By Brons Larson, AI strategy lead at Dell Technologies
Throughout this month, organizations around the world will celebrate AI’s ability to benefit society. But what are they looking to accomplish and how is the technology advancing those goals?
The AI for Good Foundation maps AI’s positive impacts against the UN’s 17 Sustainable Development Goals, which range from ending poverty and hunger to promoting quality education, clean energy and sustainable cities. These broad-ranging goals can be boiled down into two main areas.
AI is protecting people and the planet
The first of these is benefiting people. AI’s ability to find patterns in oceans of data can help to make life safer in unexpected ways. For example, its ability to spot counterfeit goods has implications beyond identifying fake Gucci handbags. It could also detect potentially dangerous counterfeit pharmaceuticals or stop sub-quality knock-off components from making their way through the supply chain into vehicle airbag systems. It could save lives.
In medicine, AI is finding patterns in patient DNA, enabling scientists to make better decisions when treating patients for tumors. Machine learning algorithms powered by high-performance computing hardware are helping doctors to understand and treat the long-term effects of COVID-19 using complex simulations of patients’ bodies, called ‘digital twins.’
AI is also helping to advance a second broad social good: environmental protection. Nonprofits are using it to prevent deforestation, some by using ground-based sensors that check for telltale chainsaw sounds, and others by analyzing satellite images.
Then, there are solutions that mix both. The state of California is using AI to predict the spread of wildfires, helping it to deploy resources that save the forest and people living nearby.
AI is demonstrably a force for social good, but like any technology, it also carries the potential for misuse. It takes effort and planning to gear its outcomes for positive social impacts.
Conventional machine learning AI, also known as second-wave AI, depends heavily on training data. Poor-quality data introduces bias, which can cause problems in areas such as civil rights. That’s why it’s important to consider ethics when using this technology.
That means understanding who you’re doing business with when providing systems and software that use AI. As a technology that amplifies existing capabilities and often creates entirely new ones, it’s important to deal with customers and partners committed to using it responsibly.
It also means understanding the technology’s limitations. As a statistical solution, traditional machine learning will encounter edge cases where it makes mistakes. That could mean harmlessly misidentifying a cat picture online. It could also cause an autonomous vehicle to crash. Biased machine learning based on poor data sets could discriminate against specific groups of people, denying someone parole, a job or a loan.
We should also consider the unexpected consequences of using AI, but without letting our fears get in the way. We should avoid unnecessary over-regulation that could prevent the positive social uses of this powerful technology.
Some of AI’s dangers will naturally disappear as the technology develops. We’re already seeing an evolution from second-wave to third-wave AI, which is both more capable and more resilient against some of those ethical dangers.
Instead of relying on data that represents objects, third-wave AI focuses on the features and properties of those objects. That makes it more accurate, but it also gives third-wave AI a capability that second-wave technology never had: reasoning.
Third-wave systems can use what they know about objects to infer new information. For example, a second-wave system might recognize a ball bouncing across the road and slow down an autonomous car to avoid hitting it. A third-wave system might know that where there’s a bouncing ball, there’s probably someone chasing it not far behind. That might cause it to stop the car altogether and wait, just in case.
Third-wave AI is already achieving socially positive outcomes while avoiding second-wave AI’s downsides. It’s less subject to data-based bias because it doesn’t use extensive data sets for training. Unlike second-wave systems, it can also explain its decisions, helping to keep results equitable.
From a social good perspective, third-wave AI’s biggest win is that, unlike machine learning, its reasoning capabilities enable it to identify things that it hasn’t seen before. A good example of this is the TIGER project, a biosensor system that SAIC developed in the early 2000s. It uses third-wave AI to identify unexpected infectious organisms. The commercial version of this technology was the first to identify SARS as a novel Coronavirus and now helps doctors to diagnose medical conditions.
As AI goes through these evolutionary step changes, it will get smarter still. Third-wave AI models will work together to tackle wider-ranging social and environmental problems at speed and at scale. Hollywood may enjoy unlikely dystopian sci-fi scenarios where AI takes over the world, but the reality is far more promising—and thanks to the AI for social good movement, it’s happening right now.