AI is poised to deliver measurable value for a variety of public and private sector applications, and is just beginning to make inroads in the enterprise. This article offers best practices for picking an AI use case, points out barriers to AI success, and advises on how to find the best AI talent.
Artificial intelligence (AI) technology dominates the headlines, but it’s still not widely used. According to GartnerOpens a new window , between 2018 and 2019, the number of organizations deploying AI grew to just 14 percent. This may lead some enterprises to wonder, is the transition to AI necessary?
If approached correctly, absolutely.
AI technologies are well-positioned to provide measurable value in the form of speed, accuracy, and/or scale for a wide-variety of public and private sector applications. Organizations are increasingly experimenting with AI technology to better understand its return. For example, in the geospatial analytics and mapping market, computer vision (a specific type of AI) could greatly enhance both the speed and scale in which companies operate.
Knowing where to deploy AI
As with any digital transformation initiative, knowing where to start is half to battle. When evaluating projects for AI infusion, begin by identifying a narrow, well-defined set of problems that your organization hopes to solve. Be sure there’s ample data to support the project, as well as a way to clearly define metrics for success.
While there may be interest or pressure to deploy AI against a company’s most challenging business problem, the ideal project candidates for an initial AI transition are those that consist of fairly simple, repeatable tasks that are typically time and cost-consuming. Those projects traditionally offer a tightly-bound problem or series of set challenges to train an AI model(s). They are also likely to motivate and recruit early champions, as staff often support projects that can preserve human resources.
Set Realistic Expectations
We have learned two hard-earned lessons when it comes to an AI transition. First, it’s difficult to avoid the hype cycle around AI, so it’s critical for companies to set expectations at every stage. AI projects are most effective and cost-efficient when developed incrementally, with lots of input from both the product management team and data scientists. Don’t swing for the fences with a bold, flashy application of AI: as thought leaders in the space like Hilary Mason point out, machine learning is most effective when it’s boring. Start small, make gradual, ongoing improvements, and keep the cost-benefit ratio of implementing AI in mind at all times.
Second, it is essential for companies to take a hard look at the data they have available and figure out how to curate for AI workflows. All too often, companies rush to develop a massive data set without appropriately assessing what they need. This can lead to cost-overruns, schedule delays, and frustration among early advocates. Deep analyses of the necessary data early on will pay massive dividends as the AI project matures.
Be Mindful of AI Barriers
As you embark on an AI application, be mindful of the four major barriers that block deploying AI/machine learning products:
- Inability to explain a models’ decisions,
- Lack of quality training data, and
- Availability of appropriate, fair model training datasets
The high cost of model development or deployment compared to existing manual analysis makes some tasks bad automation targets. Specifically, the additional cost of acquiring an IT environment sufficient to train and test models might end up being cost prohibitive to some organizations. Second, environments which require a degree of explainability beyond what AI models can provide â€“ for example, some medical or financial systems â€“ may remain intractable until sufficient understanding of models is achieved. Third, while many organizations are collecting more data than ever before, those data sets may not be sufficiently curated for AI models. Organizations may not know how to curate those data or have the necessary discretionary funds to quickly build a curated data set. Fourth and finally, commercial datasets for some tasks include bias that cannot be separated from decision-making variables given their structure, which can introduce bias to automations.
Find and Engage the Best AI Talent
The success of any technology endeavor is driven largely by the team behind it. In this day in age, organizations must be creative in how they find and leverage technology talent, which means they have to go outside their internal team to find technologists with coveted skills in emerging technologies like AI. Earlier this year, Topcoder joined as a SpaceNet partner alongside IQT CosmiQ Works, Maxar Technologies, Intel AI, Amazon Web Services, and Capella Space to help crowdsource the work of developing highly complex algorithms that analyze our collected satellite imagery to its open workforce community of gig economy data scientists and developers. Crowdsourcing helps SpaceNet accelerate innovation by simultaneously exploring a number of algorithmic approaches, enabling our geospatial experts to rapidly identify the best method and spend more time generating insights from the data.
A critical piece of outsourcing projects to a talent network with an open workforce model is to ensure they have the proven knowledge and skills necessary to do the job. The vetting process a company like Topcoder takes on is something SpaceNet wouldn’t have the global reach, or project scope, to do.
This type of advanced developer participates in crowdsourcing competitions and contract work as a way to hone skills, earn a living and gain access to projects that they wouldn’t have otherwise. SpaceNet wants greater visibility into those communities and competitions to give top AI talent across the globe a chance to better get to know what we and their partners are trying to accomplish. One example of how we’re doing that is by participating in the world’s largest programming and design tournament, the Topcoder Open (TCO) November 13-16 in Houston. SpaceNet will lead a master class on geospatial analytics at the TCO to empower the Topcoder Community with even more AI and data science skills specific to SpaceNet’s work, which will help further develop mutually beneficial technology relationships and project opportunities.
It’s a Marathon, Not a Sprint
AI is still an emerging technology. As you incorporate it into an enterprise technology strategy, be flexible and patient. Most products go through several iterations of question and method tuning before they achieve deployability, so plan accordingly.
Finally, make sure to share AI lessons learned, code, and data sets across teams even if they are focused on different applications or lines of business. This will foster the development of an AI community within your IT and, hopefully, proliferate throughout the entire company.