
In this post, we will provide some guidelines on how to evaluate the AI/ML systems for your organization that you can start with to introduce cognitive automation to your organization.
The automation that is done by writing code is repetitive task automation. The modes of interaction with these type of automation for the user is most touching, type and gesture. Cognitive automation is the application of artificial intelligence and machine learning algorithms to solve a problem where the algorithm is trained instead of being customized by modifying the code. The algorithms learn from the data and identify patterns and fine-tune the accuracy of the output. There are two areas in cognitive computing where we see a lot of potentials for companies to adopt and use cognitive computing one is Natural Language Processing and the other is Computer Vision.
Evolution of Machines and Tools
Since the industrial revolution man and the machine and the tools have become inseparable parts of the work. These tools and machines have evolved over the centuries from being Assistive to being Autonomous. You can read the book Sales Ex Machina: How Artificial Intelligence is Changing the World of Selling. It lists 5 stages of evolution that are listed below.
- Assistive – Hand and horse-powered machines, steam-powered machines and finally electrified machines.
- Administrative – For knowledge workers, different types of machines were needed to manage resources and computers finally became a panacea that solved all the problems in managing the resources.
- Automation – Machines and software that can perform jobs for hours without user interactions based on a set of rules and conditions to check for.
- Augment – These types of machines helped users to become better and see clear pictures. Eg: using AI to train Chess players.
- Autonomy – This is the most advanced stage in machinery where the machines can learn, train, decide and execute the tasks by themselves without human intervention. eg: self-driving cars, drones.
Another book I recommend reading here is Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins by Garry Kasparov.
What's stopping companies from evaluting AI?
The main reason why companies fall short in adopting AI is that they lack knowledge is asking AI companies basic questions to evaluate their software. Here are a few thoughts on how to evaluate the AI solution offered by others to you. You can ask about the following 6 steps that any AI system uses to solve the problem. These are questions asked using common words and not technical jargon and hence the answer from the vendor should also be simple and plain to understand easily. If the vendor complicates the explanation then they have not understood your use cases properly. Let us dive into the 6 questions to ask the AI Vendor.
Question 1: Feeding the problem
How do I input my problem statement into the AI system? The answer should be a simple explanation of how to specify the domain of my problem first. Eg: I want to train the AI system to identify fruits and hence want to set the context to fruits so that when a user queries about Apple it shows the fruit and not the Apple Inc logo. The second question is how to feed the data about the problem you are trying to solve into the system. This should be as simple as exporting related data to a flat file of rows and columns and import into the AI system at the click of a button. Eg: FAQ documents can be input into a chatbot by providing a set of questions and answers in an Excel format. The third question is to know how the problem is stored in the system internally and is easy to export it to a file. This is where complex technical answers are delivered but need to be distilled to common sense.
Question 2: Knowledge Management
The system is being trained using various sets of data to solve the problem accurately. Knowledge can be classified into two types – Tacit which is like talent and is very difficult to transfer to others and the other one is Explicit which can be documented and imparted to others via training. We want to know how the system stores this explicit knowledge through is learning from data and understand the outliers, constraints, rules and pattern matching. If this can be explained by the vendor in simple terms and you can understand that easily then it will help you evaluate the tool deeply.
Question 3: Pattern Finding for Solution
How does the system use pattern matching to find the rough solution for the problem? This is the question that needs to be answered to validate the accuracy stated by the vendor. When we start engaging with clients we conservatively say that accuracy will be 60% in the beginning and only after extensive training we can take it to 90% or more. Ask for any similar problems the vendor has solved in the past and understand the accuracy improvement steps to be done.
Question 4: Problem Decomposition
This is taking a deep dive into the AI system where you get to know how your specific problem at hand will be decomposed by their system and if they can provide a simple flow of that then it will help you gain confidence in solving the problem through the AI system. If the vendor is open about it then ask some frameworks being used to solve the problems. Most AI companies don’t develop their own algorithm and they just train an open-source one and tweak it. You can ask the algorithm used to see if it is a popular and proven one.
Question 5: Identifying the best training data
To hit the ground running the best collaboration exercise you can do with the vendor is to identify the best training data for doing a proof-of-value project. The proof-of-value project is a mini project that we do for a prospect that has efforts less than 40 hours to show them the value of the product being adopted in their enterprise. Some POV projects take bigger effort and companies pay for it and evaluate it for the complete use case. Identify the problem you want to solve and then the data in your system that need to provide to the AI/ML system.
Hope this post provided you with guidelines on how to evaluate the AI/ML products for your company. In the next post, we will cover some use cases for which AI/ML is very useful. If you have any questions you can reach out to us on the contact page.
Founder & CEO
H.Thirukkumaran has over 20 years of experience in the IT industry. He worked in US for over 13 years for leading companies in various sectors like retail and ecommerce, investment banking, stock market, automobile and real estate He is the author of the book Learning Google BigQuery which explains how to build big data systems using Google BigQuery. He holds a masters in blockchain from Zigurat Innovation and Technology Business School from Barcelona Spain. He is also the India chapter lead for the Global Blockchain Initiative a non-profit from Germany that provides free education on blockchain. He currently lives in Chennai India.