AI systems are becoming more complex, requiring extensive data, computation, and engineering to develop, deploy, and maintain. However, this complexity also creates technical debt, which is the cost of closing the gap between acceptable and optimal AI functionality. Gartner predicts that by 2027, data science organizations will cut AI technical debt by 70% by using simulation platforms and technologies to manage the complexity of AI systems.[1] These platforms will help accelerate discovery and innovation processes while improving the quality and performance of AI solutions. As a result, simulation platforms will become essential tools for data science organizations to navigate complexity in enterprise decision-making.
The presence of technical debt is a formidable hindrance for AI systems to achieve optimal performance and stability for businesses. Several factors contribute to this challenge, including outdated hardware, inefficient software architecture, inadequate maintenance of data pipelines, and suboptimal model performance. Addressing these issues will be crucial to ensure that AI systems remain scalable, reliable, and efficient while minimizing the risk of errors and failures. According to The AnyLogic Company, leaders consistently miss big opportunities when faced with disruption because traditional analysis methods are inadequate when faced with innovation. [2] AI is frequently promoted as a way to simplify processes, but it can actually leave companies burdened with technical debt. According to McKinsey & Company, CIOs estimate that tech debt amounts to 20 to 40 percent of the value of their entire technology estate (before depreciation).[3] Therefore, it’s essential to continuously evaluate and eliminate technical debt in AI systems to sustain their effectiveness and enhance their long-term performance.
Walmart, the world’s largest retailer by revenue, was looking for an automation technology that would help complete orders faster and at a lower cost in the company’s fast-growing online grocery business. They wanted to evaluate Alert Innovation’s Goods-to-Person (GTP) concept, Alphabot®, that could automate online grocery at the store level by using autonomous mobile robots capable of operating in all three dimensions within a multilevel storage structure. Alphabot robots, or “bots”, are self-driving vehicles that can gather items in ambient, chilled, and frozen temperature zones in a high-density storage system and bring them to associates that pick individual items to build a customer’s order. Alert Innovation presented Alphabot® as a technology that would make the in-store fulfillment of online orders faster and more efficient. Before making a financial commitment and deploying the system in a Walmart store, it was decided to task MOSIMTEC, a simulation consulting firm, with designing a material handling simulation model for an independent technology feasibility assessment. MOSIMTEC chose AnyLogic material handling simulation software for the project. MOSIMTEC’s and AnyLogic’s abilities to dynamically build facility layouts from data inputs, without accessing the development environment for each layout change, would help cut model development time significantly and enable faster evaluation of multiple Alphabot® configurations. AnyLogic also offered unparallel ease of deployment, so that multiple Walmart engineers could run the material handling design model, without the need to install additional software or purchase developer’s licenses. AnyLogic was also selected because the Alphabot® system would require extensive control algorithms. AnyLogic’s ability to integrate with Java eliminated excessive time spent translating algorithm ideas back and forth between a propriety scripting language and a format that programmers would be comfortable with. By estimating equipment requirements needed to meet various turn-around-time thresholds, the outputs from the initial material handling design model informed the business case for deploying Alphabot® more widely. The simulation model quantified system performance capability under unconstrained demand conditions to benchmark its limits. The model showed that Alphabot® would be able to pick 95% of online grocery orders in less than eight minutes, with an average pick time to be under five minutes. The initial model was later updated and expanded to understand the impact of various detailed design alternatives. The model helped Alert Innovation determine which design alternatives would result in the largest ROI, along with better sizing out the system for future deployments. Walmart and Alert Innovation launched a proof-of-concept pilot of Alphabot® at a Walmart supercenter in Salem, New Hampshire in March 2019. [4]
Simulation platforms and technologies have become critical assets for data science organizations as they tackle the challenge of managing complex AI systems. These platforms provide data scientists with the ability to create accurate and dynamic reproductions of their AI ecosystems through various methods, such as discrete events, agent-based, system dynamics, and material handling modeling. The use of simulation platforms offers numerous advantages, including the ability to train AI agents with reinforcement learning, test or embed machine learning models, generate synthetic data, optimize parameters, and evaluate outcomes. These platforms enable data scientists to significantly reduce the time, resources, and manual coding required to develop and enhance their AI systems, thereby improving their overall efficiency and effectiveness.
Prominent players in the industry are displaying proactive behavior by investing in simulation platforms and advanced technologies that can help them maintain a competitive edge. These companies are fostering a culture of innovation by experimenting with new ideas and concepts, streamlining their AI workflows, and effectively managing their systems through AI. In addition, they are cultivating a team of proficient AI professionals who can offer valuable insights and guidance on the latest AI trends and developments. According to KPMG, companies should consider a positive and dynamic approach to building in control as part of their AI development strategy.[5] These measures are crucial for reducing AI technical debt and realizing the full potential of AI for their business. By staying ahead of the game, leading companies are better equipped to tackle challenges and capitalize on opportunities in the ever-evolving landscape of AI.
