AI has now made it onto CEOs' agendas. While the topic certainly isn't new, CEOs have learned that the idea of AI is far simpler than its effective application. To get there, companies need to start with their business objectives and then use AI in ways that advance those objectives rather than just implementing AI for AI's sake and hoping it can add value later.
Meanwhile, CEO attitudes about AI and machine learning or ML (a subset of AI techniques) have been changing as it relates to digital disruption. In the beginning, it was about understanding what digital disrupters do and how they do it. Now, they're beginning to realize that they need to create value on their own terms. That's not to say that they won't take advantage of some of the accelerators the digital giants have made freely available. However, a me-too only strategy is ultimately a risky proposition.
"The top-most priority for the CEOs of leading companies is reinventing the future of their companies, powered with AI, powered with data," said Arnab Chakraborty, global managing director, applied intelligence North America Lead at multinational consulting firm Accenture. "It's about unlocking the value with AI by looking at how they optimize their existing business whether it's in sales and marketing, supply chain, finance, HR, all those functions."
Not surprisingly, AI is also on CIOs' agendas. In fact, global professional services company Genpact recently published a report with MIT Sloan CIO Symposium in which 48% of the 500 CIOs surveyed said that AI is their No. 1 investment priority.
"CIOs are saying, we've got to invest in it. The question is, why now?" said Sanjay Srivastava, chief digital officer at global professional services firm Genpact. "Three things have happened: the technology has gotten very good; it has become very affordable; and the need has gone up."
Following are 10 AI and ML trends to watch.
Partnerships Will Change
Companies have historically had a core competency, such as selling groceries or selling prescription and over-the-counter drugs. Yet, their products overlap somewhat for customers' convenience and to boost the retailer's share of wallet. Traditionally, there hasn't been much synergy between the two types of companies.
"Walgreens is collaborating with Kroger to provide a seamless experience for customers who are looking for pharmacy products and groceries. We'll start to see those kinds of partnerships [create] new data-driven business models that are going to be [powered] by AI," said Accenture's Chakraborty.
To do that, enterprises need to do three things:
Low-Code/No Code Data Science
Low-code and no-code software development has been disrupting organizations, enabling power users (low-code users) and average line-of-business professionals (no-code users) to build simple applications. A similar trend is occurring with data science on different levels such as data integration, augmented analytics and even model building. However, there are some concerns.
"If the company's main business is not in AI and they are interested in applying AI for business insights only, using one of the no-code or low-code AI and ML platforms may serve their needs. However, they should be careful to test them with validated benchmark data, since this is essentially a black box approach with only limited possibilities to customize the inner workings of the models," said Silke Dodel, PhD and Machine Learning Architect at multilingual customer support tool provider Language I/O. "Companies that want to make AI part of their business model could start to use the existing models churned out by the giants in the field and use transfer learning to adapt them to their purpose."
Investing in a few highly trained ML data scientists that understand the mathematics behind the models as well would help ensure that the best model could be used for the user's niche application.
Transfer Learning
Transfer learning applies machine learning to a different, but related problem. Practically speaking, companies will use (and are using) some of the ML models built by the digital giants instead of building them from scratch.
"One of the biggest issues in the AI/ML space has been the cost of developing AI solutions that can effectively translate locally in regions across the globe. Voice AI systems struggle with understanding accents and discerning dialects -- even within covered languages [for which] they have trained data sets," said Ahmer Inam, chief AI officer at global digital and technology services company Pactera EDGE. "Generally, the current paradigm of how we build, train and scale voice AI systems needs to change. There are advances that are happening [such as] training data, algorithms, training frameworks, and how voice AI experiences are designed and built with the end-user in mind (a human-centric approach).
One step in the right direction is Facebook AI's new wave2vec, which learns the structure of speech from raw audio. This unsupervised approach to machine learning rivals the best supervised systems that are time, data and compute (hence cost) heavy. Inam said that wave2vec will enable significantly faster training of voice AI applications with smaller data and thus lower compute cost since it doesn't require transcribed speech data.
Better ROI
Some organizations struggle to drive ROI from their AI and ML investments. Sometimes the problem is unrealistic expectations. Other times, it's the inability to translate what's happening in the lab to a production environment.
"Artificial Intelligence technology is moving from science to actual engineering. In recent decades, the focus was on R&D for new algorithms and techniques. Today, the focus has pivoted back to data strategy, preparation of data and data use," said Ben Johnson, CEO of data analytics and custom software firm Freya Systems. "Data scientists use a catalog of algorithms and techniques to approach problems and develop solutions, applying the well-established algorithms to data. What is significantly important to me is focusing on problems and reducing the 80+% failure rate of projects. This refocus[ing] will result in an actual ROI for AI and ML projects."
Organizations Will Access More Data
Many organizations have already supplemented their own data with third-party data. While the trend has been enabled by the API economy, organizations will share more data with others in the future.
"This is not just about generating a higher volume of data but rather generating and using data more strategically," said Justin Silver, an AI strategist and data scientist at AI-powered digital selling platform PROS. "In addition to organizations improving their own proprietary data, we will see collaborative efforts on improving data -- sharing of data, say between businesses, for mutual benefit while respecting privacy/confidentiality. For example, airlines sharing data to identify trends that helped them navigate impact of COVID-19 on their business."
AI-Savvy Business Professionals
The human-machine partnership is evolving, and with it, expanded knowledge and a richer skill set.
"The real question is, how do you scale yourself and your employee base to get ready for the future of work?" said Genpact's Srivastava. "The world needs more people that understand finance, accounting, and machine learning so they can apply machine learning to cashflows. The world also needs mechanical engineers who understand computer vision so they can apply that to automate manufacturing and workflow."
More AutoML
As organizations scale AI-led transformation, they'll find that they need to automate more work.
"A lot of the initial part of the AI value chain will be automated -- data ingestion, capture, building the data pipeline, building the ML pipeline so that the [human] talent can focus more on the higher end of the value chain which is building the ML models, doing the feature engineering and testing the models rather than doing all the plumbing work that is required to get to the model," said Accenture's Chakraborty. "We're already seeing the platform players -- Google and others -- creating a lot of ML capabilities which go a long way in automating those standard activities so businesspeople and data scientists can work on more high-end, value-added activities."
Federated Learning Focuses on Intelligence at the Edge
Federated learning enables models to be pushed to the edges where it will be trained without necessarily requiring any data sharing across the ecosystem. The trained models are then pushed back from the edge to a central repository, minus the training data.
"That creates a much faster way of training and deploying the models without risking any data privacy concerns," said Accenture's Chakraborty. "[If you're in the healthcare industry,] that helps you to create a new service that is powered by AI to serve the patients in a much more effective way. You're creating new experiences and better patient outcomes."
Cybersecurity Will Become More Complex and Intelligent
Cyberwarfare is happening for real, and not just on the electronic battlefield between political foes. Cyberterrorists and cyber spies are attempting to upend businesses by adding AI-powered attacks that are increasingly subtle, complex and nuanced.
"AI is an application that runs on data to generate results so now it is no longer about the application," said Genpact's Srivastava. "I'm just going to corrupt your data, so now your AI is going to pick the wrong data and come up with the wrong results. You've just expanded your threat surface. It needs to be factored into any enterprise AI governance plan."
AI Will Triage Medical Cases
AI can "see" what doctors can't see because, for example, in a computer vision context, there can be subtle clues in X-rays, for example, that can make a difference in a diagnosis. However, more generally, AI will make doctors more proficient at what they do.
"It is an inconvenient truth that AI is much better than human doctors at triage and initial diagnosis. This is not an attempt to take jobs from doctors, but rather to increase the throughputs for doctors by relieving them of the tedium and the time-consuming efforts associated with these tasks," said Christopher J. Hughey, president and CEO at advisory and consulting firm Fast Layne Solutions. "This will be critical in the decades to come, as the US faces a critical shortage of physicians and other healthcare workers.
Why the shortage of healthcare workers? Well, COVID-19-related stress levels didn't help. In fact, the Association of American Medical Colleges already expected a physician shortfall of between 54,100 and 139,000 by 2033 before the pandemic hit. More recently, a Washington Post-Kaiser Family Foundation poll indicated that three in 10 healthcare workers are considering leaving the profession due to burnout, PTSD, and other stressors related to the pandemic, Hughey said.
Check out other InformationWeek slideshows.
Lisa Morgan is a freelance writer who covers big data and BI for InformationWeek. She has contributed articles, reports, and other types of content to various publications and sites ranging from SD Times to the Economist Intelligent Unit. Frequent areas of coverage include ... View Full Bio
Comment Now