As modern organizations adopt machine learning to optimize, automate and facilitate better predictions (such as during a global pandemic), AI is providing a clear competitive advantage.
Ritika Gunnar, VP Data and AI Expert Services & Learning at IBM, has worked with thousands of organizations across different industries and nations in her 21 years with the company. She has found that there are three core ingredients that must be present in order to successfully integrate and get AI projects to scale.
2. Do you have the right process?
3. Do you have the right data?
Start with culture
This cannot be underestimated. “Culture is one of the most important things to get AI projects off the ground and to scale,” says Ritika. It’s not enough to just have a group of data scientists or specialists in your organization.
“You need to have buy-in at every level starting from the foundation.”
As AI projects have multiple iterations, it’s also key to develop projects that are both bottom-up and top-down in order to give your people empowerment to fail. Look for buy-in from those who will be using and managing production applications, she suggests. “The [AI] skills trajectory [applies] to everybody in the organization, and that is a cultural transformation.”
Prepare for change
As both data and process will change during the creation of successful AI applications, it’s important to both monitor and manage your project differently, says Ritika. “It requires iterative, rapidly experimental capabilities that only agility can get you.”
As AI is an ever-changing component that learns, understands and evolves based on the data it sees, organizations can’t have good AI without having good information architecture.
“That means the dev-ops process changes, as we are not working with something static,” Ritika explains. “The same person who built an AI application now needs to understand what it means to have an ever-changing component through development, testing and production.”
And no matter what, data is the foundation.
Once they have data, successful organizations iterate on their processes. “Allowing time to iteratively try things, release them and perfect them becomes really important,” says Ritika. It is imperative to start on projects that are important to the organization, but not so important that there is reluctance to iterate them.
“Do not start with your most complicated problem or success will be difficult,” she cautions. “Start with problems that are more tractable in nature, where you can figure out how the organization’s processes, applications and management needs to change.”
Start small, create a foundation, and then when you set up what Ritika calls “a centre of excellence,” you can scale the culture, process and project to the rest of your organization.
“Entrepreneurs need to get on the train of leveraging these digital and AI technologies now,” she says. “[It’s] the only way to stay relevant and then to stay competitive… so take advantage of new AI technologies in this day and age.”
During Ritika’s C2 CONVERSATIONS – LIVE session on May 27, she was asked a number of topical questions by the audience. Here are a select few that she took time to answer on and offline:
What timeline do you see for mass adoption AI use? Does the pandemic affect this development?
“[In] the past five or seven years, machine learning was being used for proof-of-concept engagements and scenarios that are differentiating but not mainstream. I truly believe that this is the decade of ubiquitous AI, which means that it will be integrated into everything you do; everything that you interact with will be embedded with AI.
What we’ve seen is that organizations that were going to adopt AI at a slow pace are now adopting it at a much more rapid pace. For use cases, that involves a few things: cost savings, risk management, and the optimization of information sharing. These are three areas we’ve seen AI being used more now, and at an extremely accelerated rate.
I think we’re going to see a lot more AI adopted exponentially as information out there needs to be disseminated to organizations, customers and employees alike in a much more rapid way.”
What should organizations with smaller data sets do to compete in AI and is that even possible? What AI tech/solutions would you like to see companies start to implement more often, particularly small businesses and startups?
“No amount of AI will overcome the lack of data. Many clients that we work with have a lack of accessible data, not necessarily no data. For that, building a proper information architecture is critical. In other cases, however, large data sets do not exist. Building machine learning models from scratch with limited data is incredibly difficult if not impossible.
“When helping clients with their AI journey, oftentimes we’re using pre-trained models. We’ve already put in the heavy lifting of training a general model with a massive data set. We architect the models in a way where they are able to be extended with client data related to a specific scenario or industry. This enables us to train high-performing models with less data.
“I’ve seen many small businesses benefit greatly from this model. For example, many of our clients are currently experiencing higher-than-normal customer or employee inquiries that their staff or customer support employees are not able to handle. In a week, we can stand up a Watson Assistant instance to help answer customer questions related to COVID-19 closures, business procedures and more. Using pre-trained models, we can bring a lot of business value with assistant AI.”
How has your journey been getting to where you are? Can you talk about some of the challenges you encountered and how you dealt with them? How do you pave the way for more women coming in?
“I have come to believe diversity — not just in terms of gender, but broadly conceived — is essential to the development of high-quality AI. On my own team, I’ve seen how people from different backgrounds, with different points of view, can challenge one another to create the best ideas. It’s also critical to have diverse teams involved in the development and implementation of AI to help ensure that human bias doesn’t creep into the technology.
“If we want to increase the numbers of women in AI — and increase diversity in the field across every dimension — we must celebrate the diversity that exists. We must make sure that people representing different groups and backgrounds have supportive tech communities where they feel comfortable asking questions, making mistakes and venturing into unfamiliar territory — all necessary parts of learning. These support systems are critical. It’s all about creating a culture where people feel they can continually stretch themselves.”
(*You can read more about this here.)
How can we make sure to build inclusive AI?
“I believe in the 3Cs: community, confidence and continuous curiosity. First, you need to be able to create the community. If you are new to AI, regardless of your background, you need to find a community that you are comfortable with and that you can relate to. Next, you need to have an innate sense of confidence — the ability to see yourself in those roles. And continuous curiosity. The lead time for any organization to have tech skills is three to four years, but for AI, it’s 18 to 24 months. If you are not continuously learning, you aren’t staying relevant. So to have hunger to stay in the field is always extremely important.
(*You can read more about women in machine learning here.)
How do you address the cold start problem?
“I find that a lot of problems in delivering technology can be solved with this formula: successful outcomes = skills/expertise + technology + methodology.“When we think about delivering the outcomes clients need, we have to consider more than just the technology and the techniques we’re using to architect a solution. We have to have the right systems in place to ensure the technology is performing well. This includes people: having the right people to understand not just the technology, but also the business problem. You need to be able to define the problem in a detailed way that includes an industry lens and outcome-based approach. The right experts can examine that data you have and then consider the following: Where are there holes? What problems might I run into? Where and when should a human be in the loop? How will I measure and track performance?“The point is that SMEs need to bring a methodology that takes a data-centric and design-centric approach, so when an algorithm encounters something like the cold start problem, you have the systems in place to quickly address it.”
WATCH AND LEARN
Dive even more deeply into AI and transformation with these wide-ranging, featured video playlists from IBM.
Check out Ritika’s full C2 Conversation
Questions or comments? Drop us a line at firstname.lastname@example.org