Power and Prediction: The Future of AI Systems
Across the globe, 59 percent of companies have begun to implement artificial intelligence (AI) business strategies, however, of those companies, only 11 percent have received a significant financial benefit (Ajay et al.). There is a clear interest in the development of AI and the introduction of those systems into the workplace, however, many business leaders have struggled to effectively implement AI into their current business models. In Power and Prediction, the authors Ajay Agrawal, Joshua Gans, and Avi Goldfarb explore the necessary systems to scale AI beyond its current uses. Through their research, they make an insightful argument on the problems business leaders must overcome in order to successfully integrate AI technologies into the workplace.
Prediction is one of the most time-consuming aspects of decision-making. AI will transform the decision-making process by creating more accurate and faster predictions therefore reducing the cost of decision making for businesses. This is exemplified in decisions such as choosing a restaurant or picking clothes in the morning. Many high-ranking leaders depend on routine to limit the time consumption connected to making low risk decisions. For example, Barack Obama and Mark Zuckerberg routinely wear the same dark-colored clothing to limit the cognitive load associated with deciding what to wear (Ajay et al.). This allows them to have excess time for other more demanding decisions related to their jobs. High-risk decisions require more prediction and demand individuals spend additional time contemplating different options and scenarios. However, AI can quickly predict what the user will want and present them with high-quality options so that they can focus on using judgment to select the best choice. The authors define this process as decoupling and identify it as the key to creating system-level innovation with AI. The authors suggest “that by decoupling prediction from the other aspects of a decision and transferring prediction from humans to machines, AI enables system-level innovation (Ajay et al.)” This is possible because once the prediction capabilities of AI are perfected, then AI will be able to make better and quicker suggestions than a person. For systems change, we will need to make foundational developments to our businesses that shift prediction tasks to machines and reallocate that time for people to make stronger judgment calls. These two methods working in tandem with each other will result in timely and quality decisions.
The authors share how artificial intelligence was used in Flint Michigan to determine which pipes might be contaminated with lead. The machines used data to run thousands of tests to create accurate and quick suggestions of which pipes were most likely to have lead. The machines could predict the location of lead pipes with 80% accuracy (Ajay et al.). City officials were then able to use their judgment to make the final decision of where to dig based on the AI predictions. This use of AI was significant because in this situation there was an urgent need for support in Flint, Michigan. This was a very high stakes situation that without AI would require significant funding to dig in each neighborhood to check for the presence of lead pipes. According to the MIT technology review, “The algorithm is saving about $10 million as part of an effort to replace the city’s water infrastructure (Winick)”. By using AI in the pipe selection process people were able to receive help faster without the additional cost to the city for digging. Authors suggest that by using this same predictive power as a baseline for strategic management business leaders and politicians will be equipped with the power to disrupt decision-making for future industries.
The strength of the book is its ability to inspire emerging professionals and aspiring entrepreneurs to implement systemic change using artificial intelligence. They argue that now is a pivotal moment where machines have the power to disrupt our entire economy if leveraged correctly. Leaders who have been able to utilize AI in their decision-making process have not only removed toxic pipes in Flint Michigan but also increased early diagnosis rates of cardiovascular disease, and coached sailing teams to victory. In 2021, the New Zealand sailing team used artificial intelligence to run hundreds of scenarios for the best sailing techniques to use for the America’s Cup competition course. In 8 weeks, the AI was able to make technique simulations that beat the New Zealand sailing team, many of whom have trained their entire lives (Ajay et al.). The New Zealand sailing team then used the methods tested by the AI to surpass their competition in that year’s race (Ajay et al.). The New Zealand sailing team’s success stems from transforming their training habits to advantage of AI’s predictive power. The authors urge emerging leaders and new businesses to redesign industries to be centered around artificial intelligence. The result will be revolutionary and look completely different than the way systems function currently. By reworking the setup to be centered around artificial intelligence business leaders will be supplied with accurate predictions which streamlines the decision-making process and makes business models more efficient and accurate.
The authors draw on historical, current, and hypothetical uses of artificial intelligence to illustrate how implementing AI prediction into decision-making will unlock the ability to rethink our systems with large-scale impact. The authors identify the present as the “In Between Times” where the introduction of the technology and supplementary systems must be created that allow the technology to be adopted widely (Ajay et al.). They argue that this trajectory of artificial intelligence parallels the history of electricity and the internet. This means that within a few decades, artificial intelligence will be as essential to our everyday lives as light bulbs. In reading this text new business owners will be excited to learn that they can be part of the change. Technology such as electricity has been able to dominate our currency society because business leaders began recreating systems to be centered around electricity such as redesigning factories, households, and institutions to depend on electricity in fundamental areas of living (Ajay et al.).
The authors argue that the barrier to the revolution of artificial intelligence is the current application of point solutions rather than systems change solutions. Many of the current uses of AI are focused on solving one problem within a system rather than changing the system itself. For example, insurance companies use AI to predict if homeowners are at risk of flooding or house fire in order to give premiums. The authors argue that a better system change level solution would be “shifting from a business model predicated on pooling risk among customers and transferring risk from customers to carriers to also mitigating risk for their customers (Ajay et al.).” In this example rather than using the prediction to charge clients after the incident, the insurance companies can prevent the problem altogether. With the traditional system, this would lead to a significant drop in income. However, from a systems change point of view, this new AI could be an opportunity to change the business model to provide the service of preventative care (Ajay et al). Where the homeowner would pay to have the insurance company predict the best changes to make in order to prevent their problems and challenges. The authors are encouraging all business leaders to change their mindset in this way to find the best way to implement AI.
Despite the compelling narrative surrounding systems change, the suggested steps towards creating the system’s change are vague and pose significant difficulties in implementation. In established companies, creating artificial intelligence-centered systemic change requires a full revamp of all technology and rules. One example the authors used to illustrate artificial intelligence based systems change was Amazon’s predictive shipping. Similar to how targeted ads will pop up when scrolling on the Amazon home page. Amazon has a plan of implementing predictive shipping where the AI algorithm will predict what you want to buy, but instead of the suggestion being at the front of your home page, they will show up on your front door. Then users will shop from the materials that suggested for them (Ajay et al.). In order to implement Amazon’s predictive shipping there must be additional technology created to supplement the predictive elements of artificial intelligence. The unwanted packages shipped to customers’ doors will need to be returned to the Amazon facility. Since Amazon currently doesn’t have a formal return policy, Amazon workers often dispose of the items that are returned because it is easier and more cost-effective to dispose of the items than restock them in their place. As a result, Amazon’s predictive shipping contributes to excessive waste and unsustainable business practices. Additionally, even though it is exciting technology many customers will be unsatisfied with having packages sent to their homes without their knowledge or consent.
Contrary to the hopeful solutions that are suggested in Power and Prediction, the creation of successful systems change requires a lot of strategic planning and time that is not fully outlined in the book. The methods to center artificial intelligence are supported by a few vague examples of modular changes and horizontal changes used for systems change in the restaurant industry. In the Designing Reliable Systems chapter the authors attempt to use restaurants ordering food from a farm as a suggestion for how to create AI systems change. The examples and suggestions lack concrete guidance on overcoming practical hurdles in implementing artificial intelligence. Hard to implement suggestions are a common pitfall of research in the realm of artificial intelligence. According to researchers in the product innovation management journal, “Even in academic studies, most contributions offer evidence based mainly on a few successful cases, but not explicitly rooted in any theoretical lens.” “Power and Prediction” falls into this category by using the same three or four examples many times throughout the text without clear evidence for supplemental research. The example of pipes and Flint Michigan was used many times throughout the text as evidence.
After AI systems change is implemented there are still significant risks associated with the refinement and upkeep of the new systems. Possible risks include: “a) algorithmic bias and allocative harms; b) unequal access and benefits; c) cascading failures and external disruptions, and d) trade-offs between efficiency and resilience. (Galaz).” Like many forms of change, the new systems are vulnerable to the introduction of systemic problems that can disadvantage marginalized communities. Artificial intelligence bias was briefly mentioned in the last pages of the book, however, it was not followed up with current or future methods for mitigating this bias. Many leaders have already identified artificial intelligence bias as a major roadblock to the full adoption of artificial intelligence. As a result, readers may be left unaware of the risk of bias in the new systems and how to be proactive in mitigating discrimination with the new AI.
While “Power and Prediction” did not provide clear next steps it is extremely convincing in getting future leaders to think about using artificial intelligence in ways that haven’t been attempted before. Since the publishing of the book, the author’s research has become the center of conversations about prediction and artificial intelligence systems change. The authors have been invited to share their findings with Harvard Business Review, TechTalks, Forbes, The Economist, Economist Innovation Summit, and the Trento Festival of Economics (Goldfarb). Their work has already begun discussions with well-known businesses to incorporate more AI into their strategic models. The theoretical hypothesis for the future of AI was exciting and captured the attention of many leaders. This book will likely start as a jumping-off point for many companies to do system change work. In response to the first book Prediction Machines the author’s findings we used to support the research of Turing Award winners, three new Canadian machine learning institutions, and global corporations such as Nvidia, Johnson and Johnson, and Uber (Ajay et. al.). The authors’ positions as leaders in artificial intelligence research have already shaped the trajectory of AI systems change in artificial intelligence. The next contributions and development of their theories will create an even larger impact amongst business leaders and politicians.
It is unlikely that the findings in this book will decrease the time it takes to implement technology from 40 to 10 years, like the authors suggest. However, this additional time will allow for the development of regulations and protection against the dangers of systemic injustice. Addressing issues of AI bias and inclusivity is essential to realizing the full potential of artificial intelligence. The “In Between Time” will allow artificial intelligence developers to make advances in systems and source data necessary to implement AI in a way that is inclusive of all users. Artificial intelligence developers should take this time to fully understand the gravity of the future of artificial intelligence and its impact on our society. Then use further research to inform decisions when developing the road map for artificial intelligence systems change.
Works Cited
Agrawal, Ajay,, Joshua Gans, and Avi Goldfarb. Power and Prediction: The Disruptive Economics of Artificial Intelligence Harvard Business Review Press, 2022.
Galaz, Victor, et al. “Artificial Intelligence, Systemic Risks, and Sustainability.” Technology in Society, vol. 67, 2021, pp. 101741-, https://doi.org/10.1016/j.techsoc.2021.101741.
Gama, Fábio, and Stefano Magistretti. “Artificial Intelligence in Innovation Management: A Review of Innovation Capabilities and a Taxonomy of AI Applications.” The Journal of Product Innovation Management, 2023, https://doi.org/10.1111/jpim.12698.
Goldfarb, Avi. “Artificial Intelligence.” AVI GOLDFARB, www.avigoldfarb.com/artificial-intelligence. Accessed 1 Apr. 2024.
Winick, Erin. “AI Is Helping Find Lead Pipes in Flint, Michigan.” MIT Technology Review.Com, 2018.
