Entrusting your money to a bank once seemed strange and risky. Similarly, entrusting all of your data to a company and letting its algorithms build a detailed model of you from it might seem to be an odd or even dangerous idea, but we’ll all soon take it for granted.
A decade from now, your personal model will be more indispensable than your smartphone, and the company that provides it may well be the world’s first trillion-dollar business. So it is time to start getting acquainted with our digital alter egos—and what they’ll mean for our lives.
Today, several different companies gather information about you and use machine-learning algorithms—computer programs that build models from data—to predict what you may want to buy and figure out how to sell stuff to you.
Privacy concerns aside, this poses two problems. First, companies have a conflict of interest: They want to serve you, but they also want to make money.
For example, Google predicts how likely you are to click on an ad to show you the most profitable ones. The choice also depends on the advertisers’ bids, but you’d probably rather just see the ads most relevant to you. Google co-founder Sergey Brin says that Google wants to be the third half of your brain, but nobody wants part of their brain constantly trying to show them ads.
The second problem is that a model of you derived from fragments of your data—Google’s model based on your searches, Amazon’s from your purchases and so on—can only ever have a very limited understanding of who you are and what you want. A single model assembled from all the data you’ve ever produced would be much more accurate: The more data, the better the model. For privacy reasons, you’d want the data and the model under your control, not a third party’s.
Soon enough, facing the fog of life without a good model to guide you will seem unendurable.
To solve both of these problems, we need a new kind of company that is to your data like your bank is to your money—storing it, keeping it safe and investing it on your behalf. For a subscription fee, such a firm would record your every interaction with the digital world, build and maintain a 360-degree model of you, and use it to negotiate with other people’s models.
No major technical obstacles would prevent doing this: The main requirement would be routing your interactions through what’s called a proxy server. If all your interactions with the digital world—through your smartphone, desktop computer or any other device—pass through a “middleman” computer in the cloud en route to their destination, the middleman can record them all.
The companies that now offer to consolidate all your data somewhere in the cloud are forerunners of tomorrow’s personal databanks. Once a firm has your data in one place, it can create a complete model of you using one of the major machine-learning techniques: inducing rules, mimicking the way neurons in the brain learn, simulating evolution, probabilistically weighing the evidence for different hypotheses or reasoning by analogy. Then you can go to town with your model, which you’d own and control like you do your money, rather than letting companies such as Apple, Google and Facebook FB -3.77 % fight for control of it.
With this in mind, here’s a future suggestion for LinkedIn: Add a “Find Me a Job” button. When you click it, your digital model would “interview” instantly for all the open positions that match your specifications, interacting at high speed with human-resources departments’ recruiting models. LinkedIn could then return a list of the most promising jobs for you.
While one copy of your model is doing this, another online alter ego could be looking for a car for you, exhaustively researching the options and haggling with the auto-dealer bots so you don’t have to.
At any moment, millions of copies of your model could roam the Internet, doing all the things you’d do if only you had the time. From these, your model selects the best few options for you to choose—then learns from what you decided, making the model more accurate the next time around.
To offset organizations’ data-gathering advantages, like-minded individuals will pool the data in their banks and use the models learned from that information.
As the models improve, their interactions will become increasingly like real-world ones—just millions of times faster and in silicon. Your model will go on a thousand digital dates with each of a thousand possible spouses and rank them by how well the dates went.
Tomorrow’s cyberspace will be a society of models, a vast parallel world that selects only the most promising things to try out in the real one—the new, global subconscious of the human race.
This will all probably happen in years, not decades. Apple’s Siri, Microsoft MSFT -0.90 % ’s Cortana and Google Now all include efforts to build complete models of you from the data captured by your smartphone, and they’re making rapid progress. Like a personal assistant, they try to help you accomplish your daily tasks, either in response to your commands (Siri) or on their own (Google Now). But to do that, they need to understand you, and they’ll use any data they can to do so—from the smartphone’s sensors to your emails and calendar.
The Web pages you see every day are already the result of complex interactions among the models that content providers, advertising networks and advertisers are deriving. Learning algorithms trade against one another in the stock market. Last May, a Hong Kong venture fund named Deep Knowledge Ventures appointed an algorithm to its board, voting on investment decisions alongside the five human directors, according to Business Insider.
Today’s models don’t yet interact with us: You can’t tell them they’re wrong or ask them questions. Machine-learning algorithms are black boxes that only computer scientists can open up. But that will change as more of us realize how important machine learning is and demand a say in how it occurs.
Eventually, your model will be like your best friend, but with infinitely more patience. What will you ask it? You might not like some of its answers, but that would be all the more reason to ponder them. Your model—your digital half—might even help you become a better person.
—Dr. Domingos is a professor of computer science at the University of Washington and the author of “The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World” (Basic Books).
Click here to view original web page at Get Ready for Your Digital Model
Entrusting your money to a bank once seemed strange and risky. Similarly, entrusting all of your data to a company and letting its algorithms build a detailed model of you from it might seem to be an odd or even dangerous idea, but we’ll all soon take it for granted.
A decade from now, your personal model will be more indispensable than your smartphone, and the company that provides it may well be the world’s first trillion-dollar business. So it is time to start getting acquainted with our digital alter egos—and what they’ll mean for our lives.
Today, several different companies gather information about you and use machine-learning algorithms—computer programs that build models from data—to predict what you may want to buy and figure out how to sell stuff to you.
Privacy concerns aside, this poses two problems. First, companies have a conflict of interest: They want to serve you, but they also want to make money.
For example, Google predicts how likely you are to click on an ad to show you the most profitable ones. The choice also depends on the advertisers’ bids, but you’d probably rather just see the ads most relevant to you. Google co-founder Sergey Brin says that Google wants to be the third half of your brain, but nobody wants part of their brain constantly trying to show them ads.
The second problem is that a model of you derived from fragments of your data—Google’s model based on your searches, Amazon’s from your purchases and so on—can only ever have a very limited understanding of who you are and what you want. A single model assembled from all the data you’ve ever produced would be much more accurate: The more data, the better the model. For privacy reasons, you’d want the data and the model under your control, not a third party’s.
Soon enough, facing the fog of life without a good model to guide you will seem unendurable.
To solve both of these problems, we need a new kind of company that is to your data like your bank is to your money—storing it, keeping it safe and investing it on your behalf. For a subscription fee, such a firm would record your every interaction with the digital world, build and maintain a 360-degree model of you, and use it to negotiate with other people’s models.
No major technical obstacles would prevent doing this: The main requirement would be routing your interactions through what’s called a proxy server. If all your interactions with the digital world—through your smartphone, desktop computer or any other device—pass through a “middleman” computer in the cloud en route to their destination, the middleman can record them all.
The companies that now offer to consolidate all your data somewhere in the cloud are forerunners of tomorrow’s personal databanks. Once a firm has your data in one place, it can create a complete model of you using one of the major machine-learning techniques: inducing rules, mimicking the way neurons in the brain learn, simulating evolution, probabilistically weighing the evidence for different hypotheses or reasoning by analogy. Then you can go to town with your model, which you’d own and control like you do your money, rather than letting companies such as Apple, Google and Facebook FB -3.77 % fight for control of it.
With this in mind, here’s a future suggestion for LinkedIn: Add a “Find Me a Job” button. When you click it, your digital model would “interview” instantly for all the open positions that match your specifications, interacting at high speed with human-resources departments’ recruiting models. LinkedIn could then return a list of the most promising jobs for you.
While one copy of your model is doing this, another online alter ego could be looking for a car for you, exhaustively researching the options and haggling with the auto-dealer bots so you don’t have to.
At any moment, millions of copies of your model could roam the Internet, doing all the things you’d do if only you had the time. From these, your model selects the best few options for you to choose—then learns from what you decided, making the model more accurate the next time around.
To offset organizations’ data-gathering advantages, like-minded individuals will pool the data in their banks and use the models learned from that information.
As the models improve, their interactions will become increasingly like real-world ones—just millions of times faster and in silicon. Your model will go on a thousand digital dates with each of a thousand possible spouses and rank them by how well the dates went.
Tomorrow’s cyberspace will be a society of models, a vast parallel world that selects only the most promising things to try out in the real one—the new, global subconscious of the human race.
This will all probably happen in years, not decades. Apple’s Siri, Microsoft MSFT -0.90 % ’s Cortana and Google Now all include efforts to build complete models of you from the data captured by your smartphone, and they’re making rapid progress. Like a personal assistant, they try to help you accomplish your daily tasks, either in response to your commands (Siri) or on their own (Google Now). But to do that, they need to understand you, and they’ll use any data they can to do so—from the smartphone’s sensors to your emails and calendar.
The Web pages you see every day are already the result of complex interactions among the models that content providers, advertising networks and advertisers are deriving. Learning algorithms trade against one another in the stock market. Last May, a Hong Kong venture fund named Deep Knowledge Ventures appointed an algorithm to its board, voting on investment decisions alongside the five human directors, according to Business Insider.
Today’s models don’t yet interact with us: You can’t tell them they’re wrong or ask them questions. Machine-learning algorithms are black boxes that only computer scientists can open up. But that will change as more of us realize how important machine learning is and demand a say in how it occurs.
Eventually, your model will be like your best friend, but with infinitely more patience. What will you ask it? You might not like some of its answers, but that would be all the more reason to ponder them. Your model—your digital half—might even help you become a better person.
—Dr. Domingos is a professor of computer science at the University of Washington and the author of “The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World” (Basic Books).
