Challenges of Implementing AI: How to Overcome Them

Artificial intelligence (AI) is the icing on the cake of every data-driven business. According to the 451 Research Report[1], 68% of the interviewed companies said that they are already using artificial intelligence or intend to within the next three years. And 92% of those who have already adopted this technology have upbeat views regarding the performance of their artificial intelligence projects. However, artificial intelligence implementation remains a common challenge for many organizations. That’s why in this article, we are going to show you how to overcome artificial intelligence challenges.

According to a Deloitte report[2], about 94% of companies encounter potential challenges during artificial intelligence implementation. From AI algorithm bias to creating the appropriate data inputs and workflows, artificial intelligence implementation can be stressful, particularly if you’re not conversant with its inner workings.

Challenges of artificial intelligence implementation

According to 451 Research’s report, businesses face these three major problems of artificial intelligence implementation:

• Shortage of skilled personnel
• Data access and preparation
• Limited budget

Shortage of Skilled Personnel

The respondents on the 451 research report pointed out the shortage of skilled personnel as a top concern in artificial intelligence implementation. 40% of the respondents cited it as a hurdle, while 21% said it was a major bottleneck. Moreover, another survey by Forbes Insights[3] says that 67% of 300 executives concur that the quality and availability of skilled personnel is a major problem of artificial intelligence implementation.

This is not surprising at all since the number of data scientists is low compared to the demand. Therefore, they have not been able to meet the rising interest in artificial intelligence implementation. Companies require data scientists and data engineers to analyze data and generate meaningful insights and developers to create AI systems.

Data Access and Preparation

33% of the respondents in the 451 report mentioned data access and preparation difficulties as a stumbling block in artificial intelligence implementation. Artificial intelligence requires data to understand things. It depends on a large body of datasets from which it monitors trends and behaviors and makes correct insights.

“Insights generated from AI are only as accurate and timely as your data sets. Bad data means you will not have a trustworthy, transparent AI.”

These are the words of Rob Thomas, the author of The All Ladder: Demystifying AI Challenges[4].

Another key element is the quality of the data applied in predictive analytics. Ideally, the data sets should be highly representative and balanced, or else the AI will end up adopting algorithm bias.

For example, datasets used by artificial intelligence technologies can contain human-based decisions influenced by sex, race, gender, or private information. According to a report by Reuters[5], Amazon discovered that its AI recruiting tool was not rating women for job openings in a gender-neutral way. It penalized resumes that included the term ‘woman’ or ‘women’. In simple terms, the AI engine was biased against women applicants. So, this clearly depicts how poor data can be a limiting factor in AI implementation.

Limited Budgets

32% of the interviewees also cited limited budgets as an impediment to artificial intelligence implementation. AI technology is a costly affair, and therefore firms need to set aside a separate budget for its adoption and implementation.

It might be interesting for you: Reduce Operating Costs and Improve Efficiency Using AI 

Challenge of implementing AI? Data management to the rescue!

Artificial intelligence and data management have a symbiotic relationship. When integrated into a data management tool, AI can optimize the path used by a query to access data, thus increasing efficiency. It also improves query outcomes and accuracy, even if data dwells outside the original search parameters. On the flip side, data management is crucial for a successful AI implementation. Also, without a robust data management infrastructure and strategy in place, AI adoption would be a difficult feat.

Ideally, you should opt for an ecosystem that has built-in AI technologies. This will allow you to run data management within one system. You’ll also be able to query data where it dwells, eliminating the need for time-consuming and expensive data movement.

A data management ecosystem featuring a built-in AI as described above can help ease the earlier stated three main problems of artificial intelligence implementation. The shortage of data engineers and scientists is mitigated in the following ways:

• Database developers and scientists have more time on their hands to focus on value-added tasks. This is because AI promotes the optimization of regular tasks like query optimization and the integration of development languages and systems.
• Increased connectivity involving different data warehouses in the environment allows personnel to derive their own data insights. This consequently cuts down the numerous calls to database engineers and scientists.

The challenge of access to quality data is mitigated by picking the right data management network. For example, data virtualization brings unprecedented connectivity, allowing unlimited access to various data types across several warehouses. All of this is possible without the need for data movement. Consequently, this simplifies data access and preparation.

Of course, any data management solution cannot solve upfront budget constraints. Instead, it can show efficiencies and return-on-investment (ROI) that make the investment in artificial intelligence more encouraging. For example, highly accurate predictions generated by AI-integrated data management solutions can open up new, money-making ventures that could have gone unnoticed.

Other bottlenecks that affect artificial intelligence implementation include:

1. Scalability of AI

Despite the significant interest in artificial intelligence, a lot of implementations remain in the experimental or pilot phases. An AI implementation may be seamless as a proof of concept. However, scaling it to tackle business tasks on an industrial scale may present a big challenge. Around 58% of the executives interviewed in the Forbes Insights survey indicate that their organizations are keen on AI adoption, but its implementation has been at a snail pace.

To function appropriately, AI needs a back-end infrastructure that supports:

• Data acquisition
• Robust applications
• Model development
• Real-time release of insights as per enterprise requirements

Oftentimes, a proof of concept or an experimental project is performed on small servers or personal computers. However, scaling to the business production stage can only be successful with the adoption of more powerful GPU and CPU processing that can meet the heavy lifting of business requirements.

The solution

To tackle the scalability issue, executives should cast their nets wide in search of new market systems and cloud-based technologies that can provide the much-needed high-grade computing capabilities as well as more bandwidth.

Begin with smaller, feasible projects and expand from there. Failure should not discourage your efforts. An IDC survey reveals that organizations experience a 50% failure rate with their AI projects[6]. Therefore, organizations should not stop experimenting while learning from their previous AI efforts.

2. Routine Updates For AI Algorithms

The shelf life of AI algorithms is pretty short. This is because markets, employees, customers, and business conditions change quickly and rely on algorithms running on outmoded regulations. That means there’s a higher chance of AI algorithms going stale. Company employees often undergo training to learn how to work with machine learning systems. The same is true for AI algorithms which also require ‘training’ to stay up to date with the constantly changing business landscape.

For this reason, organizations are faced with the challenge of repeatedly updating the models on which AI algorithms run using new data training sets. The training could be weekly or even daily. Typically, the new data training sets are designed using a labeling process that involves text, image, video, or audio annotation. Photos, for example, constantly require labeling using identifying text.

The ability of AI algorithms to provide correct, timely, and unbiased insights depends on continually refreshed or updated training sets. As such, the data scientists and AI developers will spend a significant amount of time developing and fine-tuning models.

The solution

With the shortage of data scientists vis-à-vis the demand for updating AI algorithms, the solution lies in automating the process. Thanks to the currently available hi-tech tools and training sets, businesses can frequently refresh their algorithms without the need for data scientists. Such capabilities should form part of the data management infrastructure that’s integrated with AI.

3. Trusting AI Decisions

Enterprises rely on machine insights to inform their workflows and decision-making processes with artificial intelligence. Over time, AI-powered decisions will occupy almost all day-to-day business operations. But for this to happen, the stakeholders must trust the machine outputs.

The trust issue stems from AI bias which may be attributed to bias from data sets or humans developing those algorithms. In a recent survey, merely 25-30 percent of businesses are making investments in creating control frameworks and processes that promote transparency and trust in artificial intelligence[7].

For this reason, proper governance and ethics should form part of any AI discussions. Remember, artificial intelligence depends on the data sets and regulations supporting it. And biased data and algorithms will automatically translate to AI bias.

The solution

Organizations need to create processes and formulate policies that manage the impact of artificial intelligence on decision-making by machines as well as humans. Moreover, AI teams tasked with designing, developing, and managing AI implementations should undergo routine training on ethics matters and accountability within AI systems. And lastly, AI systems and related insights should be subject to routine audits.

4. Calculating The ROI on AI Projects

According to the Forbes Insights survey, 42% of executives indicate that a poor return-on-investment (ROI) on artificial intelligence projects poses a great challenge in adopting the technology. Poor ROI also contributes to a lack of top management support for machine learning, resulting in budget cuts for AI projects. Therefore, this is true, especially in organizations where the implementation of AI requires a significant change in company processes[8].

Ken Nixon, the former head of sales and business development for Intro-Blue says[9]:

“Artificial intelligence requires fresh ways of measuring value gained, away from conventional benchmarks and basic efficiency. The pros are both soft and hard. AI accounts for our company’s entire value addition, as well as the competitive advantage over our market competitors. Our clients benefit from AI, and it informs our decisions on expanding the commodity for the future while using more data sets and well-designed algorithms for quality results.”

The problem is many executives have a limited expectation of artificial intelligence. Most of them look at AI systems to pick up manual, routine tasks.

However, looking beyond the automation of repetitive tasks by AI will see the technology become more transformative and contribute to better ROI.

The solution

With artificial intelligence implementations, there’s a need to redefine ROI. For example, AI saves time by automating manual, repetitive tasks. AI can also automatically highlight outlying aspects that the human eye cannot see within a large body of data. This means that employees can pursue other value-additive duties during that extra free time.

Therefore, businesses should develop a more extensive and open-minded perspective of AI. This will help unearth novel approaches where artificial intelligence can boost operations, customer experience, productivity, and ultimately ROI.

Challenges of artificial intelligence implementation: The summary

AI creates new possibilities even in areas that didn’t look promising in the past. This technology can revolutionize business operations provided organizations can tackle the problems of implementation.

However, it is important to note that AI projects experience as many failures as successes. That means you have to learn how to overcome artificial intelligence challenges to move forward. If you succeed in implementing artificial intelligence across the entire organization, you can enjoy a greater competitive edge in a world where machines and humans work jointly.

If you’re looking for AI consulting services, the Addepto team is at your disposal!

References

[1] Ibm.com. 451 Research Report. URL: https://www.ibm.com/analytics/resources/451-research-report/. Accessed February 10, 2022
[2]Deloitte.com. Deloitte State of Cognitive Survey. URL: https://www2.deloitte.com/content/dam/Deloitte/us/Documents/deloitte-analytics/us-da-2017-deloitte-state-of-cognitive-survey.pdf. Accessed February 10, 2022
[3] Forbes.com. Indicators of the state of artificial intelligence. URL: https://www.forbes.com/sites/gilpress/2019/04/30/7-indicators-of-the-state-of-artificial-intelligence-ai-april-2019/?sh=4826e9766009. Accessed February 14, 2022
[4] The AI Ladder: Demystifying AI Challenges, Rob Thomas, O’Reilly Media, August 2019.
[5] Reuters.com. Amazon scraps secret AI recruiting tool that showed bias against women. URL: https://reut.rs/3JykdY4.
[6] Artificial Intelligence Global Adoption Trends & Strategies, IDC, July 2019, Accessed February 14, 2022
[7] AI Transforming the Enterprise: Eight Key AI Adoption Trends, KPMG, January 2020., Accessed February 14, 2022
[8] Caesarius, L.M. and J. Hohenthal, Searching For Big Data: How Incumbents Explore A Possible Adoption Of Big Data Technologies. Scandinavian Journal of Management, 2018. 34(2): p. 129-140., Accessed February 14, 2022
[9] The Future Applications of AI for Large Enterprises, Forbes Insights, November 2, 2019., Accessed February 14, 2022