I’ve had more than one expert tell me this: businesses that are not yet using artificial intelligence feel a compelling need – an urgent need – to begin using this emerging technology. Many of their competitors are already using AI to drive competitive advantage. The time is now to get on board.
But clearly, artificial intelligence is far too complex for most businesses to build in-house. They need to outsource their AI solution, from a cloud source or other specialized AI provider. This process is itself complex. Where to begin? How to select a provider? How do we know a provider will still be a good fit years from now, given that AI changes so fast.
In short, how can businesses best start their AI journey?
To provide guidance, I’ll spoke with one of the AI sector’s leading experts, Dinesh Nirmal, Vice President, IBM Data and AI Development; Site Executive, IBM Silicon Valley Lab.
DOWNLOAD THE PODCAST:
The Current State of AI: Early Adopters (4:52)
“If you look at one of the recent surveys, only 4% of the enterprises have really adopted and started developing in AI. If you look at AI right, you have to be an enterprise who is embracing AI. I think that’s the only way to survive going forward, especially with the competition. So, it becomes very critical for every enterprise to adopt AI. Many of the enterprise are talking about it, but they haven’t really adopted or deployed AI.
“Developing AI is about developing the models, having a data science team getting that right. But deployment becomes very challenging for enterprises, because they have applications that were written in the ’60s and ’70s. They have third-party software that cannot be touched. So, how do you infuse AI or deploy AI into the midst of it becomes really challenging for enterprises. That’s where a lot of customers or enterprises get challenged. I mean, developing a model becomes a little more easier, I’m not saying it’s really trivial but when it comes to deployment, it becomes very challenging.
“So I look at it as three Ds, I call it: The data, the development of models and deployment. Those are the three phases if you look at the AI. I also think that the first D and the last D are the most critical, which is, how do you get the data, the trusted data, how do you clean the data?
“How do you make it accessible to the data scientist in a really fast manner? And then the development piece is getting more and more commoditized, meaning you can use a lot of the open source tools that are available to build those models. But then the last D which is deployment becomes also very critical. Because now you have to infuse that AI or those models that you built into the enterprises, it’s pretty complex involvement. There involves software that was third party that software that you have no access, no skills, all those things. So it becomes very challenging.”
How Companies are Purchasing AI
“Let me start out by saying, we came up with something called an AI ladder. How do you climb up the ladder to get to where you described James, which is basically being able to be an AI-centric enterprise?
“The AI ladder involves: how do you collect or connect? Meaning collect your data, right? Or connect your data because it could be an existing data source that’s already there. The second is, how do you organize that data, right? The organization of data becomes very critical – what data is sensitive, which data is not sensitive. So you’ve got to organize the data or add some governance around the data to make it available.
“The third piece is to analyze the data, meaning, how do you analyze the data? What are the tools that you need to analyze the data? And then the last piece is to infuse the data or deploy those models that you have built. So we kind of come up with an AI ladder with these four runs.
“So when I say [purchasing AI through the] cloud, I would say it’s a hybrid cloud. Because think about it, this is data, right? I know [to have] the data to build models you have to take the data to train the models. And most enterprises are not there yet, whether they’re willing to move very sensitive data on the public cloud. So, it’s going to be a hybrid cloud approach whereby which you do all the training behind the firewall using the data, and then when you are ready to deploy the model, you can deploy it on public cloud.”
Challenges the Companies Encounter as they Deploy AI
“It’s a huge challenge for enterprises to get the right set of skills. The second one is culture that we talked about, how do we make sure that everyone is bought into the AI mindset? So that’s another one.
“The third piece is that, the deployment. This is a story that I often tell is that, I was with a major bank in Denmark. And I met with them and the CEO said, “Dinesh, I get it. We have developed the model in four weeks but it has been 11 months, we still haven’t deployed it.” And that’s the reality. You developed a model, but then how do you get that to the deployment phase? So, I would say the culture, the organizational aspect, the skills, how do you get the right set of skills? How do you deploy it? And how do you get that data, the trusted data to…our data scientists in the fast fashion? Those are the key challenges that enterprises face today in making AI a reality.
“I mean those [AI] skills are very rare. Statistics, mathematics, some level of data engineering. So it’s a combination of skills that data scientists hold and it’s very valuable, right? And so it’s very hard to get those skills.”
Near Term Future of AI, as Business Deploy It for Competitive Advantage
“I strongly believe that unless you are willing to embrace AI, the enterprises will lose their competitive advantage. We are seeing it more and more. Whether if you take a chatbot for example, even if it can only go a two-layered discussion, customers are noticing it. How can you make a happy customer out of a discussion?
“So, because now a chatbot, you can immediately go get an answer, instant gratification is there. You don’t have to get on a phone and wait 15 minutes for somebody to answer that question. So those kind of things becomes very critical for an enterprise to adopt. So adoption of AI in the next four, five years is going to accelerate probably from 4% to I don’t know, 40% or 50% because it becomes very critical for everybody to embrace.
“Now, there are places where we will see more and more AI, for example, customer service, because that’s a place where we could say, do sentimental analysis to say, “What was the sentiment?” And based on that sentiment, maybe we can offer this, right? So there might be areas like NLP sentiment analysis where there will be much more adoption.
“So there is another customer I met with and they said to me, “Dinesh, one of the biggest benefit of AI, we believe, is the skills.” Because if I have an underwriter who has 30 years of knowledge, tomorrow, he or she quit the knowledge walks out with them. But if I train an AI model, it’s there for lifetime. It keeps learning, learning, learning. There is no retirement, there’s no leaving the company.
“That model is going to be there and continuously learning. So the skills becomes very critical, retention of the knowledge and skills becomes very critical. So that’s why I said, AI adoption becomes very critical for enterprises, because that’s what will drive them from a competitive nature to make sure that they survive and thrive. The other example is like, for example, if I’m a manufacturing company, how do I detect defects before it goes out to the customer? And that could save me millions of dollars a year, if I can find those defects before it goes out.
“So that’s like pattern detection, right? Pattern analysis, those kind of things. I think we will see it in the industry from an AI adoption perspective. I don’t think any industry will not adopt AI, and I do think the acceleration of AI will happen across enterprises.”