With considerations that include user experience, business impact, technical design, and risk management, it’s easy to get lost in the many priorities of building AI. And without adopting the right mindset and approach to responsible AI design, your organization risks a number of unintended consequences.
I spent the last 12 months attending over 14 AI and technology conferences, featuring wickedly smart AI leaders who are working to solve a range of challenges. Many draw on their own experiences, while also looking at organizations that are successfully building AI, for solutions.
And as you piece together the stories these experts have shared, patterns begin to emerge. In this article, I touch on the six most common characteristics of companies that are successfully building AI, and what we can learn from them.
1. Obsess Over Quantifying Impact
The impact of an AI/ML model can be measured in money saved, revenue added, risk avoided, time saved, and other metrics. Organizations that are successfully building AI solutions know the value of the problems they’re trying to solve and prioritize their data science resources accordingly.
It’s easy to get stuck in the never-ending loop of making AI models more accurate, but this will only lengthen the time it takes to achieve positive ROI from your investment. This is why time to value, or the amount of time it takes to realize the worth of an AI solution, is of the highest priority to top AI builders. They rapidly prototype ML models to arrive at the right strategy, ultimately enhancing the return on their AI investment.
In some of my most profound conversations, AI leaders hire elite teams of internal consultants to bridge the gap between functional business areas and the engineers building their models. This way, the expected impact of the solution is clearly communicated and understood among all involved stakeholders.
2. Approach AI With a Portfolio Mindset
Originally a financial management concept, managing with a portfolio mindset means allocating resources to different investments or projects, creating a diverse portfolio of assets with varying levels of risk.
Successful AI builders are approaching AI with a portfolio mindset.
For example, Apple tries to balance many simple predictive analytics solutions (spreadsheets and regression) with a handful of moonshot ideas. This approach yields a steady stream of AI wins for the team, creating continuous excitement and faith in their data science teams’ abilities.
3. Prioritize the Human-AI Interaction
Organizations that are successfully building AI understand how AI impacts teams’ workflows and decision-making. They focus on what it takes for team members to trust and embrace the addition of AI to their day-to-day processes.
In some cases, as Jennifer Strong has mentioned, this means making AI “boring”, or a normal part of workflows. In others, this means still leaving the decision-making power in team members’ hands, even if the model is more accurate.
4. Learn To Do More With Less Data
While AI/ML models once required mass amounts of data, top AI builders are shifting to what Andrew Ng calls data-centric AI, or “the discipline of systematically engineering the data needed to build a successful AI system.”
The rapid evolution of transfer learning and pre-existing models have amplified our ability to solve problems. Organizations that previously lacked sufficient data can now build models using the right data, supplemented by third-party sources as needed.
5. Recognize the Importance of AI Governance
With the European Union’s AI Act, New York City’s automated employment regulation, and Canada’s Artificial Intelligence & Data Act proposal, the AI regulatory landscape is about to change from suggested frameworks to more permanent laws.
Organizations that are successfully building AI are not only prioritizing AI governance, but are also considering how to make compliance effortless for their teams. If regulatory checklists are too long or complicated, there is a much higher risk of making mistakes. Instead, companies are implementing solutions, like putting parameters in place to flag data inconsistency, to improve the governance process.
6. Know What To Buy Vs. Build
There is no longer a need to build every system that will contribute to your AI/ML solution.
This shift is similar to what we’d see in software engineering. If you want to write a computer program, you purchase an operating system like Windows, find a laptop that meets your specs, and use pre-existing code collaboration software like GitHub. From there, you write the code.
Similarly in data science, data labeling tools and workflows are increasingly low-code and can be purchased directly from big cloud vendors, like AWS. And pre-existing models can be found in a number of AI marketplaces. Companies that understand this are saving significant resources otherwise spent on building their own data science collaboration tools.
If you’ve ever been nervous about building AI, now is the time to move past that fear. It has never been easier for companies to get started with AI—especially when the right people and processes are put in place for success.
About the Author: Cal Al-Dhubaib is a globally recognized data scientist and AI strategist in trustworthy artificial intelligence, as well as the founder and CEO of Pandata, a Cleveland-based AI consultancy, design, and development firm.
Featured Image Credit: Canva