By Nicholas Larsen, International Banker
Largely through the advancements that have been made in machine learning (ML) and deep learning (DL) training capabilities in recent years, artificial intelligence (AI) has progressed by leaps and bounds in terms of what it can achieve, how it can be applied and the extent to which the world can benefit from it. Through the ingestion and interpretation of data, AI algorithms can more accurately generate insights and predictions that continue to underpin smarter decision-making. With adaptive AI systems proving adept at continuously responding, learning and modifying their outputs from ingesting new data, this technology’s capabilities look set to be dramatically upgraded.
While traditional ML consumes raw data to be applied in the real world, the process is largely a static one in which the learning process is completed and remains unchanged while the system is in use. As long as the world’s conditions remain the same, the model should be able to deliver successful results. But, of course, our world is ever-changing, which means that the model’s accuracy will diminish the less it reflects this changing environment.
Adaptive AI dynamically incorporates new data from its operating environment to generate more accurate insights on a real-time basis. It is increasingly regarded as artificial intelligence’s next evolutionary stage. By incorporating a more responsive learning methodology, such as agent-based modelling (ABM) and reinforcement learning (RL) techniques, adaptive AI systems are more reactive to the changing world around them and can thus more seamlessly adapt to new environments and circumstances that were not present during the earlier stages of the AI system’s development. This can be achieved by working on new data in runtime and development environments that enable models to adapt and update their own codes, thus allowing the AI system to dynamically retrain, learn and improve following those changes to the environment.
Adaptive AI systems support “a decision-making framework centered around making faster decisions while remaining flexible to adjust as issues arise,” Gartner noted, whilst recognising that such systems aim to learn continuously based on new data in runtime environments in order to adapt more quickly to changes in real-world circumstances. This data may represent changes in behaviours by consumers or businesses in real-time, which means that adaptive AI can maintain its accuracy continuously. “Flexibility and adaptability are now vital, as many businesses have learned during recent health and climate crises,” Gartner Research’s Distinguished VP Analyst Erick Brethenoux observed in October 2022. “Adaptive AI systems aim to continuously retrain models or apply other mechanisms to adapt and learn within runtime and development environments—making them more adaptive and resilient to change.”
This adaptability is certain to prove critical over the coming years, during which the likes of the Internet of things (IoT) and autonomous vehicles are expected to expand greatly in popularity. Such applications must continuously consume massive quantities of data to reflect ongoing changes in the external environment in real-time. Simply put, current static machine-learning models are unable to satiate this thirst for new and continuous data and thus can’t be deployed effectively for such use cases. Adaptive AI models, however, have the capacity to accommodate—and respond to—an endless stream of data.
The applications of adaptive AI could even prove life-saving, especially given the potential it has to improve the healthcare industry’s performance. The ability to consistently analyse data related to thousands, if not millions, of patient symptoms and vital signs can enable adaptive AI systems to optimise the clinical recommendations they produce.
Adaptive AI can even adapt to differentiate between a mix of patients in various regions of the country. “A hospital in Minneapolis may see a very different mix of patients than one in Baton Rouge, 1,200 miles down the Mississippi River, in terms of age, comorbidities such as obesity or diabetes, and other factors,” Sam Surette, head of regulatory affairs and quality assurance at AI-focused medical company Caption Health and former U.S. Food and Drug Administration (FDA) reviewer, wrote in an October 2020 article for health publication Stat. “Because clinically appropriate performance is set in part on factors such as disease prevalence, having access to local data can help fine-tune performance to match the needs of each institution. Adaptive AI might even learn subtle differences between institutions, such as how frequently they perform certain blood tests, which are otherwise difficult to factor into calculations.”
Over the long term, then, adaptive AI delivers faster, more accurate outcomes, which should mean that more meaningful insights can be gleaned by enterprises to further optimise decision-making. It also implies that adaptive AI systems can exercise more autonomy when delivering such outcomes—by having the ability to independently upgrade their own learning capacities in response to changes in the real world that were previously absent when each system was first developed. Thus, enterprises reliant on AI need not employ as much human capital within the process as was perhaps once required.
That said, adaptive-learning models rely on sound training in machine-learning capabilities. “It works best when trained on millions, or even billions, of customer interactions across different geographies, industries and use cases,” Mike Gozzo, chief product officer at Ada, an AI-based automated brand-platform provider, told digital customer-experience publication CMSWire in October 2022. “This creates a rich data set that drives personalized and proactive experiences for each customer, in every interaction.”
One must also acknowledge that simply having access to new data does not necessarily mean that an adaptive AI system will improve its performance, particularly if it learns the “wrong” lessons from the new data. Algorithms are designed to make predictions based on the information they consume. But if said information consistently demonstrates bias in terms of its sourcing, the system may exhibit those same biases when operating.
In 2018, for example, Amazon was forced to scrap its recruitment engine, which used AI to determine which resumes belonged to the best job applicants, because the models showed clear biases against women. This was largely because machine-learning models were trained using patterns in resumes submitted to the tech giant over the previous 10 years. As men submitted the overwhelming majority of resumes—thus reflecting the dominance of males within the tech industry—the AI system “learned” that male candidates were preferable to females.
And using healthcare again as an example, a 2019 research paper found clear evidence of racial biases in one commonly adopted algorithm, such that black patients assigned the same level of risk as white patients by the algorithm were actually more ill than the white patients. “The authors estimated that this racial bias reduces the number of Black patients identified for extra care by more than half,” the study noted. “Bias occurs because the algorithm uses health costs as a proxy for health needs. Less money is spent on Black patients who have the same level of need, and the algorithm thus falsely concludes that Black patients are healthier than equally sick White patients.”
Nonetheless, as the world increasingly relies on real-time data processing and analysis through the likes of IoT and as enterprises become more data-driven, adaptive AI systems will be built more frequently as the need to adapt to environmental changes spontaneously becomes ever more urgent. Gartner has predicted that by 2026, enterprises that have adopted AI engineering practices to build and manage adaptive AI systems will outperform their peers in the time and the number of processes it takes to operationalise AI models by at least 25 percent.
In the meantime, enterprises may have to rejig their business models to optimise their decision-making processes to exploit adaptive AI fully. “First, create the foundations of adaptive AI systems by complementing current AI implementations with continuous intelligence design patterns and event-stream capabilities—eventually moving toward agent-based methods to give more autonomy to systems components,” according to Gartner’s Brethenoux. “To make it easier for business users to adopt AI and contribute toward managing adaptive AI systems, incorporate explicit and measurable business indicators through operationalized systems, as well as incorporate trust within the decisioning framework.”
Enterprises will also almost certainly have to encompass ethical considerations regarding the appropriate and compliant use of AI when engaging in this reengineering, while regulators are bound to have their say in how adaptive AI should be deployed in a controlled manner.