Artificial intelligence (AI) has the propensity to push the art of predictive analytics to new levels. Its ability to parse massive amounts of data in short periods allows it to detect patterns and relationships between disparate data sets while at the same time keeping an eye on potential disruptions to those patterns that could result in unexpected outcomes.
This makes it invaluable for applications ranging from sales and marketing to business planning, product development, and cybersecurity. But how accurate is AI proving to be? And how effective is it at forecasting the behavior of complex systems and environments, especially when those environments will become increasingly influenced by AI itself?
AI Predicting the Hits
Recent research from the Center for Neuroeconomics Studies at California’s Claremont Graduate University suggests that applying AI to the traditional linear statistical models used to predict whether a given song will become a hit or not increases the accuracy rate from 69 percent to an impressive 97 percent. If true, this offers tremendous potential for the music industry to focus its production, promotion, and other resources on the most promising titles with virtual certainty that they will make a profit.
Streaming services, in particular, could see much greater customer retention and revenue generation if they populate their new music offerings with tunes that stand a high chance of being accepted.
At the moment, however, only about 4% of new songs become hits. By using AI to more closely align listeners’ interests with new releases, that percentage should rise. However, this could lead to blandness in the music industry because most recommendations are based on what users have already selected, not what is new and fresh.
A Healthier Outlook
In healthcare, providers are improving the accuracy of diagnoses, remedies, and even drug discovery and treatment development by putting large language models (LLMs) like ChatGPT to work interpreting the notes of physicians, clinicians, researchers, and others. A new model dubbed NYUTron, developed at New York University’s Grossman School of Medicine, has shown it can ingest and analyze this structured and unstructured data despite its wide variety of styles and formats.
The model has shown a 15 percent improvement over standard predictive tools for critical areas like patient readmissions, in-hospital mortality, and insurance denial. This is expected to improve health outcomes by reducing the risk of infection, drug interactions, and other potential hazards. It can also streamline the billing process and even lower the overall cost of treatment.
Not all attempts at AI-driven predictive analytics have met with success, however. OpenAI, the creator of ChatGPT, recently pulled the plug on a model designed to help detect AI-generated text after achieving a success rate of only 26 percent. Meanwhile, the false positive rate, in which human text was labeled AI text, hovered around 9 percent.
Currently, the model is still publicly available in the hopes that user feedback will help develop a more reliable system. OpenAI is also researching new data provenance methods and is looking to develop similar AI classifiers for audio and video.
If successful, these kinds of programs would greatly help educators, fraud investigators, and organizations fighting misinformation and disinformation on both social and traditional media. Even if the success rate is improved, OpenAI does not recommend that its models be used as primary decision-making tools – that responsibility should remain with humans.
Picking the Winners
Accurately predicting the future is also the key to making a killing in the stock market, so it’s no wonder that interest is keen on creating an AI model that can do just that. One developer claiming success in this area is VantagePoint Software. The company says its new AI-driven platform “accurately predicts future prices and trend changes in various markets including U.S. stocks, Canadian stocks, ETFs, Futures, Forex, and Crypto.”
The firm, which has been developing trading software since the 1980s, says this latest version provides 87.4 percent “proven accuracy” by leveraging deep learning and neural networks to enhance prediction, optimization, pattern recognition, classification, and other factors used to spot potential opportunities, and pitfalls, in financial markets.
If AI prediction is improving, then there is no reason to think it cannot benefit investing and trading to the same extent as marketing and product development.
But might this lead to a dilemma in the future? If everybody is making the right call nearly all the time, essentially lessening risk and enhancing reward, is there a danger of magnifying the consequences of miscalculations when they do occur?
If everyone starts using AI to buy the right stocks or launch a new business, what happens when those models suddenly start flashing warning signs? Will the correction or the market contraction be more severe and of longer duration than if only some players were getting it right?
It’s been nearly a hundred years since the world suffered a catastrophic economic collapse, which was partly fueled by the widespread assumption that markets and valuations would rise forever. It would be a shame to repeat that history simply because an algorithm says it’s for real this time.