Science literature but what im most interested in is, number one, how to help policymakers to make the decisions about Technology Policy and so that involves understanding the details of our Computer Systems and Network Systems work. And probably more importantly im really interested in how to give all our Computing Technology so that its more responsive to Public Policy needs. Ive been able to do that because ive worked with and continue to work with an Extraordinary Group of computer scientists and other researchers at mit and around the world. Host can you give one example of how you influenced policy . Guest the people are probably well aware as an example that theres a big debate about the use of Encryption Technology. We need encryption to protect our email exchanges, our financial transactions, political speech that goes on, we hope privately online when it needs to. Governments for decades have been concerned that the use of strong Encryption Technology would swarm the ability for Law Enforcement to conduct electronic surveillance. From a technical perspective we looked at that question, and while acknowledging encryption can post some barriers for the police, we has shown through Technical Research that if you try to force all encryption systems to be palpable or have backdoors, then youll end up harming the security of all the billions of people around the world to use the internet and continue in Computing Technology. While this is an ongoing debate even the governments that are trying to control encryption have acknowledged that they shouldnt do so in a way that introduces systemic weaknesses. Thats want everywhere think think without a real effect. Host to up us explore some of those other areas, some of the policy initiatives that are involved with technology today, kyle daly who is the Technology Editor with axios is joining us. Thank you peter. Thank you, professor. I want to stick on encryption for for a moment because as you alluded to there is an appetite in washington. We have seen from bill barr. Weve seen from some on the hill to impose backdoors on companies. Are you concerned about that . Guest im certainly concerned that governments make decisions that really get the balance right, that really look at all the interests on all sides of this debate and make sure to make the right decisions. I certainly understand the frustration that Law Enforcement has. They are in many ways digitally behind, but thats a problem that goes far beyond just Encryption Technology. They certainly need help any better training and a better equipment to do investigations in the Digital World and i think we should be providing Law Enforcement the help to be able to function in this environment. The problem is that trying to regulate encryption is kind of a quick fix. It might feel good but not really going out because the concerted criminal activities was going to find ways to hide their communication one what are the other. This leads all the rest of us anymore vulnerable state. So i am concerned that policymakers really should look at the whole picture when they are making this choice. You mentioned policymakers are always playing catchup, always a little behind. Do you see any remedy for that . Guest hopefully, the program i teach at mit is part of the remedy for that where we are constantly trying to understand how to educate our Computer Science students so they are more aware of Public Policy needs. We also spend a lot of time working with policymakers all around the world to try to understand the challenges they face to help them to be smart about how to approach them, and help were possible to make sure the systems we are building meet the needs of society has. Yes, i think we have treated a lot of technology as a kind of fixed quantity, that is, theres some sort of absolute wall between the Public Policy world and the technical world and would very much want to bring that wall down, both because we think we can do a better job of designing systems and because we think the policymakers can do a better job of making policy if they are well informed. What does that look like from washington side . You guys can better educate Computer Science students on policy, but how do you educate policymakers on Computer Science . Guest ill give you an example. Were doing a lot of research right now on election security. A group of my students have looked very carefully at some of the mobile voting apps that are out there, some of the internet online voting at that are out there. What weve learned is in some cases the apps that some election jurisdictions are choosing have very significant flaws, and weve been glad to see that dhs as an example has worked very hard to try to bring information about those vulnerabilities to election jurisdictions all around the country in looking at some of the Internet Voting Services where there are kind of different choices about how the systems are designed, whether they are used to enable people to submit ballots electronically or perhaps just get copies of ballots that they can then mail in on paper. Weve been able to show that one is a pretty safe approach, that is the electronic ballot delivery with mail in return, and one is a really dangerous approach where there could be hundreds of thousands or millions of ballots sitting on unprotected servers. Sometimes the policy world misses the nuance in a certain way. What our research has shown is there are techniques that can be used that could be productive, i can expand access to the ballot for people who need it and want it and may face barriers. But there are some things that just shouldnt be done. As long as we Pay Attention to the details, Technical Details and policy details, we can make progress on these things. Theres been a lot of talk about a number of tech policy issues, whether its election security, privacy, ai, facial recognition and were not seeing a lot of firm action. Do you think theres lack of leadership in washington on some of these issues . Guest so face recognition is a great example. A lot of concerned about face recognition actually arose from research that was done by another student of ours at mit who found that the widely used as Recognition Systems from some of our leading Technology Companies are dramatically less accurate if youre a person of color or woman. The result of this has been a really deep investigation into just what it takes to build more accurate Face Recognition Technology. There are parts of the government that it worked really hard to try to support that. The National Institute of standards and technology is doing more and more testing in this area about how to tell when you have a good face recognition system and when jeff one that is not good but i think there is more need for policy leadership in this area. We are going to be relying on all kinds of Artificial Intelligence based technology for really critical decisions for everything, whether you get arrested or not to whether you are going to get a load or not or whether you will be hired for a job or not. The fact is right now as we saw with face recognition, we lack the ability to make solid Technical Assessment of whether the systems are Accurate Enough for the purpose of their using. There are always going to be companies that could have different technologies, try this or that, this is fine for low stakes usage before highstakes usage when peoples liberty or lives or economic livelihood is on the line, i think the government does have two step in and set some standards and make it clear if companies are deploying technologies that are substandard, at minimum they should not be able to sell it and to be on that beyond that they should probably be financially responsible for the harm. Host professor weitzner, can you explain how in laymans terms how facial recognition is developed and how it recognizes faces and how it is being used today . Guest sure. Face recognition uses a range of Artificial Intelligence called Machine Learning. The way that Machine Learning works generally, so its a technique to try to get machines to learn things, for example, to try to match names to faces. And in a nutshell the way that you teach the machine to learn how to recognize faces is you give it lots of images of faces and associated names, at the computer looks for patterns and all of those images. It may look at millions of images to try to figure out how to recognize one from the other. In some cases face recognition is about matching one phase to another. If you want use your face as the key to enter building, for example, that kind of Face Recognition Technology will try to figure out whether the video image, for example, that its ease of peter is like the one thats on file. If its a question of putting names to faces, then it would do a different kind of technique, but either way its about Teaching Computers to find patterns in very large amount of data and then recognizing that battered pattern to get in some new data. But the key to face recognition or any other kind of Machine Learning technology is that the pattern recognition is only as good as the data, whats called the Training Data that is initially presented to the computer. For example, if you try to train a computer to do face recognition and all the images of the people with darker skin are poorly lit and dont have adequate contrast, then the system is not going to learn how to recognize people with darker skin as well as people with lighter skin. This is a problem, this is really an engineering problem which is a solvable problem if enough effort is put into it. But because these technologies are developed very often in a commercial context, companies are going to expand as much effort as a feel the need to expand to sell the product but no more. Part of what we have to do is make sure with the right kind of standards in place and the right kind of responsibility in place in case the system doesnt work properly. Host but just to go back to what kyle said earlier, theres a real issue about privacy here, isnt there, and a potential abuse . Guest absolutely. Theres no question that in order to make Machine Learning techniques work well, work accurately, they need a lot of data. In a lot of cases thats personal data. There are different approaches you can take to that problem. You can try to secure the data to make sure even the one company perhaps gets together a lot of data for training purposes, its protected data to make sure no one could get into that data. Its pretty hard to build perfectly secure systems. There are also techniques that are being developed in my lab at mit and by people all around the world for what are called private learning techniques so that you can do the same kind of Machine Learning training that we talked about but do it in ia way that the data actually stays with the owner of the data and that the computation remains private. So the computer can learn the patterns that needs to learn in order to recognize whatever it is, the face or a credit risk or the incidence of cancer perhaps in an xray or anything else while at the same time preserving the privacy of that information by allowing it to stay in control of the person who has it originally. Academia has been out front on a lot of these issues you are alluding to, the way that facial Recognition Systems perform more poorly on people of color. We are starting to kind of seed industry catch up. A number of Major Tech Companies have imposed moratorium on use of their facial Recognition Technology by Police Departments at the very least. Do you see a reckoning going on . Do you see more sincere awareness in Silicon Valley of how products may be exacerbating existing inequality . Guest i think society generally is taking a more critical look at a lot of the technology that is being put in front of us, that is being offered for use. If you look at the way the internet developed initially commercially in the 1990s, it had a lot of excitement about it. Some people called it the wild west. It was very enticing new technology and there was a real spirit of experimentation then which was great. It led to a lot of innovation and a lot of advances, and i think i am very happy with the expanded access to information that it created for all of us in society. But i think theres a recognition also that maybe some of the concerns about privacy were not attended to as carefully as they ought to have been. So now and what almost feels like a second wave of data intensive technology, that is Artificial IntelligenceMachine Learning, i think everyone is looking more critically and carefully partly because the companies that are developing a lot of these technologies went from the proverbial couple of inventors in a garage to enormous Global Entities that have strong and in some cases, it positions in markets. We are appropriate look at these things will carefully. We also recognize our lives depend on them a lot more urgently, where in doing this technology with the power to drive us around, with the power to make very important decisions about our lives. So yes, at a think we are right in the middle of that process of trying to figure out what it means to hold Companies Accountable for protecting peoples privacy, for building the products and Services According to a certain set of standards. Its a kind of legal detail, but for a long time and to this day software and relief really werg the software here exclusively pretty much, software isnt subject to the normal Product Liability rules we think about for things like automobiles or consumer appliances or whatever else. If your car has an accident when it shouldnt or explodes or catches fire, the car company is responsible for a lot of money. That is not the case for software and Internet Service providers. We are i think in the process of trying to forget what kind of responsibility should we put on these companies when we are really going to depend on the products and services for our lives. You mentioned our lives do depend on technology more than ever right now were in the middle of a global pandemic. What are maybe some areas where you see, whats something you are hopeful about that technology could solve as we try to fight covid . Guest im spending a lot of time myself working in an area called digital Contact Tracing. Your viewers, many of them, probably know that apple and google in april announced a global offering, a a number of technologies would make it possible for people to determine if they been in close proximity with someone who has been identified as infected with covid19 virus. The design for those systems came out of some work that we did in my lab and labs around the world, in switzerland and germany and elsewhere. This all happened very, very quickly in response to the pandemic. We were all looking for what we could do to help and came to this view that we could actually use the cell phones that are in our pockets, in our hands, for many, many people around the world to help with the epidemiological process, with the Public Health process of making sure people who have been exposed to this Infectious Disease take appropriate measures. So its really interesting because the Technology Developed super quickly. We put out a the design in abot three weeks, and a week later apple and google said okay, we are going to do that. And a month later it was already deployed in both of their mobile operating systems. We are now in a more complicated process where Public Health authorities all around the world a kind of figure how to use that. We are working very close with the number of them to try to make this work. We moved from what is a narrow technical question, that is, how to get phones to recognize when you are in proximity with each other, to how to integrate a system like this into a very complicated Public Health process known as Contact Tracing, people talking about a lot. I am hopeful that this will in the end be able to make the contribution to the building of Public Health authorities to help contain the disease in different places around the world. Theres a lot of complexity to it and there still a lot we have to learn, partly technically but probably more for me Health Policy perspective how to do