Top 5 challenges of implementing AITop 5 challenges of implementing AI
An op-ed by John Harding, NVIDIA regional director for UK and Ireland
March 21, 2022
John Harding is the NVIDIA regional director for UK and Ireland.
The adoption of AI is for enterprises of all sizes, not just reserved for big global players. The technology can keep you ahead of competition, provide better solutions to customers by gaining actionable insights from data and maintain a leading position in the market.
But with new technologies come new challenges: Only 53% of AI projects make it into production, according to Gartner. The following are key challenges shared by European business leaders.
1. Recruiting and keeping top talent
How to identify, recruit and keep top talent for such a rapidly advancing technology is a challenge many businesses are facing. Knowing what questions to ask when recruiting new staff isn’t always easy when the subject matter is complex.
When working on rapidly developing, cutting-edge technology, it becomes hard to find people with specific experience in those areas. Recruiting for roles working with new technology needs an open mindset, so that hiring managers can create a team of skilled people that can work together to develop skills and drive the business.
When you have a team in place, then comes the retention. Supporting people with the resources they need, regular training and continuing to work on cutting-edge, exciting innovations such as conversational AI, autonomous vehicles and fundamental research into AI is key. It’s down to the company to give the individual the means to do their life’s work.
2. Navigating local data security regulations
Data regulations vary from one country to another. This can make it frustrating for global companies to gather data insights from around the world in a timely manner. Driven by regulations that change frequently and differ by country is a big hurdle.
European legislation, GDPR and the banking sector are very protective. This strict control means businesses aren’t able to implement AI solutions at the level they want, because they’re unable to get the data needed to do so. And while regulations do need to protect consumer rights, they should still work for businesses to be able to use the data they’ve collected.
Data is collected in different ways across different regions, which means when the data is consolidated, it has to be mapped and cleansed. That’s a lot of groundwork to get to the anonymous data needed for analysis.
External regulations, internal governance and organizational structure can significantly slow down the time it takes to deliver data. Federated learning, a technique most commonly used in the medical industry, is knocking down barriers to anonymized data and rapidly increasing the usability of secure data.
It decentralizes deep learning by removing the need to pool data into a single location. Instead, the model is trained in multiple iterations at different sites. Especially for global organisations with data sources around the world, this opens up the possibility for teams to build larger, more diverse datasets for training their AI algorithms.
3. Using the data effectively
If it takes analysts weeks or even months to get access to anonymous data, they can find themselves in a place where they’re trying to solve last year’s problem, not tomorrow’s. Understanding previous data is important in itself for businesses, rather than trying to use incorrect data to discern future trends.
Federated learning can play a role here, too. Companies and research institutions developing AI models are typically limited by the data available to them. To gather enough training data for a robust, generalizable model, most organizations would need to pool data with their peers.
But in many cases, data privacy regulations limit the ability to directly share data — such as patient medical records or proprietary datasets — on a common supercomputer or cloud server. That’s where federated learning comes in. Privacy-preserving federated learning techniques can enable the creation of robust AI models that work well across organizations, even in industries constrained by confidential or sparse data.
4. Building AI into the business strategy
Arguably the most important challenge to implementing AI is aligning and building it into the business strategy.
The success that comes from using AI is highest when businesses integrate it fully into their strategy, rather than creating an AI solution and retro-fitting use cases into it. Looking at a longer-term approach is one way to get senior management on board. Understanding where the business needs to be in one to five years, and what is needed to get there, can highlight the importance of a long-term strategy.
An AI solution isn’t a bolt-on to the IT department within an organization. It should be used to enhance opportunities business-wide and implemented collaboratively across both IT and the senior leadership teams.
Investing in AI means better customer service and improved internal operations, enabling businesses to reduce costs while increasing outputs. A well-invested and strategically implemented AI makes clear the benefits of business profits and performance.
5. AI and data centers: On-premises vs. cloud
While on-premises data centers can reduce the speed of access to data, it’s debated whether the cloud is safe enough, and meets the rigorous data regulations that a global organization might face. Combining data sets in the cloud, for example, is possible in some regions, but not all. Getting data to the cloud at scale is a challenge particularly due to international regulations. Even after following those regulations, it can take six to nine months to get access to the data.
While an on-premises data center offers faster access to data and less pressure from regulations, it cannot offer the same level of algorithm training as the cloud. The strength of an AI algorithm comes from the amount of data it is fed, meaning a cloud-based algorithm with access to a wider range of data is more likely to be better trained.
As AI technology advances, the processes and costs are improving, giving organizations more opportunity to harness their data in the way that works for them.
Through the democratization of AI, access to AI models, processes and resources becomes simpler. AI models for example, often don’t have to be built from scratch. Leading AI organizations such as NVIDIA or Google have built huge catalogs of models that can be tailored to specific business needs for their own data.
Guest commentaries published by AI Business are vetted to meet our editorial news guidelines.