AI is undoubtedly a hot topic these days, stirring up discussions everywhere. Can it outperform us in writing? Can it generate art? Is it going to take over our jobs? Interestingly, while AI might exhibit almost human-like abilities and even surpass us in certain tasks, it still falters on simple questions such as the current time or performing basic multiplication. Seems less impressive now, doesn’t it?
Arindam Paul, vice president of data science at Saifr® recently explored what the next stages of AI might be and how it could continue to evolve in the coming years.
As we delve deeper into the current state of AI and its potential future trajectory, Paul stated that we must consider the evolution from single-step to multi-step problem-solving. Up to this point, the AI we have interacted with focuses primarily on single-step problems. Given an input, it performs computations and yields an output. An example could be sentiment analysis of a text or identifying creditworthiness based on loan default data of the past decade.
Nevertheless, a vast array of problems in the real world are not single step. Herein lies the next frontier for AI – tackling multi-step problems. Humans check a watch or use a calculator for time or multiplication, respectively. Eventually, computers should be capable of consulting external tools to accomplish tasks. This potential future of AI, termed as action-oriented or autonomous AI, signifies computers learning to interact with the world to derive conclusions and make predictions.
Consider an AI assisting an analyst with research. Initially, the AI would utilise its “memory” or training data to generate an output – a single-step problem. However, for a more comprehensive output, the AI could leverage internet searches, APIs, and more – effectively a multi-step process. This could potentially be the direction in which AI is headed, Paul said.
As with any forward-looking statements, predictions are inherently uncertain and speculative. However, the field of AI is evolving at an unprecedented pace. Although the techniques are still maturing, we are witnessing remarkable advancements. If developments proceed as expected, AI could help tackle currently insurmountable challenges and redefine what’s possible, heralding an exciting new era.
Furthermore, expect to see progress in various AI areas including improvements in foundation models, enhanced zero-shot capabilities, better sample efficiency, increased reliability and control, more affordable dedicated hardware for training and inference, and optimised model architecture.
Read the full story here.
Keep up with all the latest FinTech news here.
Copyright © 2023 FinTech Global
