Highlights
▪ any individual machine tool—such as a drill, a metal cutter, or a press—had to be totally redesigned to take advantage of having an individual unit electrical engine.5 (How AI can be application solution?)
▪ Nasdaq to Buy Anti-Financial Crime Firm Verafin for $2.75 Billion.” Verafin is headquartered in St. John’s, Newfoundlan
▪ What NASDAQ was buying was AI. Verafin had invested heavily and built tools that could predict fraud and validate the identity of bank customers. These were key functions of financial institutions both in terms of their operation and also in terms of their regulatory compliance. To do this requires big data, and bank and credit union data was the biggest of them all.
▪ economic properties of AI itself—lowering the cost of prediction—we underestimated the economics of building the new systems in which AIs must be embedded.
▪ employed large teams of data scientists for predicting fraud, money laundering, sanction noncompliance, and other criminal behavior in financial transactions.4
▪ “[I]f you work as a radiologist, you’re like the coyote that’s already over the edge of the cliff, but hasn’t yet looked down so doesn’t realize there’s no ground underneath him. People should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do better than radiologists.”6
▪ exploring neural networks to exploring human cognition (how we make decisions), social behavior (why people in some industries are keen to embrace AI quickly while others are resistant), production systems (how some decisions depend on others), and industry structures (how we’ve hidden certain decisions to shield ourselves from uncertainty).
▪ connect the dots and assemble an economic framework that distinguishes between point solutions and system solutions that would not only solve the Verafin puzzle but also provide a forecast for the next wave of AI adoption. By focusing on system solutions rather than point solutions, we could explain how this technology will eventually sweep across industries, entrenching some incumbents and disrupting others.
▪ Throughout the entire Industrial Revolution, factories were designed to leverage steam. As we have seen, a single source of power into the factory was distributed to individual machines through a central shaft upon which belts and pulleys were hung.
▪ it was a contraption where hundreds of moving parts were tied to a single entry point for power. Having a new type of power didn’t change that. But having new devices caused some entrepreneurs to rethink the factory.
▪ Electricity equalized the economic value of space, providing flexibility
▪ Ford was a car entrepreneur. But he was largely a system solution entrepreneur. These system changes altered the industrial landscape. Only then did electrification finally show up in the productivity statistics, and in a big way.6
▪ entrepreneur pitching electricity in 1920 would have figured out that the key value proposition was not “saving fuel costs” but rather “enabling vastly more productive factory design.”
▪ application solutions that require a redesign of devices or products around AI. All those robots powered by AI are applications
▪ What we have yet to see are the plethora of high-value system solutions for AI that are likely to emerge
▪ Given what we now know about AI, how would we design our products or services or factories if we were starting from scratch? The new flat factory architectures did not emerge first in traditional industries but rather in the newly emerging ones in the 1900s such as tobacco, fabricated metals, transportation equipment, and electrical machinery itself.
▪ For AI, we can ask these same two questions: (1) What is AI really giving us? (2) If we are designing our business from scratch, how would we build our processes and business models? If electricity was not “lower cost of energy” but rather “enable vastly more productive factory design,” so too, perhaps, is AI not “lower cost of prediction” but rather “enable vastly more productive products, services, and organizational design.”
▪ AI enables system-level innovation. Decisions are the key building block for such systems, and AI enhances decision-making.
▪ While a factory layout may be easy to see, the procedures, capabilities, and training underlying the new system may be less visible and hard to replicate. What is more, new systems can enable scale.
▪ New systems are hard to develop and also, as we will explore, difficult to copy because they are often complex. That creates opportunities for those who can innovate on systems.
▪ Computers were appearing all over the place without measured improvements in productivity
▪ In the first wave of electricity, light bulbs replaced candles and electric motors replaced steam engines. These were point solutions, with no restructuring required. The economy did not transform.
▪ AI will only reach its true potential when its benefits in providing prediction can be fully leveraged
▪ how decisions are made that the entire system of decision-making and its processes in organizations will need to adjust. Only then will AI adoption really take off.
▪ The promise is clear, but the path to achieving that promise is not. There needs to be a way to use machine predictions to do things better. That means using predictions to make better decisions.
▪ The US Census Bureau asked over 300,000 companies about their use of AI. The large firms that had adopted it overwhelmingly emphasized the use of AI for automating and improving existing processes. In other words, their AIs are point solutions and application solutions, so there was no change in the system.
▪ disruptive, we mean that it changes the roles of many people and companies within industries and, alongside those changes, causes shifts in power. That is, there are likely to be economic winners and losers, especially if system change occurs relatively quickly.
▪ it can be easier to build a new system from scratch than to change an existing system. So, historically, new entrants and startups often outperform established businesses when a total system redesign is required for optimization. Thus, system-level change is a path to disruption of incumbent firms.
▪ Many will likely experiment and fail because they misunderstand demand, or they can’t get the unit economics to work.
▪ If an AI prediction creates value by enhancing the focal decision and that value creation is independent of any other changes to the system, then a point solution (enhanced existing decision) or application solution (new decision) is feasible. However, if the value of the enhanced decision is not independent but rather requires other substantive changes to the system in orders to create value, then a system solution is required.
▪ What we have is an advance in statistical techniques rather than something that thinks. But the advance in statistical techniques is very significant. As that advance reaches its potential, it will dramatically reduce the cost of prediction. And prediction is something we do everywhere.
▪ The goal, therefore, was to guess the most likely correct label, which became the prediction.
▪ predictions are a key input into our decision-making.
▪ two other key inputs into decisions: judgment and data.
▪ on. As AIs acquire more high-quality data, the predictions improve. By quality, we mean that you have data about the context in which you are trying to predic
▪ Causal inference challenges limit the usefulness of AI to places where it is possible to collect the relevant data
▪ DeepMind ran millions of simulated experiments, and the machine learned to predict winning strategies by simulating what would happen if it tried several different approaches.3
▪ Randomized experiments are the main tool for statisticians to discover what causes what. They are the gold standard for approval of new medical treatments.
▪ War is the realm of uncertainty.” Prediction could reduce uncertainty and so generate substantial military advantage. The challenge, however, is that wars involve adversaries. In war, “if AI becomes good at optimizing the solution for any given problem, then an intelligent enemy has incentives to change the problem.”4 The enemy will go beyond the training set, and peacetime data will be of little use.
▪ A transaction is proposed that involves a request for payment, which is a transfer from one account to another. If the transaction is approved, the money changes hands, which itself triggers the exchange of real goods and services. If the transaction is not approved, no money is moved, which may impede the underlying real obligations
▪ AI is the means by which banks become better at the guessing game and reduce errors.
▪ AI is actually doing is improving banks’ ability to sort legitimate from fraudulent transactions at a much lower cost—that is, prediction.
▪ Banks and other financial institutions used to conduct their own prediction functions. Approval is what their business is. The better those decisions are made, the better they are at their job.
▪ If your business wants to adopt AI, it will likely have to clear the brush and maybe an entire forest before being in a position to implement it. This book is about that clearing process—identifying what needs to change and the dilemmas and challenges that will face you in implementing such change
▪ Many people consider shopping a burden, so by being able to provide that cheaply, better prediction offered a solution for that application.
▪ ship-then-shop is a system solution because it impacts other key decisions and requires a redesign of Amazon’s system to facilitate a much more cost-effective way of handling returns.
▪ Prediction is the process by which you convert information that you have into information that you need.
▪ the cost of prediction falls, the value of substitutes for machine prediction (e.g., human prediction) will fall.
▪ simple act of making decisions degrades one’s ability to make further decisions.
▪ Think about yourself for a bit, and you realize that most of what you decide are not actual decisions but latent ones, things you could choose but choose not to.
▪ bounded rationality
Ajay Agrawal, Joshua Gans, Avi Goldfarb - Power and Prediction_ The Disruptive Economics of Artificial Intelligence (2022).epub (Highlight: 42; Note: 0)
───────────────
▪ standard operating procedures (or SOPs). These are detailed documents describing procedures for doing things all over an organization.
▪ construction industry often breaks down the whole process into simpler tasks. There is a construction schedule with a line-by-line and day-by-day listing of every task to accomplish and in what order.9 The outcomes of those tasks are planned in advance. No one person on site has to think about more than their task.
▪ fixed set of SOPs can make it hard to change and adapt.
▪ point of AI is to allow for decisions, but when decisions are made, coordination becomes difficult.
▪ AIs provide little value for rules. AIs generate predictions, and predictions are a key information input to decision-making. So, as AIs become more powerful, they lower the cost of information (prediction) and increase the relative returns to decision-making compared to using rules.
▪ greenhouse gives the farmer extraordinary control over temperature, humidity, and irrigation.14 This control doesn’t come cheap. Heating, cooling, and supplemental light all require energy. The energy required is predictable and can be managed.
▪ If AI for pest prediction gets good enough, then greenhouses can operate differently. Farmers can grow pest-sensitive crops. Larger greenhouses become possible.
▪ If AI companies like Ecoation do a good enough job controlling pests, then we can replace existing rules and build a new system
▪ checklist exists because of uncertainty.
▪ checklists are not simply indicators that something has been done. Instead, they are the manifestation of rules and the need to follow them. They are there to ensure reliability and reduce error.
▪ Large businesses have checklists. They also have standard operating procedures (SOPs), which serve a similar role. As discussed in chapter 4, SOPs are large manuals identifying all the steps people need to follow, including checking off whether they have done them. The SOPs make complex organizations function. But we have to recognize them for what they represent. They are rules to follow rather than decisions to make.
▪ intensive personalized training works best. The challenge is how to deliver that personalized education at scale.
▪ entrepreneurship training to hundreds of thousands of new sellers. The training program involved dozens of possible modules and focused on setting up a website, marketing strategy, and customer service. For example, the training might provide a checklist for best practices in product descriptions so that customers understood what they were buying. Another aspect of the training focuses on search engine optimization and keyword selection.
▪ order to take advantage of prediction machines, we want to turn rules into decisions
▪ The better pitches were ones that were not focused on replacement but on value. These pitches demonstrated how an AI product could allow businesses to generate more profits by, say, supplying higher quality products to their own customers
▪ value-enhancing approach to AI, rather than a cost-savings approach, is more likely to find real traction for AI adoption.6
▪ Deep Medicine: How Artificial Intelligence Can Make HealthCare Human Again
▪ where AI has the most potential for transforming the economy, it is well upstream from most ordinary business activities: in the system of innovation and inventio
▪ new avenues of inquiry and improves productivity within the lab.5 As a new way to create products, rather than an improvement on a specific product, the economic impact of research tools is not limited to their ability to reduce the cost of innovation.6 Instead, they alter the playbook for innovation.
▪ relatively simple innovation systems like content recommendation engines and more complex systems like those for drug development
▪ recommender engine was an AI point solution that fit into the existing workflow
▪ future requires more labs that convert known protein structures into useful treatments.
▪ University of Toronto professor Alán Aspuru-Guzik is using AI for chemistry. An AI that predicts which hypotheses to test is integrated as part of a system solution involving AI-controlled robotic arms and a fully stocked portable lab for running automated experiments. He calls the system a “self-driving chemistry lab.”11
▪ Advances in lens-grinding technology led to innovations in the personal optics market (e.g., eyeglasses) but also in the research tools market (e.g., microscopes), which enabled further innovations in the innovation system
▪ There is work to do in order to innovate at a system level so that AI reaches its transformative potential.
▪ Prediction opened up a new way of organizing the electricity industry, with decentralized electricity generation. Better prediction meant that those events where demand greatly exceeded what was expected became rarer in real time.
▪ incumbent firms can find themselves “asking the wrong questions” regarding new technologies and their value to customers. Thus, they shy away from certain technologies that offer few advantages to their own customers
▪ When the prediction is good enough and the judgment and action are clear, automation is possible. Otherwise, let the human decide. This process is called judgment by exception.
▪ Machines don’t make decisions. Instead, machines can change the people who make the decisions, from individuals deciding at the moment a decision is made, to individuals judging what matters before the specific decision arrives.
▪ nobody ever lost a job to a robot. They lost their job because of the way someone decided to program a robot.
▪ The machine has suffered for the sins of capitalism.”6 Indeed, the whole term—capitalism—seems to evoke machine power. In reality, it is the humans who apply judgment as coded in machines that have that power. Those humans are responsible, and there is a need for the legal and regulatory systems to understand that.
▪ Using machine learning to read, classify, and then present insights from scientific research, BenchSci found that scientists could run half as many experiments as normally required to advance a drug to clinical trials.
▪ Technology and Market Structure, London School of Economics professor John Sutton
▪ Insights is a trigger word for us because it represents precisely the wrong way of thinking about how an AI advance will create value.
▪ AI only has value if it leads to better decision-making. And this means that the new opportunities for value creation from AI are all about how they improve decisions.
▪ Decisions are what put the “general” in AI as a general purpose technology.
▪ • Prediction and judgment are the two primary ingredients for decision-making. In a decision tree, prediction generates the probability that each branch in the tree will occur. Judgment generates the payoffs associated with the outcomes at the ends of each branch. Usually, we make decisions without recognizing that the predictions and judgment are two separate inputs as they are both in the mind of the same person (the decision-maker). When we introduce AI, we shift the prediction from a person to a machine, and thus we decouple the prediction from the judgment. That may change who provides the judgment.
▪ Thinking in bets requires you to recognize that predictions are uncertain, and to understand that the outcomes you experience are partly determined by luck.
▪ These are the two ways judgment is built. Either it involves planned learning from someone else by reading, instructions, or culture, or it is learned by experience
▪ modularity made innovation easier; for example, when we switched from analog to digital telephones, that involved changing the dialing device but not the network itself.
▪ consider the operations of Amazon, which supplies millions of products worldwide. Amazon procures them, stores them in warehouses, captures customer orders, and ships to those customers. But it also involves helping the customer figure out what to purchase in the first place, that is, providing them with recommendations.
▪ Successful AI systems enable coordination where possible and modularity where necessary.