The Right Artificial Intelligence Approach

By : |March 19, 2019 0

By Kamal Kishore, Head of Data and AI Practice, ThoughtWorks

AI is among the most talked technology nowadays. It is being implemented by various industries and it has resulted in tremendous business growth. But without the right artificial intelligence approach, it won’t be possible. As we all know Artificial Intelligence has gripped our interest for much longer than it has been a part of pop culture – read; sci-fi books, shows and movies. Today, the only-limited-by-imagination-kind-of possibilities of AI is not only recognized but is becoming commonplace at most enterprises.

Let’s step back for a moment and look at how Gartner’s defines AI; ‘a technology that appears to emulate human performance typically by learning, coming to its own conclusions, appearing to understand complex content, engaging in natural dialogues with people, enhancing human cognitive performance (also known as cognitive computing) or replacing people on execution of nonroutine tasks.’

Add to this, the confluence of a few factors that are listed below, which encourage AI and Machine Learning or ML to be at the center of most business conversations –

                                 

___________________________________________________________________________________________________________

# Easy access to expansive compute and storage

# System connectivity that allows the collection of massive terabytes of data

# Advanced learning algorithms and tools that effectively make use of the above two resources

Why wouldn’t an organization want to take advantage of such technology? According to Gartner, Inc, AI-derived business value is set to reach $3.9 trillion in 2022. Given this burgeoning scenario, here are a few use cases of AI and ML that are providing enterprises with immediate value and positioning the latter as game-changers in their respective industries.

Enhanced customer experience:

* AI-powered chatbots and digital assistants

      = ensure seamless customer service with automation, speech recognition and improved call routing

Optimized supply chain/logistics:

* AI-driven logistics

      = provide value across the supply chain – from forecasting to sourcing raw materials, production, warehousing to distribution logistics

Efficient manufacturing and predictive maintenance

* AI-powered smart robotics

= speed up manufacturing processes

= eliminate human errors

= save on resources

= capitalize on analytics to predict maintenance requirements

Stronger security:

* AI diffuses threats

= detects unusual activities and patterns

= prepares multiple response-strategies to any eventuality

 

How does one get their AI approach right?

AI solution needs copious amounts of data to be successful. Managing vast data, ensuring the right data is used for the job and identifying the right tools for the task are amongst an organization’s biggest hurdles. Here are a few ways executive teams can address these issues –

Understand AI and its scope: leaders need to understand the realistic possibilities of AI within the context of their business and market scenario.

Take a long-term view: leaders’ clear vision should guide the choice of new-age and advanced analytics tools to deploy. Also, they should provide timely sponsorship of AI initiatives across the organization.

Build capability: building a holistic team comprised of data scientists, data engineers and analytic experts, will help enterprises identify the areas that AI can impact, cull out and break down the right data, and build insights.

Define a Data Strategy: an efficient data strategy will build a glossary of data stores that’s spread across the organization and establish an enterprise-wide framework to define both, data and the format for collaboration.

AI-ready infrastructure: enterprises should support their AI plans by ramping up storage and processing capabilities to support AI use cases

Identify the right pilot project: a pilot AI use case that is low risk with high potential ROI, has clean and consistent data available is a sure way to get some traction on organizations’ AI journeys. Examples of such projects include adding intelligence to an existing application, or to a manual-labour intensive process etc.

Go the last mile of analytics: leaders should empower cross-functional access to their intelligence and analytics to make use of newfound insights that could improve the way business is done.

Now, that we have an approach down, how do we efficiently build an AI use case?

# Identify appropriate datasets that can help run the learning algorithm, tune parameters, select features and make the right decisions that will positively impact the algorithm’s performance

# Iterate fast by crunching multiple performance metrics into one. Evaluate the algorithm’s performance against these metrics.

# Investigate errors that will define measures to improve the algorithm. To ensure effective diagnosis, choose similar sources of data to test and train the algorithm.

# Choose a large enough data set (that can be split into small groups) and will help discover changes in overall accuracy and performance

# Define general performance parameters that will help the algorithm identify bias, variance and data issues

# Manage data unavailability by working with artificial data and allowing for sufficient randomness.

 

Summing up, the reality is that AI is both a complex and iterative process. The key is to iterate fast rather than to strive for a perfect model.

AI

No Comments so fars

Jump into a conversation

No Comments Yet!

You can be the one to start a conversation.