Predictive AI: What It Should and Shouldn’t Predict

One of the age-old dilemmas around technology has to do with managing its potential to create as many problems as it solves. You might remember the classic line delivered by Jeff Goldblum’s leather-clad, snarky scientist Ian Malcolm in “Jurassic Park”: “Your scientists were so preoccupied with whether or not they could,” he quipped, “that they didn’t stop to think if they should.” At TackleAI, we’re bullish on all the societal and business solutions that could be pioneered by artificial intelligence, but we’re also aware that powerful technology – whether it’s the automobile or nuclear fission – can come at great cost. Rather than waiting to see the results of these developments, we should be thinking proactively about which fields should be employing predictive AI and which should proceed with extreme caution or avoid it altogether.

For instance, here are a few things AI should predict:

Supply chains. If you’ve recently tried purchasing any of a variety of goods, from seafood to solar panels, you know that finding the stuff you need isn’t as easy as it once was. America has been pretty fortunate over the years, but we’re now seeing how a handful of significant events (the pandemic, Russian sanctions) can put a squeeze on the global supply chain. AI has the capability to better track inventory, account for lulls in production and build efficiencies into shipping, delivery and other critical operations that help keep store shelves and stock rooms full.

Climate change. On a somewhat smaller scale, AI can help in the battle against climate change through the design of smarter, energy-efficient homes and buildings, while improving power storage and optimizing energy deployment. With its predictive powers and real-time response rates, AI can improve the effectiveness of disaster alerts and save lives. Over time, deep learning may help reveal larger solutions that can be enacted to make a real dent in global emissions and other contributors to climate change.

Workplace risks. In the private sector, artificial intelligence is already making a positive difference for many companies. The tech can detect and predict injury risks on a factory floor, identify potential maintenance needs and pinpoint cybersecurity threats that can hurt the reputation of an organization and cost it money.

Potential illnesses. In a study at Penn State University, machine learning is being used to scan electronic health data and identity information to predict patients’ risk levels for disease. “Say we want to predict whether a patient will suffer from diabetes in the future,” said principal investigator Fenglong Ma. “We will use the patient’s historical data, which in some ways may be related to diabetes, such as high blood pressure or heart failure, and those are risk factors for the target disease.” AI can assist in identifying illness while also raising flags before patients actually get sick, keeping us all healthier.

And a few more – particularly those that rely heavily on historic data – that shouldn’t use artificial intelligence:

Hiring and firing trends. Some predictive AI-driven efforts are too sensitive to be relied on without a deep understanding (and correction for) their partialities. “In data science, bias is defined as an error that arises from faulty assumptions in the learning algorithm,” writes technology and IT enterprise website CIO. “Train your algorithms with data that doesn’t reflect the current landscape, and you will derive erroneous results. As such, with hiring, especially in an industry like IT, that has had historical issues with diversity, training an algorithm on historical hiring data can be a big mistake.”

Personal finance. Decisions about loans, interest rates and other banking or personal finance matters shouldn’t be made solely on societal trends or even an individual’s credit score. Banks and credit unions have a job to do, of course, and they must weigh the risk against the reward of loaning money. But feeding an algorithm with historical loan data that has been skewed by well-documented demographic and socioeconomic biases will yield the sort of corrupted results that perpetuate unfair lending practices.

Elections. So what if all the predictive models get it wrong when they call for a candidate’s win or loss, whether in a photo finish or by a landslide? No big deal, right? Wrong. Politicians have begun gaming the system, pushing the buttons of the electorate in a way that could foster voter disenfranchisement and affect an election in the moment. Tracking and predicting elections may be great fun, but the stakes are too high for the results to be swayed by some cable news talking head’s big board.