How best to use AI for human development

The excitement surrounding artificial intelligence (AI) today reflects not only how AI applications could transform businesses and economies, but also the hope that they can address challenges such as cancer and climate change. The idea that AI could revolutionize people's well-being is obviously appealing.

But just how realistic is it?

To answer that question, we (at McKinsey Global Institute) examined more than 150 scenarios in which AI is being applied or could be applied for social good. What we found is that AI could make a powerful contribution to resolving many types of societal challenges, but it is not a silver bullet. While AI's reach is broad, development bottlenecks and application risks must be overcome before the benefits can be realized on a global scale.

AI is already changing how we tackle human development challenges. In 2017, for example, object-detection software and satellite imagery aided rescuers in Houston as they navigated the aftermath of Hurricane Harvey. In Africa, algorithms have helped reduce poaching in wildlife parks. In Denmark, voice-recognition programs are used in emergency calls to detect whether callers are experiencing cardiac arrest. And at the Massachusetts Institute of Technology Media Lab near Boston, researchers have used "reinforcement learning" in simulated clinical trials involving patients with glioblastoma, the most aggressive form of brain cancer, to reduce chemotherapy doses.

Moreover, this is only a fraction of what is possible. AI can already detect early signs of diabetes from heart rate sensor data, help children with autism manage their emotions, and guide the visually impaired. If these innovations were widely available and used, the health and social benefits would be immense. In fact, our assessment concludes that AI technologies could accelerate progress on each of the 17 UN Sustainable Development Goals.

But if any of these AI solutions are to make a difference globally, their use must be scaled up dramatically. To do that, we must first address developmental obstacles and mitigate risks that could render AI technologies more harmful than helpful.

On the development side, data accessibility is among the most significant hurdles. In many cases, sensitive or commercially viable data that have societal applications are privately owned and not accessible to nongovernmental organizations. In other cases, bureaucratic inertia keeps otherwise useful data locked up.

So-called last-mile implementation challenges are another common problem. Even in cases where data are available and the technology is mature, the dearth of data scientists can make it difficult to apply AI solutions locally. One way to address the shortage of workers with the skills needed to strengthen and implement AI capabilities is for companies that employ such workers to devote more time and resources to beneficial causes. They should encourage AI experts to take on pro bono projects and reward them for doing so.

There are of course risks. AI's tools and techniques can be misused, intentionally or inadvertently. For example, biases can be embedded in AI algorithms or datasets, which in turn can amplify existing inequalities when the applications are used. According to one academic study, error rates for facial analysis software are less than 1 percent for light-skinned men, but as high as 35 percent for dark-skinned women, which raises important questions about how to account for human prejudice in AI programming. Another obvious risk is misuse of AI by those intent on threatening individuals' physical, digital, financial and emotional security.

Stakeholders from the private and public sectors must work together to address these issues. To increase the availability of data, for example, public officials and private actors should grant broader access to those seeking to use data for initiatives that serve the public good. Already, satellite companies have participated in an international agreement that commits them to provide open access during emergencies. Such data-dependent partnerships must be expanded so that it becomes a feature of companies' operational routines.

AI is fast becoming an invaluable part of the human development toolkit. But if AI's potential to do good globally is to be fully realized, proponents must focus less on the hype and make more efforts to removes the hurdles preventing its uptake.

Michael Chui is a partner at the McKinsey Global Institute, and Martin Harrysson is a partner in McKinsey & Company's Silicon Valley office. Project Syndicate