AI has become an indispensable resource for many businesses in 2020 amid the challenges of COVID-19. Despite economic stress, 61% of high-performing companies invested more in AI in 2020. As recovery from pandemic losses gets underway, more may turn to AI to strengthen their digital transformation.
Increased investments in AI could help the economy recover faster, but they also come with some risks. As such, 2021 could be a turning point for the technology as businesses push to resolve both long-standing and newly found AI challenges. Here are seven of these issues that face AI adoption and how companies can overcome them.
Table of Contents
1. Poor Data Quality
Perhaps the most substantial barrier to profitable AI adoption is the prevalence of low-quality data. Despite lofty promises about the technology, 65% of global executives say they have not seen value from their recent AI investments. Poor data quality is a significant part of this problem, as AI applications are only as good as the information they can access.
Many companies collect far too much data, leaving them with redundancies and inconsistencies. Finding and applying high-quality information is a matter of streamlining collection processes and paying more attention to cleansing, labeling, and warehousing. These workflow changes, along with better cataloging software, will provide companies with higher-quality data in 2021.
2. AI Ethics
As companies bring AI to more processes, ethics becomes a pressing concern. AI tends to amplify and bring seemingly scientific credence to human biases, casting a dark shadow over its potential for decision-making applications. Thankfully, this is not a problem without an answer.
The growing awareness over this issue is a promising sign, as the first step is acknowledging AI’s potential for bias. As companies train their AI, they must actively work to counter prejudiced data, specifically programming AI to be anti-bias. Teams must also carefully analyze training data before feeding it into the algorithm, ensuring it can’t lead to unethical conclusions.
3. Data Storage Limitations
As AI becomes more prominent, companies have to collect and store more data. That’s becoming an issue, as traditional storage technologies are limited and often expensive. Recent technological breakthroughs have provided a solution.
For example, QLC flash is 25% denser than TLC and offers lower costs per gigabyte. Other innovations, like NVMe, have become increasingly popular, too, making flash storage more affordable and reliable than ever. Businesses can now turn to flash storage for AI applications instead of the less scalable, more expensive tradition of using hard disks.
4. Edge AI Security
Edge computing is ideal for AI applications, given its lower latency and better load balancing. Many organizations may turn to edge AI in 2021, but these deployments come with their own challenges. Most notably, edge infrastructure is vulnerable to accidental damage, especially outside of a workplace.
Proper physical security for edge infrastructure comes in three stages: monitoring, control, and supervision. IoT sensors can monitor the area around edge devices, detecting and reacting to physical hazards. Companies can control and supervise their edge installments with similar technology, capitalizing on the IoT to restrict access to this infrastructure.
5. Data Governance Concerns
People have become increasingly concerned about how companies access and use their personal information. Businesses using customer-facing AI need to account for this in their future deployments. Responsible data governance is more crucial than ever, especially in the face of rising cybercrime.
The key here is visibility and segmentation. Companies need to ensure they can see how their AI algorithms use data at all stages and restrict it. Segmentation will mitigate the impact of a breach, keeping user information as safe as possible. Being transparent about data collection policies will also help assuage customer concerns over AI.
6. CPU Bottlenecks
As computing demands and applications have risen, developers are facing CPU bottlenecks. It’s becoming evident that Moore’s Law might not hold up past a certain point, as transistors can only get so small. These limits are a roadblock in AI advancement, but companies may overcome them by bypassing CPUs.
Since GPUs can perform parallel operations on multiple datasets, they’re ideal for machine learning tasks. While these processors can’t replace CPUs entirely, they can handle the majority of analytics work. AI processes in 2021 will lean more heavily on GPUs as a result.
7. Regulatory Compliance
As AI and other data-centric operations become more prominent, they face increasing legal regulations. In 2020, at least 38 states introduced or considered new cybersecurity legislation. Consequently, AI developers and users will have to keep legal restrictions in mind in 2021.
Data regulations are a new and developing matter for U.S. companies, so rules are likely to change over the next few years. As a result, AI companies must remain flexible and adopt high privacy and governance standards even before they become law. Third-party auditors will also become more in-demand in the face of increasing regulations.
2021 Could Be a Landmark Year for AI
The COVID-19 pandemic may have slowed AI’s growth, but it will likely have the opposite effect as it fades. AI will drive economic recovery, and this surge in adoption will spur developers and users alike to push through challenges. As a result, 2021 could end up as a substantial turning point for AI development and implementation.