4 Tech Giant Executives Resolve Four Major AI/ML Adoption Obstacles in the Enterprise
Artificial Intelligence (AI) no longer only appears in scientific textbooks but is now an evolving reality. Businesses are starting to realize how important this cutting-edge technology is as 56% of enterprises reported that there would be an AI deployment in at least one function, according to a 2021 McKinsey survey
Yet, AI adoption is not straightforward. What are some major obstacles keeping firms from utilizing the enormous potential of this new technology?
Four AI executives from large tech corporations sat down to dissect the 4 major challenges to AI/ML adoption and offered solutions in a panel discussion during Worldwide AI Webinar 2022.
1. Lack of AI literacy
, VP of AI at SAP
pointed out that the lack of organizational AI literacy was the first challenge. Oftentimes, even top-level executives mistake AI projects for IT or data project that has a clear start and end date. Rather, AI adoption is a continuous journey that involves research, tests, and trials.
“I see that when you build AI models and solutions from scratch with the teams, a lot of times, they don't know what they don't know. [...]. And then also there is this spectrum between that excitement of what you read in the press and the fear of uncertainty within the organization.” - Andreas Welsch, SAP’s VP of AI
, Cloud Adoption Director at Oracle
agreed with Andreas’ viewpoint, adding that the journey from planning to implementation is complex and demanding, thus requiring thorough comprehension from the deployment team.
“Creating models is very easy, anybody with some Python skills and a few hours of training can copy some code and run in their machine [...], but later when this needs to be put on production, a lot of things need to be solved, getting data from production systems, moving it securely, retrain model and of course scale, maintain a secure and high available platform, that it a lot of effort” - Nestor Camilo, Director Cloud Adoption Public Sector of Oracle
To tackle this problem, Aamar Hussain
, Director of Azure Data at Microsoft
, suggested not treating AI adoption as any software development project and understanding the current maturity level of the business. Assessing your organization’s skillsets, budget, AI awareness, and data availability is highly recommended to ensure successful adoption.
“The nontechnological emphasis and focus are really important. Don't jump into it straight away and think that you can just switch it on and off.” - Aamar Hussain, Director of Azure Data of Microsoft
2. Lack of strategic approach to AI adoption
This second challenge was specified by Aamar. He emphasized the need for a data-centric approach and a suitable framework, methodology, and strategy. Issues identification, data governance, ethical considerations, constant AI model monitoring, and management are equally vital.
“I think it's important to understand what the actual use case is and then choose the technology that's appropriate. I would say that any traditional mechanism should be broken down into what you would do with software engineering and with machine learning. So there's a segment of the problem that is easily addressable through traditional software engineering practices and best practices, and there's a segment which potentially could benefit from unlocking data. And that unlocking and activating data would be through machine learning. So if you can partition the use case into those two categories, what's traditional, what you do for scalable architecture versus what you have data for input and output data that you can then use to unlock capabilities within AIML kinds of scenarios.”
3. Fear of the uncertain
Fear is the third common problem. Andreas Welsch mentioned people’s fear of losing their jobs to AI while Ali Arsanjani addressed the uncertainty of return on investment. He suggested:
“If an executive is met with some uncertain technology, unproven technology, they're probably going to push back until the value of that technology is demonstrated.”
Nestor built on his point, affirming “risk of failure” is not an easy matter to overcome and offering a solution:
“We have to make it easy for companies to use ML. If you don’t have the experts or the domain knowledge, democratizing AI access to the majority is a must. [...] First, we must embed ML models in our business apps [...] Then we make it easy to use your data in your production systems to train and predict without the need to do complex ETL.” - Nestor Camilo, Director Cloud Adoption Public Sector of Oracle
Additionally, Andreas Welsch highlighted the importance of building the trust of business users. Explaining and informing business stakeholders of the tremendous benefits of the AI/ML system so the involved parties can start capitalizing on its potential are paramount, as declared by Andreas.
Ali Arsanjani had a different perspective. He believed that education should come first; with enough knowledge, comes the ability to weigh and make knowledge-based and experience-based evaluations, which can be used to build trust.
“To have a conversation on the adoption of technology, I would say that we need to do more in terms of not just trying to convince, but invest in education and upskilling; then we can have more aligned conversations with someone. If they're in a completely different frame of mind and a completely different knowledge base, there is almost no way we can convince them.”
Nestor Camilo seconded Ali’s opinion, saying that providing formal AI training is crucial as “you can do a lot better if they know what you are talking about.”
Aamar Hussain discussed how education can impact the culture in a positive way. He considered that culture held a critical role in the conversation. You can either have a dialogue where leaders force people to follow or you can have a culture in which the leader brings people on board with education and reinforcement.
4. Cultural barriers
The last major challenge is cultural barriers, or as Ali Arsanjani put it, “resistance to change”. Aamar Hussain further noted that some executives tend to stick with the way things have been done for years. It often takes some persuasion before they see that adopting new processes will be worth the overall gains they will bring.
He also spoke highly about executives being prime examples for their followers. You can’t preach about sustainability and be one of the main sources of CO2 emissions. The advocate for change and new technology adoption must be the change themselves.
Ali Arsanjani proposed a possible solution to this problem:
“So starting out small, using projects that are demonstrating key performance indicators that can be enhanced. So something that someone cares about. If I come to you with some proposition and you don't really care about that proposition or it's not really aligned with your business objectives, it's not going to be important. But if it is, you will listen. And if there's a project that will pinpoint some advancement in that area, you're going to tend to listen to it more. And then the next phase of maturity there would be the education and adoption and documentation of the practices, automation of the practices, gathering data on the tasks that have been done and then continuous improvement.”
All in all, the four speakers agreed that the best ways to tackle some major AI/ML adoption challenges in enterprises are understanding business needs, identifying the organizational concerns and seeing how the new technologies can help, bringing everybody on board with your journey, and decreasing the risk of disinformation and misinformation. Most importantly, businesses can achieve transformational development by anticipating adoption barriers and adopting a strategic AI implementation using the maturity model described by Ali.
“With a business-driven machine learning MVP you can convince reluctant people that AI is mature for a lot of business use cases and don’t try to do moon shoot project first, but baby steps that generate a lot of value and confidence” - Nestor Camilo, Director Cloud Adoption Public Sector of Oracle