Virtual Ai Designer. Source: iStock video portfolio # 2101156127
getty
“There are no solutions, only tradeoffs,” Thomas Sowell wrote. Implementation of AI systems needs to reflect the difficult tradeoffs inherent in many decisions, with human values reflected in final decisions.
Let’s say a wholesaler’s warehouse was laid out 50 years ago, then adjusted now and then as space needs changed: new products were introduced, old ones discontinued and sales volumes for individual items grew or shrank. The warehouse is a mess, but AI can fix that. An AI-powered tool will be fed data on all of the products to be stored, including sizes and weights as well as sales volumes. The AI will probably also be given a catalog of shelving and material handling options, such as fork lifts and conveyor belts. Safety standards would also be provided to the system.
This sounds like a great approach until one considers goals and tradeoffs.
Capital-Labor Tradeoffs In AI Decisions
Humans have to decide what the AI should be asked to do. One obvious goal for the warehouse redesign might be to minimize human time spent moving things. But hold on—what warehouse fixtures catalog is the AI looking at? There might be a low-budget catalogue or a Neiman-Marcus approach. There’s always a tradeoff between capital costs and labor costs. The decision on capital expenditures needs information on how long the capital investment will last and how labor costs will change in the future. The choice will also be influenced by likely improvements in material handling technology in the coming years, such as robots. In a static world, the decision may be complex but deterministic: the inputs uniquely determine the output. But in a changing world, some judgment must be made about the future of technology and the outlook for labor costs. And the judgments may have be fairly granular, noting the difference in price trajectory between robotic forklifts and roller tracks.
This tradeoff between capital and labor spending occurs in many areas: order kiosks in a fast food restaurant, online appointment management at a health clinic, red light cameras in lieu of police officers. If interest rates are low, more automation is justified—but still dependent on a forecast of future labor costs. We are used to wage rates rising, but will AI reduce demand for labor so much that human workers will actually be cheaper in the future? We can ask an AI system to forecast technological development and future labor costs, but we should not expect perfection. Human judgment will be needed for critical business judgments.
Current Needs Versus Flexibility Tradeoffs In AI Decisions
Designs of many business systems and almost any physical structure have another tradeoff: efficiency in meeting current needs versus flexibility for change. That warehouse will need to accommodate new products, new sales volumes and possibly different patterns of deliveries and shipments. A system optimized for today may be costly to change.
AI can help develop information for analysis of future change by looking at the past. How frequently are new products added and old ones discontinued? Is there a trend toward heavier boxes? But these are predictions that AI will not make perfectly. An AI model may come to be better at predictions than humans, but humans should be cautious about turning major business decisions over to the AI.
Again, this tradeoff applies to many business process decisions. If we separate marketing to enterprise clients from marketing to small businesses, will we want to live with that decision forever? If not, will it be hard to change? If we stop recruiting engineers from State U, will it be hard to get back into their job fair next year?
AI systems can optimize many functions, but how easy to adapt to changing circumstances must be part of any decision. And that requires a judgement about both the likelihood of wanting a change in the future and the cost of implementing a change.
Safety Tradeoffs In AI Decisions
The warehouse example suggested providing the AI system with safety requirements. In many operations, safety standards are created by government regulators or trade associations, or at least by common practice within an industry. Would it ever make sense to go beyond the usual standards?
On-the-job accidents in a particular situation might be reduced by a higher-cost design. How much is the company willing to pay to reduce injuries? There could be a dollars-and-cents calculation based on medical costs and out-of-work time. American companies under the workers’ compensation system can get an idea of the cost of accidents from their insurer. But a numerical computation won’t cover the pain to the injured person nor the fear that other employees may have after witnessing a workplace injury. The company’s values and culture come into play.
On-the-job safety comes to mind quickly, but consumer product safety merits attention. Data security goes beyond the dollars-and-cents calculus to include customer satisfaction—which will drop after their data are hacked.
Tradeoffs And Values
Many important decisions have human values at the center of the tradeoff. The comedy video series Murderbot has a robot puzzled by humans taking personal risks out of group loyalty. But that’s something that we humans do. We care about others: their welfare, their safety, their feelings. When human values come into play, humans must drive the choices. AI can help us understand the tradeoffs, but ultimately people must make the critical decisions.
