Virtual AI designer. Source: iStock Video Portfolio # 2101156127
getty
“There are no solutions, only tradeoffs,” writes Thomas Sowell. The implementation of AI systems must reflect human values in the final decision-making, as well as the difficult trade-offs inherent in many decisions.
Let’s say a wholesaler’s warehouse was laid out 50 years ago and then adjusted from time to time as space needs changed. These include the introduction of new products, the discontinuation of old products, and increases or decreases in the sales volume of individual products. Warehouses are a mess, but AI will take care of it. The AI-powered tool is populated with data about every product you store, including size, weight, and sales volume. The AI will likely also be given a catalog of shelving and material handling options, such as forklifts and conveyor belts. Safety standards are also provided to the system.
This sounds like a great approach until you consider your goals and tradeoffs.
Capital-labor trade-offs in AI decision-making
Humans must decide what to make AI do. One obvious goal of warehouse redesign might be to minimize the time people spend moving things. But wait a minute. What kind of warehouse supplies catalog is the AI looking at? It could be a low-budget catalog or a Neiman Marcus approach. There is always a trade-off between capital costs and labor costs. Capital investment decisions require information about how long the capital investment will last and how labor costs will change in the future. This choice will also be influenced by improvements in material handling technology expected in the coming years, including robotics. In a static world, decisions can be complex but decisive. That is, the input uniquely determines the output. But in a changing world, some decisions need to be made about the future of technology and the outlook for labor costs. And by focusing on the differences in price trends between robotic forklifts and roller trucks, the ruling may have been quite granular.
This trade-off between capital and labor spending occurs in many areas, such as ordering kiosks in fast-food restaurants, online reservation management in medical clinics, and red-light cameras replacing police officers. Lower interest rates justify further automation, but it still depends on forecasting future labor costs. We’re used to rising wages, but will AI reduce the demand for labor so much that human workers will actually be cheaper in the future? We can ask AI systems to develop technology and predict future labor costs, but we shouldn’t expect them to be perfect. Important business decisions require human judgment.
Current needs and flexibility trade-offs in AI decision-making
The design of many business systems and nearly all physical structures involves another trade-off: efficiency to meet current needs and flexibility to change. That warehouse must accommodate new products, new sales volumes, and possibly different distribution and shipping patterns. Modifying a system that is currently optimized can be costly.
AI helps develop information to analyze future changes by looking back. How often are new products added and old products discontinued? Do boxes tend to get heavier? But these aren’t things that AI can perfectly predict. AI models may be better at making predictions than humans, but humans should be wary of entrusting important business decisions to AI.
Again, this tradeoff applies to many business process decisions. If we separate marketing to enterprise customers from marketing to small businesses, will we continue to accept that decision forever? If not, will it be difficult to change? If we stop hiring engineers from universities, will it be difficult to attend job fairs again next year?
AI systems can optimize many functions, but how easily they can adapt to changing circumstances needs to be part of their decision-making. And this requires determining both the likelihood that changes will be needed in the future and the cost of implementing the changes.
Safety tradeoffs in AI decision making
In the warehouse example, we proposed providing safety requirements to the AI system. In many businesses, safety standards are created by government regulators, industry associations, or at least by common practices within the industry. Is there any point in going beyond normal standards?
Occupational accidents in certain situations may be reduced by higher cost designs. How much is the company willing to pay to reduce injuries? It can be calculated in dollars and cents based on medical costs and overtime. U.S. companies under the workers’ compensation system can learn the cost of an accident from their insurance company. But number crunching alone cannot cover the pain of the injured person or the fear that other employees may have after witnessing an injury at work. The company’s values and culture come into play.
Workplace safety immediately comes to mind, but consumer product safety also deserves attention. Data security goes beyond counting a few cents and also includes customer satisfaction, which decreases when data is hacked.
Trade-offs and values
Human values are at the center of trade-offs in many important decisions. The comedy video series “Murderbots” depicts robots baffled by humans who abandon group loyalty and take personal risks. But that’s what we humans do. We care about the welfare, safety, and feelings of others. When human values are involved, humans must drive the choices. AI can help understand trade-offs, but ultimately humans must make the key decisions.
