AI has been called the most important general-purpose technology of our era, comparably disruptive to 20th century innovations like the internal combustion engine. By 2030, analysts predict that AI will have a global economic impact of $15.8 trillion.
While 72% of executives expect large effects from AI within five years, deployment of scalable AI is still low (18% in 2017), and extensive deployment even lower (5%). There is no shortage of fuel for AI—more data was created in 2016 and 2017 alone than in the previous 5,000 years of human existence—but less than 0.2% of that data is being used for AI purposes.
CEOs across a range of firm sizes and verticals are reimagining processes with AI in mind—including improved data handling, better UX, intelligent supply chains and customer solutions. Generally, however, actual AI deployment is still largely centralized to five types of companies:
- AI foundational development (e.g., Amazon, DeepMind, Apple, Baidu, Microsoft, Facebook)
- AI architectural engineering (e.g., Nvidia, Qualcomm, Google)
- AI data collection, analytics, optimization, deployment, and monetization
- AI developed for specific applications (e.g., Tesla, Spotify, Snapchat, Johnson & Johnson)
- AI borrowed (via APIs/integrations) for specific applications (e.g., Chevrolet, GE)
Further AI adoption is held up by a series of obstacles: lack of analyzable data, lack of computing power, personnel shortages, and a paradigmatic misunderstanding of what AI is.
One sizable difference between those successfully and unsuccessfully incorporating AI is their approach to data. A company may have a lot of data generally, but to train AI algorithms, that data must be robust, reliable, and analyzable. And yet, in one study, researchers found that 97% of business leaders considered their data at least somewhat unreliable.
For example, published research data skews positive because null results often go unpublished. To build an unbiased database, however, data must be unbiased; you need null data. Likewise, researcher biases can emerge in data collection or sourcing. The ownership of data remains an issue, as many companies rely on others’ data, but don’t necessarily get sufficient access. Health data, for example, is strictly regulated for privacy concerns.
For supervised machine learning, it’s also critical that data be organized: tagged, structured, and defragmented across sources. This requires a lot of work from data scientists and investment in data oversight, a field that IBM predicts will grow by 28% between 2017 and 2020.
As with data scientists, there is a shortage of AI specialists in general. Chinese tech giant Tencent estimates that there are just 300,000 AI researchers and practitioners worldwide, but there is market demand for millions. LinkedIn ranks machine learning engineers, data scientists, and big data engineers among the five fastest-growing fields. There were 9.8 times as many machine learning engineers working in 2017 than 2012.
Enterprise AI also requires use-case engineers who can build AI, experts who can coordinate how humans and machines work together, and innovators who improve AI’s capabilities through groundbreaking research and development. To fully realize AI’s potential, businesses will need to re-skill current employees and compete to hire new ones with niche AI-related skill sets.
Businesses should also look at crowdsourcing for filling talent gaps. Increasingly, subcontracting platforms are aggregating work from contingent specialists, who can provide AI, machine learning, and design expertise in a distributed problem-solving model that offers collective intelligence and improved quality control. AI specialist crowdsourcing is growing fast and demonstrates the potential for capturing a much larger and significant share of the market.
Beyond AI specialists, there is a shortage of AI leadership. At Davos 2016, eBay CEO David Wenig charged executives with one goal above all: Learn basic computer programming. Leaders should do so not because they are all going to be engineers, “but unless you take basic computer programming, the concept of the digital economy is very abstract,” Wenig said.
If CEOs are serious about AI, they should be promoting AI literacy across their companies—from back office to boardrooms. This not only helps close talent gaps, but improves technical oversight and vision. By becoming knowledgeable AI leaders, executives and managers will be able to think more strategically about what AI can do, how to allocate resources, how to benchmark potential hires, how to foster human-AI collaboration, and where innovative applications can be embraced.
The sheer amount of computing power required for AI processes has become another bottleneck to adoption. In recent years, cloud computing and parallel processing provided short term solutions, but as data volume grows and as deep learning drives automated creation of increasingly complex algorithms, we’re due for another revolution in AI infrastructure. Today, companies are building innovative hardware—neuromorphic chips and tensor processing units (TPUs), for example—in order to improve computing power for scalable AI.
The final hindrance to AI adoption is a concern regarding ethics and security. The most common misconceptions about AI tend to fall on two extreme ends of the spectrum: On one hand, leaders expect AI to be a panacea, capable of 100% human-like performance on day one of implementation. On the other hand, there are concerns that AI will replace human ingenuity entirely.
Both of these assumptions are rooted in yet another fallacy—that machines are equal to humans when it comes to decision-making and cognitive capabilities. This is not the case. To date, nobody has developed an AI that is generally intelligent (AGI)—i.e., capable of fully replicating human cognition.
The vast majority of existing AI technologies employ Artificial Narrow Intelligence (ANI), meaning they are very good (better than humans, for the most part) at a specific skill set or task—but not for every task. An AI that has been developed for facial image recognition is not capable of also optimizing email subject lines. To properly employ ANI, business leaders should view the technology as a tool for driving specific outcomes, not for human replacement.
A business landscape fueled by AI is inevitable. The gap between AI expectations and achievements can only be closed with a robust talent pool, a refinement of enterprise R&D, and advancements in infrastructure and hardware.
The majority of companies integrating AI will likely offload foundational, architectural, and data development to an ecosystem of partners and experts who will guide their AI journey, from baby steps to large-scale enterprise-wide adoption. In combination, such efforts will drive an even more substantial gap—a gap between the leaders and laggards of Industry 4.0.
This article was produced on behalf of Wipro by Quartz Creative and not by the Quartz editorial staff.