^B00:00:01

>> Adam Kushner: Hi I'm Adam Kushner. I run the Sunday Outlook section at the Washington Post, which is our home for ideas and essays and arguments and most importantly today book criticism. I want to thank the Library of Congress for putting this festival together. We are extraordinarily lucky to be joined today by Thomas Malone, the other of Super Minds, The Surprising Power of People and Computers Thinking Together. Tom is the founding director of MIT's Center for Collective Intelligence. He trained as a mathematician and social psychologist. His whole career is devoted to the question of how humans can take best advantage of technology in an era of self-driving cars and home devices, like Amazon Alexa. Computers aren't just becoming smarter; we're also making them smarter and they are making us smarter. Super Minds is about how computers and people perhaps slowly over the coming decades are going to advance to amplify each other, touching almost everything we think and care about from climate change to grocery shopping to democracy. Tom is going to take questions after his speech and he's going to be signing downstairs in lane two after the talk, so please join me in welcoming Thomas Malone.

^M00:01:07
[ Applause ]
^M00:01:13
^M00:01:15

>> Thomas Malone: So good afternoon. I'd like to start by giving you the two main messages of my book. The first message is that I think we spend way too much time thinking about people or computers. And not nearly enough time thinking about people and computers. Too much time for instance thinking about how many jobs computers are going to take away from people. And not nearly enough time thinking about what people and computers can do together that could never be done before; so that's the first message. The second message includes that first one, but goes much further and deeper. The second message is really about a new way of looking at the world. In a sense this second message is about how to see ghosts. Now, I don't mean real ghosts if there even are such things. But I do mean powerful entities all around us, all the time that are mostly invisible unless we know how to look. These ghosts are super minds, which I define as groups of individuals acting together in ways that seem intelligent. Now by this definition super minds are all around us. For instance, every hierarchical company or other organization that you've ever heard of is an example of a super mind. A group of individuals acting collectively in ways that at least sometimes seem intelligent. Also every democracy is a kind of super mind, whether it's in the government or a club or some other organization. Again, not guaranteed to always be intelligent. But sometimes these democracies can be very intelligent. Another very important kind of super mind are the markets that we use for all kinds of goods and services. And another very important kind of super mind less visible than the others are communities. Whether those are global scientific communities like the ones shown here or local neighborhoods, or many other kinds of groups. Now one of the first things you realize when you look at the world this way. Is that these super minds run our world? Almost everything we humans have ever accomplished from investing writing to making the turkey sandwiches I have for lunch almost every day. Almost all of those things were not done by single individuals acting all alone. They were done by super minds. By groups of people working together often over time and space. Now like individual people, some of these super minds are really smart. Some super minds are pretty stupid. Sometimes we really like the things super minds do. Some super minds can be evil. But whether we like them or not, we can't accomplish much of anything without somehow working with or influencing these super minds. Now another thing we see when we look at the world through this perspective of super minds is that computers can help make super minds smarter. To take one of many possible examples think about Wikipedia where thousands of people and computers all over the world created a new kind of super mind. And that super mind created the largest encyclopedia our planet has ever known. So that's an example of computers creating a new and newly intelligent kind of super mind. Of course it's not guaranteed that computers will always make super minds smarter. For instance when fake news influences voters in a democracy that can make the democracy less smart. But if we use them wisely, I think we can use these computers to create human computer super minds that are smarter than anything we've ever known. And that's what I want to focus on this afternoon. One way of framing the core question here is this. How can people and computers be connected, so that collectively they act more intelligently than any person or group or computer has ever done before? Now that's a big, hard question but one thing that can help us answer that question is understanding the difference between two kinds of intelligence. The first is specialized intelligence, the ability to achieve specific goals in a given environment. The second is general intelligence. The ability to achieve a wide range of goals in a wide range of environments. Now here's something a lot of people don't know. Even the most advanced computers in the world today have only specialized intelligence. Take the IBM Watson program that beat the best human players of Jeopardy a few years ago. I know from the person who led the development of that software that that program couldn't even play tic tac toe, much less chess. It was very specialized to the particular task of playing jeopardy. At the same time, any normal human has far more general intelligence than the most advanced computers in the world. Even a five year old child can carry on a sensible conversation about a much wider range of topics than today's most advanced AI programs. Now that's today's state of the art, an obvious question is how soon will that change? When will we have human level artificial intelligence? If you do a poll asking people that question today, the average answer you'd probably get is about 20 years from now we would have human level AI. Some of you might be very excited by that possibility. Some might be scared. But here is something else a lot of people don't know. People have been asking that question "When will we have human level AI" ever since the beginning of the field of artificial intelligence in the 1950's. And for that entire time the average prediction of when we'd have human level AI has always been about 20 years in the future, for the last 60 years. So is it theoretically possible that this time the prediction could be right? Yes, theoretically possible. But I think we should be very skeptical of anyone who confidently claims that we'll have human level AI in the next few decades. In my own view, we're very likely to have that someday, but that's likely to be many, many decades in the future. And part of what that means is that in the meantime all uses of computers will have to involve people. One way people often talk about that today is to say we need to have humans in the loop. When they say that, they're often thinking about one person, one computer. There's nothing necessarily wrong with that way of thinking about things, but I think a much more useful way of thinking is to think about to start with the human groups that have accomplished almost everything we humans have ever done.

^M00:10:04

And to add computers to those groups. Then we can use the specialized intelligence of the computers to do the things they do much better than people. Like arithmetic and certain kinds of pattern recognition. And we can use the general intelligence of people to do everything else. Perhaps even more importantly we can use computers to create hyper connectivity, connecting people to other people at a scale and in rich new ways that were never possible before. Now one way of summarizing what I've just been saying is to say that we need to move from thinking about humans in the loop to computers in the group. That's my slogan for you today. But how can we do that? How can we create more intelligent human computer groups? I think useful way of thinking about this is to think about the cognitive processes that any intelligent entity needs whether that's a person or a computer or a group. You can think of those in terms of five key processes. First to act intelligently you have to decide what action to take. To do that, you usually need to create some possibilities for actions you might take. You can usually do both of those things better if you can sense the world around you and if you can remember the past. And if you're really smart, you can learn from your own experience to do all of those things better and better over time. So one way of using this framework is to think about how groups can do these different kinds of cognitive processes. But think about that, let's start with the decide process. And think about different ways groups can make decisions. I think one of the most important parts of my book is a categorization of five fundamental types of super minds for making group decisions. The first and most obviously is hierarchies, where group decisions are made by delegating them to specific individuals in the group. Another possibility is democracies where group decisions are made by voting. Another kind of super mind is markets where the group decisions are really a combination of all the parawise agreements between individual buyers and sellers. And in communities the group decisions are made by a kind of informal consensus based on shared norms and reputations. Now all four of those kinds of super minds require at least some cooperation among the group members. But if you don't have any cooperation among the group members you have the fifth kind of super mind, which I call ecosystems. And an ecosystem the group decisions are made by the law of the jungle, whoever has the most power gets what they want. And the survival of the fittest. Now one way of using this framework of different kinds of super minds is to use it for analyzing and inventing new ways of doing all kinds of things we do in the world. For instance to take just one example, which is in the news a lot recently. If you think that we need to change the ways that women are treated in the work place, you could use this a check list to think about possibilities. The most obvious way perhaps is to use democratic decision making to pass laws that are then enforced by hierarchical governments about what kind of behavior is legal. But another way, not quite so obvious would be to use communities. For instance the Me Too Movement is trying to change our shared community norms about what kind of behavior is appropriate, whether that's legal or illegal it's a community norm that can be changed. So this is a way of thinking about different ways to do things, and I think there are lots of possibilities for using this framework to do that. What I want to focus on for the rest of our time today is how computers can create or help us create new kinds of these different varieties of super minds to make the ones that already exist much smarter. Let's start with an example of a decision that's typically done by a kind of hierarchy but that could benefit from involving more community aspects as well. The decision I want to focus on is medical diagnosis and I want to talk about a project called the human diagnosis project. This is a system that lets medical conditions, doctors, nurses and others get advice from other clinicians about their difficult cases. They can enter information into the system about the patient's symptoms, the lab test results, etc. and then they can ask for opinions from other people anywhere in the world connected to this system. For instance a doctor at Mass General Hospital in Boston might get second opinions from doctors at Stanford, or more interestingly a nurse in a remote African village hundreds of miles away from the nearest doctor could use this system to get advice from doctors or others anywhere in the world. Now one of the things they find when they use this system is that by getting multiple opinions in this way they get more accurate diagnoses and medical - better clinical care. In a sense that's a result of using the system to create more hyper connectivity, connecting medical clinicians all over the world in a way that they can use their community to help make diagnostic decisions. Another thing they find when they use the system is that they gradually accumulate a knowledge base of cases. And they've applied artificial intelligence machine learning to the knowledge base they already have, and found that the programs learning from that knowledge base can already for common symptoms like chest pain, they can already give diagnoses that are almost, not quite, but almost as accurate as a human physician. What I think is most interesting about this is to imagine what will happen when case or systems like this have seen millions of cases. Far more than any individual doctor could ever see in a whole lifetime. I think then these systems will likely be able to do very accurate diagnoses of many kinds of symptoms and diseases. In fact, it's likely they'll even recognize new diseases that we humans had never even noticed before. So those are some potential benefits of using computers to make medical super minds smarter by involving many more people. Another example I'd like to talk about is how to create a new kind of democracy that's not really feasible without computers. The example I want to talk about is something called liquid democracies. Now you already know about direct democracies like in ancient Greece where all the voters can vote directly on all the questions. You also know about representative democracies like we use in this and many other countries to elect representatives, which then vote on all the important questions on our behalf. Liquid democracies are a kind of combination of those two other kinds of democracies that have the potential to give us the best of both worlds. In a liquid democracy you can always vote directly whenever you want to on a given question. But most of us don't begin to have time or inclination to do that for all the things that need to be decided in a democracy. So you can also in a liquid democracy delegate your proxy for voting to anyone else you want to. You might give your proxy for voting on one category of decisions like military decisions to one person; you might give your proxy for voting on environmental issues to somebody else. And those people can in turn delegate your voting proxy to still other people who for instance may know more about the details of a particular decision that needs to be made than the people who originally had your proxy would know. And if at any time you feel like the people who have your voting proxies aren't doing what you want, you can always take back your proxy and vote directly yourself, or give your proxy to someone else.

^M00:20:07

Now his isn't just a theoretical possibility. There have been a number of political parties in different countries around the world who've already been using this, and running on platforms that say if our representatives are elected they'll vote as the liquid democracy in an online system tells them to vote. And these parties have already had some success in Europe and Iceland and several other places around the world. Google has also used a liquid democracy like this to make some simple business decisions. And I think the key point here is that this is a new kind of democracy made possible by computers. It wouldn't be feasible without them. But it has the potential to be much more responsible to what voters actually want and to take advantage about specialized knowledge about particular problems. So that's a way of making smarter democratic super minds. Another possibility is to use markets to not just allocate resources but to predict the future. I want to talk about something called prediction markets where they're kind of like gambling, but they're different in some important ways. In a prediction market you buy and sell predictions about possible future events. For instance, if you think that your company is going to sell somewhere between 1,500 and 1,600 units of some product in the month of September, you could buy shares of that prediction. And if at the end of September the prediction turns out to be right you get $1.00 for each share. If it turns out to be wrong you get nothing. Now it turns out that the prices in these prediction markets are estimates of the probability of the future event in the collective judgment of all the people participating in the market. And it also turns out that when people have experimented with these prediction markets they turn out to be almost always at least as accurate and often even more accurate than any other prediction methods, like opinion polling or focus groups or whatever. And they've been used successfully for predicting things like product sales and movie box office receipts and election results and so forth. For instance there's some public prediction markets on the web and here is one of them showing the probability predicted for Brexit occurring in the UK by November 1. As you can see they're predicting the probability is about 57% now, that it will happen by then. And you can also look at the price history at the bottom of the screen to see that that probability has gone significantly up since late July when Boris Johnson became the Prime Minister. Here's another of those prediction markets of who is likely to win the next US Presidential election. You can see it shows Donald Trump 42% probability, Elizabeth Warren 23 and so forth. It's important to realize that these are not how many people say they're going to vote for that person. This is people's best estimates of the probability that that person will win whether they like it or not. And it's very interesting to be able to see that change over time as new information becomes available. Now in these prediction markets we've just seen I think it's very likely that the - the participants in the market were all people. But what if we could also have computers participating in these markets? That's a question that a student of mine and I began to investigate several years ago. We wanted to try to make predictions of things that were analogous to what would a competitor and business do or what would an enemy do in war time. But we wanted simpler example than that. So we tried to predict whether the next play in a football game would be a run or a pass. We showed people videos of a football game. We stopped the video just before each play began and we let people participate in a prediction market buying and selling predictions of whether the next play would be run or pass. We also did that with computer bots, who had only some simple information about what yard line the ball was on and how many yards to the next first down. And we let those bots make the predictions and then buy and sell those shares with other computer bots. And then most interestingly we had some prediction markets where both people and computers participated in the same prediction markets. So a person wouldn't ever know whether the most recent trade was made by another person or by a computer. What we found was that the prediction markets that included both people and computers were more accurate and more robust to various kinds of errors than either people alone or computers alone. So I think that's an example of how markets can provide an interesting way of combining human and computer intelligence. Now some of you may be worried about your jobs. What - are computers going to take away my job? Should we worry about this? I think we should all relax a lot about that. I think it's theoretically possible that that might happen. But I think and it's almost certain that some jobs will go away. But every time this has happened in the past, more new jobs have been created than were destroyed by computers. This has been happening ever since the Luddites in England in the 1800's. And I think it's likely to happen again. Here for instance is what happened with US employment over the last couple of hundred years. In 1800 over 90% of people in the United States were employed in agriculture in some way. Today it's less than 2%. But the difference was more than made up for by increases in manufacturing and increasingly in services. In fact, population has grown significantly during that period, so way more new jobs were created in those other industries. And while I think we should worry about individual people, whose jobs are destroyed and who can't find new jobs. I think we should worry about what to do for those individual people. I think it's very unlikely that there will be long term massive unemployment for most people in the economy. It's never happened in the past when people have worried about. I think it's unlikely to happen again. Here are a couple of examples to illustrate what I think is very likely to happen and will cause that same thing that's happened before to happen again. Let's start with an example of the technology called the printing press. Starting in about the 1400's the printing press was able to make very rapid, cheap copies of things that had previously been done by scribes copying laboriously by hand all day long. So those scribes essentially had their jobs eliminated. But think of what happened next. Since it was possible to make very cheap copies of so many things so easily, all kinds of new things became economically feasible to make and sell. We had not only copies of a few important books, but we had newspapers and magazines and novels and comic books. And think of the jobs that were created to do all those things. Not just printing press operators, but novelists and newspaper reporters and editors, and book store owners and newspaper delivers and on and on and on. So I think the same kind of thing will happen again. Here is another more recent example, think about the job of a reference librarian who helps patrons who come into a library, use the reference materials that are there in the library. I don't know the exact numbers for people employed in that job now. But I think that a new technology like google search has certainly decreased the rate at which people with that job are likely to increase. But think of the new jobs that have been created for every reference librarian job that may have been eliminated. We now have not only software developers and database analysts, but website developers and search engine optimization specialists and online advertising sales people and on and on and on. Now it's not always easy to predict what the new jobs created will be. But in my book I give a lot of example of new kinds of human computer super minds that are likely in many cases to create lots of new jobs. For instance I talk about how we could create new super minds for helping deal with global climate change involving people from all over the world. I talk about how we may be able to do a much better job of predicting terrorism by in part, involving lots more people in pieces of that. I talk about how we may be able to do a better job of developing corporate strategic plans that are much - done much more rapidly and much more innovatively and much more comprehensively.

^M00:30:10

Partly also by involving more people and also computers. And I talk about how these super minds can help deal with job loss caused by automation. Now in the book I also talk about some pretty philosophical things about super minds. For instance, I talk about whether super minds can be conscious. There's a long section about whether Apple Incorporated, the whole company itself a kind of super mind is a section about whether that super mind is conscious. I won't give you all the details today but it turns out that Apple is aware of its environment, it's aware of itself. Its goal oriented, it integrates many kinds of information. And if you're willing to exercise a little empathy I think it's quite possible that Apple as a company experiences feelings of certain types. So I think it's not crazy to think that Apple and many other companies and many other groups, in fact do have a kind of consciousness. So let me leave you with one last thought about where I think we're headed. I think in the long run as our world becomes more and more closely connected by many kinds of electronic communication, it will become increasingly useful to think of all the people and computers on our planet as part of a single global super mind. And perhaps our future as a species will depend on how well we're able to use our global super mind to make choices that are not just smart, but also wise. Thank you very much.

^M00:32:05
[ Applause ]
^M00:32:13

>> Adam Kushner: Thank you, I think we have time for some questions. I think people are supposed to come to the microphones, is that right? Okay if you have a question, please come to the mic. 

>> Yes thank you. That was fascinating. The idea that institutions or organizations are minds, and in some sense conscious, it also implies - it implies not only that they make decisions on your - on your neural processes slide, but also that they have goals. And I guess the question is as these organizations get more and more efficient - I mean sort of what you're saying I think is you're talking about processes that are not just since computers, but really go back to the beginning of at least capitalism and before that.

>> Thomas Malone: Absolutely, at least as far back as humans go and probably even further if you want to think about bacteria and so forth having collective intelligence. 

>> So how - we can get more efficient given the goals that we have. But, -- but how is it that these - I mean it's a common place since at least the 19th century that - that we work - we work for the machines, the machines don't work for us. I mean how - how is it that we get these organizations to have goals that are - that are actually goals that help humans rather than helping the organization - the super minds pursuing their own goals?

>> Thomas Malone: I talk about that question in the book. And even if there are no machines involved you could say, you could say that a hierarchical organization there's a classic sociology paper about unions, labor unions and how over time the labor unions - the leaders of the labor unions who have full time jobs doing that, come to more and more look out for the - the benefit of the union itself and their own jobs rather than the people they are supposed to be representing. There's nothing unique about labor unions, that same thing happens in businesses and many other organizations. So I think that's a risk, whether there are machines involved or not. I - I'm not sure I would agree with your premise that we're always already working for the machines, but in some cases it certainly may feel that way. I think that essentially what we need to do is think about how we can influence the goals of the super minds that are important. Sometimes we can do that in obvious and easy ways, sometimes it's not at all obvious or easy but that's what we should be thinking about. And I do, in the book talk about one perhaps cheerful possibility which I call the law of evolutionary utilitarianism. I won't go through the detail of how this all happens, but it turns out that if you believe that bigger organizations or bigger groups often have more power than smaller ones. And a way of attracting people to join a group is to do something that's good for them, that they like and want. Then in the long run in general, it's likely that super minds will more and more fulfil the desires of more and more people. Not always, not immediately but that's at least one reason for long run optimism about whether or super minds will actually fulfil our human desires are not. That make sense? Okay, next question? 

>> Hi I was wondering if you could share some of your thoughts about the future of human computer interfaces, because I wonder if at some point in the future the bottleneck is [Inaudible] intelligence in general, but the ability of humans and computers to understand each other. 

>> Thomas Malone: Ability of computers what?

>> To understand humans and vice versa?

>> Thomas Malone: Yes, well that's already true to some degree. Some computers already guess what you may be thinking when you get auto complete and typing a text message. The computer is trying to guess what you're thinking and often does a reasonable job, though certainly not always. So I think increasingly computers will get better and better at having accurate models of the humans that they're dealing with and humans also will get better at having accurate models of the computers they're dealing with. You ask I think about the long term of that. I don't talk much about this in the book though I do mention it a little bit. I think it's pretty obvious that where we're headed in the long term is more and more neural connections. That is electrical connections to our brains that connect us to computers. I think it's quite possible we'll have fairly advanced forms of that even before we have artificial general intelligence. So it's not going to happen immediately by any means but I think more and more that will happen. 

>> Thank you.

>> Thomas Malone: Thank you. Question over there, sorry.

>> Hi, I'm a physician and once we had implementation of electronic health records, physicians were very hopeful that the gathering of the data that came from this massive community of doctors would help us make better diagnosis and so forth. But the reality is at this point in time, all the data is being gathered and nothing is coming out of it, perhaps because it is no collecting - collective effort gathered the right data. And just gathering data for the sake of gathering data does not translate into outcomes. So is there any concerted effort, in the way that you presented that case to create that rather than just - because we have all sorts of data currently with the electronic health records that we have. 

>> Thomas Malone: That's a very good point, just gathering data by itself doesn't guarantee anything good will happen. There are often other things you need to do to have that data be useful. And I think what you're describing is what has happened all too often in the healthcare system that we gather these medical records and then we make electronic medical records but we don't really think very hard about how we could use them, and so the data turns out to not be very useful. One thing that I think was interesting about the example I described was that they gathered it in a very particular way where they wanted to get help from other clinicians and that led things to be structured in a way that they were able to learn useful lessons from using the computers as well as other people. But you're absolutely right, it's not guaranteed and it may take some time before that actually happens. Let's see maybe I do one more over here, because I had several over there.

>> So I'm an elementary school teacher, so this spoke to me on that level because you know part of a super mind in the classroom. And the big buzz word is community, classroom community and of course now at the computers it's global community. So my wondering is because we talked about the different decision making. Do you think that it's more such logical question, but do you think that humans have a natural inclination towards that decision making model but we just use the others because of the fact that it's more difficult to get consensus when you have those large groups of people, and as it becomes more globalized, we go back to that community thing or -

>> Thomas Malone: Yeah, it's a very interesting question. I think it is true that we humans have a natural inclination to use communities as super minds, because that's how we grew up. Hunting and gathering tribes were communities, so we all have a genetic understanding of how to be part of a community. But over time we humans have invented other ways of working together. In fact, you could even say that hierarchies and democracies and markets are a kind of artificial intelligence. They're a kind of artificial super mind invented by we humans that - and they've turned out to be surprisingly powerful and useful in many situations. They're not - no one is always best. In fact, there's another whole chapter in my book devoted to analyzing the relative advantages and disadvantages of these different types of super minds. So I think that kind of gets to your question. Okay another question here. 

>> Like the main character in 1984, I can see the role of consulting super minds. But I [Inaudible] hand over individual or collective responsibility to anything not human. But my specific question is you mention that Apple has a consciousness and we should have empathy to understand that. My question is does Apple - is Apple capable of empathy? Because we all know they're unfortunately some human beings even restricted access to empathy because of [Inaudible] based on theory of mind and all of that.

>> Thomas Malone: I heard two questions in what you said. The first one was should we ever delegate decisions to a super mind or should there always be humans making them? Unfortunately we've been delegating decisions to super minds for many millennia. Whenever a human tribe or a human democracy or a market, or a hierarchy, a government, an army or anywhere. Any time a group of people makes a group decision like that the individual humans have in some sense delegated the decision making power to a super mind that they are part of, and in many of those cases may include only other humans, no computers. But we've been doing that for a long time. The question about Apple and consciousness. When I talked about empathy, what I meant was not whether Apple has empathy, though I think it may perhaps. What I meant was can we, as humans have empathy enough to attribute feelings to a group of humans. In fact, as an individual human I don't really know what's going on inside your mind. I assume if you act in certain ways that you're angry or afraid or whatever. But I don't actually know that, it could just be a robot pretending all those things. So it requires a certain amount of empathy on the part of me to understand another human - human's emotions and I think the same thing is necessary if you choose to try to exercise it, to understand the emotions in a super mind.

>> I'll be interested to see how this plays out because I understand there's something in the structure of the healthy human mind that has that marrying capacity for empathy and can that be inserted into computers.

>> Thomas Malone: Okay unfortunately I think we're out of time. I see the wrap it up sign, so I'm sorry that the others of you won't get to ask your questions publically but I'd be glad to talk to you afterwards and anyone else. Thank you very much.

^M00:43:19
[ Applause ]
^E00:43:22