Artificial Intelligence (AI) project fails: Stop blaming the talent gap

When Artificial Intelligence initiatives fall short, the blame is often placed on a skills gap. But there's more to it. Does your organization prioritize these three foundational AI pillars?
Up
129 readers like this

Hiring the right technical talent remains a significant roadblock to Artificial Intelligence (AI) adoption for enterprise organizations. According to a recent O’Reilly survey, slightly more than one-sixth of respondents cited difficulty in hiring and retaining professionals with AI skills as a significant barrier to AI adoption in their organizations.

While the talent gap remains a large part of the dialog, this number has decreased from the previous year, signaling that other challenges are becoming top of mind for businesses exploring and deploying AI projects.

Still, the technical skills gap isn’t the biggest impediment to AI adoption, nor is it the reason so many AI projects fail. In fact, according to the same O’Reilly survey, respondents identified a lack of institutional support as the biggest problem, followed by difficulties in identifying appropriate business use cases.

[ Check out our primer on 10 key artificial intelligence terms for IT and business leaders: Cheat sheet: AI glossary. ]

Of course, this is a harder pill to swallow: It means the real challenge lies with us rather than with a limited number of professionals equipped to do the job.

3 pillars of AI project success

So how can organizations avoid the common pitfalls of AI projects? As with other technology implementations, it all comes down to proper company-wide training, the production environment, and having the right foundation in place. With these three pillars in place, you can start realizing the business value of AI earlier.

1. The right foundation

Successful AI projects require three things:

  • Data scientists must be productively tooled, have domain expertise, and access to relevant data. While AI technology is becoming well-understood—from handling bias prevention, explainability, concept drift, and similar requirements—many teams still fall short here.
  • Organizations must learn how to deploy and operate AI models in production. This requires DevOps, SecOps, and newly emerging AIOps tools and processes to be put in place so models continue working accurately in production over time.
  • Product managers and business leaders must be involved from the beginning, in order to redesign new technical capabilities and decide how they will be applied to make customers happy.

While education and tooling have improved significantly over the last several years, there’s still much room for improvement in actually operating AI models in production. In that vein, product management and user interaction design are becoming common hurdles in AI success.

These problems can be addressed by investing in hands-on education. Outside the classroom and conference halls, professionals from all across your organization must get experience actually working on AI projects, understanding what they can do and how the technology can push your business forward.

2. Company-wide collaboration and training

Certainly, talent is part of the problem, but it’s not just data science talent that’s needed. The root of the problem usually lies within business and product expertise. As important as technical talent is, understanding how AI will work within a product and how it translates to better customer experience and new revenue is just as critical – and that responsibility doesn’t fall solely on the R&D team.

For example, we have algorithms that can read X-rays as accurately as humans, but we’re just now beginning to integrate this capability into the clinical workflow. If doctors and nurses aren’t trained on how to use this technology to streamline their workflow, it holds no value for them or their patients.

Being able to train and deploy accurate AI models doesn’t address the question of how to most effectively use them to help your customers. Doing this requires educating all organizational disciplines – sales, marketing, product, design, legal, customer success, finance – on why the technology is useful and how it will impact their job function.

Done well, new AI-enabled capabilities empower product teams to completely rethink the user experience.

Done well, new AI-enabled capabilities empower product teams to completely rethink the user experience. It’s the difference between Netflix or Spotify adding recommendations as a side feature versus designing their user interface around content discovery. It makes a big difference, but it also takes a village to achieve. That’s why company-wide buy-in spearheaded by the executive team is vital to AI success.

3. Proper production environment

Not all production environments are the same, so not all outcomes will be the same. It’s important to understand the limitations of AI projects based on the talent, infrastructure, and data you have and to set clear expectations from the get-go.

For example, a recent research paper (done for the ACM Conference on Human Factors in Computing Systems (CHI) series of academic conferences) explored a new deep-learning model used to detect diabetic retinopathy from images of patients’ eyes. Scientists trained a deep-learning model to identify early stages of diabetic retinopathy in patients from pictures of corneas from eye exams over the past several years. The goal was to reduce blindness, a symptom of the disease when left untreated.

The paper describes what happened when the same accurate, effective model was used in clinics in rural Thailand: The machines used to take images of patients’ eyes were not as sophisticated as the ones used for training the model. The exam rooms used were not completely dark, as the trained model assumed. For some patients, taking another day off for follow-ups or additional testing wasn’t a viable option. To boot, not all doctors and nurses were trained to explain why this new test was necessary.

The lack of proper infrastructure and cohesive education for hospital staff, coupled with an understanding of practical limitations, is a prime example of why AI projects fail.

The AI talent gap will remain a challenge for the next few years, as education catches up to industry. But in the meantime, there are steps organizations can take to ensure their AI projects prevail.

It’s not enough to just train your models – train your organization, too. Take the time to educate every facet of your business on why you’re tackling a certain AI project, how it will impact their role and the customer experience, and what the expectations are.

The right talent will come – will your organization be ready to use it?

[ Get the eBook: Top considerations for building a production-ready AI/ML environment. ]

David Talby, PhD, MBA, is the CTO of John Snow Labs. He has spent his career making AI, big data, and Data Science solve real-world problems in healthcare, life science, and related fields.

Social Media Share Icons