Transforming your organization with AI

Authored by Praveen Srivatsa Director, Asthrasoft Consulting

Artificial Intelligence (AI) is not a new technology. However, AI solutions need a lot of computing resources - which the cloud has now made available to everyone. This has brought AI into the mainstream. Many organizations are now focusing on solutions that leverage AI to transform their business processes. 

While teams are building up a technical understanding of AI, businesses are unaware of unlocking value and monetizing AI solutions. In order to leverage AI, we need to understand the applications of AI. We then need to go about creating an AI driven culture in the organization. Let's start by understanding what AI can do for us. 

Explore what can Artificial Intelligence do 

Artificial Intelligence allows computers to mimic the human behaviour of detecting complex patterns and interpreting them. AI can interpret vision and speech by analyzing images, audio and video. It can use sensors to identify and analyze temperature, location, presence and movement. We can also use AI to drive drones or robots without human intervention. So AI has the potential to perform a whole set of actions - sometimes more efficiently - that were being performed by humans. 

However, AI looks at each action in a limited way. It lacks the common sense that a human being brings to the table. And when something unexpected happens, AI systems tend to behave unpredictably - as was experienced by the Microsoft Tay bot, which was shut down, just 16 hours after launch. So rather than thinking of AI as replacing human actions, it should be looked upon as a technology that can supplement them and make humans more efficient. 

Consume AI services and build a business case 

There are many vendors that provide AI services around vision, speech, text, language and conversations. There are also specific implementations around object detection, sensor tracking and pattern analysis. Instead of reinventing the wheel, businesses should start by consuming these AI services to automate processes within your organization. 

You can use vision AI with facial recognition as a means to track employee attendance or use it with object detection for inventory management. You can use Natural Language Processing (NLP) or the Document AI services to introduce speech or document recognition into your existing applications. Conversation bots and sentiment recognition can provide a new generation experience over your existing application interfaces. This allows the business to understand the value of AI. 

Refine AI solutions for your context 

As you delve into using external AI services you will soon realize that they are too generic and are not optimized for your scenario. For example - while a sentiment analysis API does a great job of identifying human sentiments in neutral settings, it is not a great fit for patients in a hospital. Once you have used an external AI service, you can now refine it for your context. 

This is where you bring in your data, train the model for your context and improve the AI algorithms for better accuracy. 

This is also the point where you will need to start collecting lots of data in order to fine tune the model. In the AI world, this is called labelled data - data that has lots of details of the context (for eg. sentiment based on gender, time-of-day, patient condition, age etc). While anyone can use the generic APIs, your refined algorithms - using the labelled data that you have collected - is invaluable to your organization and cannot easily be replicated by anyone else. 

Manage the bias in your datasets 

Data is the fuel for AI applications. However, just like oil needs to be refined before it is of any use, the data needs to be managed and organized before it can feed your AI algorithms. Many teams approach projects with the thought that they already have lots of data - so they can now build AI solutions around it. Unfortunately, most current sources of data are like raw, unprocessed oil and are not of much use for training AI algorithms. 

To refine the AI algorithms, we need lots of ‘relevant’ data. For example, many countries are realizing the the weather data collection was not accurate in the past. So even if we have lots of such data, it is not really helpful for future weather predictions. Also many data sets have inherent bias built in. Most of the traffic data available has an inherent bias towards developed countries. Similarly medical data is biased based on the gender, race and cultures. In order to feed our AI algorithms we need lots of unbiased labelled data for our specific context. 

Build trustworthy AI systems 

Different forms of AI solutions have been around for a long time. Spam filters have used algorithms to detect unwanted emails. However the day we realize that a few of our important emails are being marked as spam, our trust in the filters goes down dramatically. All AI systems are only as useful as they are trustworthy. When we build and deploy AI algorithms, it is important to ease this into our existing business processes and build trust by all the parties involved. This goes a long way in ensuring the success of our AI solutions.

It is also important for an organization to recognize that implementing AI solutions represents an organizational behaviour change. Most AI solutions claim a high degree of accuracy - usually above that of humans performing the same action. But what about the few inaccuracies that creep in. When a person causes an accident, he is responsible and liable for the damages. What about a self-driving car that causes an accident? 

So when working with AI technologies, we need to understand their capability and use them to supplement and empower humans to do more. This has the dual benefit of bringing in a more efficient organizational culture while maintaining the human chain of responsible behaviour. 

In summary, Artificial Intelligence is a powerful tool that can help us solve many problems and enable hitherto unexplored efficiencies and opportunities. When we think software projects, we think about process changes. When we think AI, we need to think about behavioural changes. 

To start with, explore what AI can do by inducing small behavioural changes with external AI services. Refine this with custom unbiased labelled data that now makes the AI solutions more relevant to your context. Continuously train the models and supplement your teams with recommendations from the AI engine. Finally, build a trustworthy AI system that can open up enormous opportunities for your organization.