Accidents, the particulars, the details, the things that are all different but all affect all the time. I am much happier with that metaphor been with the people in power and the restof us metaphors. We have to share a microphone. So rosenberg in his book is at the end that Machine Learning does operate alike like a rake. It tests a whole bunch of possibilities and settleson one. Four factors we dont even necessarily know, we cant explain it, but we do that. In line with that i would love if you would tell the crowd about chicken flexors. Chicken sectors are an example used by philosophers of knowledge. This is all real. So its important to chicken producers to be able to tell the sex of a chicken because generally they want the layers also known as a female and the males are just fed and just expense so theywant to get rid of the males immediately. So there are chicken sectors who can look at a chicken and tell you what sex is. They can be really fast and incredibly accurate in doing this and they cannot tell you what the observable differences are between amale and female chicken. They just cant. We still dont know and so if you want to become a chicken sex or, just so you can say or have a card printed up, then you get trained by an existing chicken sex or and ive never said the word chicken sexer so many times in my life where you pick up a chicken and you say male . No, next one. Mail . Next one and you gas and over time you get this power, youre able to tell what sex it is without being able to tell how. This isvery , its a good example for philosophers of knowledge who are committed generally, traditionally to the idea that knowledge is something you can explain and justify because otherwise its just a guess. When you start its just a guess. When you know you should be able to point out why or say look at these wingtips are pointy but we dont know. And Machine Learning, this is one of the cases where Machine Learning works this way. Indeed, our brains operate without explanation, with the chicken sexer and Machine Learning are similar and our brains operate like that. We dont know why it works, but it works. So theres lots of examples of this but you can do a regular scan, so it to a Machine Learning system thats been trained in a particular way, a particular application and it will tell you things about the person who owns theeye. What gender, i think some age stuff and a whole bunch of heart Health Related things. One of the cholesterol, i dont remember which and the like. The thing thats interesting about it is that neither the Computer Scientist nor human doctors can tell whats in the scan thecomputer is paying attention to. Maybe theyll figure it out but as of now, they cant tell and it may just be some odd combination of a set of factors that we are really not put together. Maybe jackson sexer can train somebody, might be an interesting experiment but Machine Learning can work this way. It works without giving us any general rule or thing to look forward to, pointy wingtips or anything like that. As a vegetarian you do not want to watch videos because the male chicks are thrown onto a Conveyor Belt and ground up alive. Enjoy. Thanks for that, chicken fingers come next. So i want to talk for a minute about the Regulatory Regime thats starting up in the world. We must be able to control it, we cant control the universe but we can control that. Indeed when i started working with computers, im no developer either that anything we have to do with a computer was, you could always find out what was wrong. There was always a reason what was wrong and you can always fix it. You usually do that by blaming microsoft for something but you fixed it. Thats not the way the world operates now but now you have regulators coming in saying we must have accountability and theres a lot being written in congress to look for algorithmic accountability. To try to make the algorithms accountable. One thing you hear often is be transparent, show me the algorithm. No one would know what the hell to do with the source code of an algorithm. Its gibberish, it changes all the time, it depends on the data and another says open up the algorithm to know its impact. We might know how to do part of that but we dont really know how to do that. But yet our lives are more and more going to be served by algorithms that predict reliability what youre going to want next and theyre going to do a good job of it. And they may have bias as you mentioned before. They may have other problems. Do you see a way that authorities either cultural or governmental will be able to regulate and manage this algorithmic Machine Learning ai world or is it beyond them . I am worried about this as well. There are certainly areas where we want regulation. Not to mention there is very strong evidence, convincing evidence some of these algorithms have amplified systemic prejudice. Unacceptable. There are areas where we want regulation. Autonomous vehicles, self driving cars powered by Machine Learning, we need somebody other than the Car Manufacturers to decide what he systems should be lets it optimized for. What are their goals . One goal is fewer fatalities. You could end the list, dozens of things it could aim for, so to speak, including lower environmental impact, shorter travel times, comfort counts. We dont want, i dont want the Car Manufacturers to be in charge whether anybody looking over the shoulder about which of those. We are allowed to say as a culture, youre a big, rich car company, luxury car into going for comfort. If that turns out is going to optimize it for that means youll get worse mileage, we are killing the planet, we have a right to say no. There are tradeoffs among all these different values. It makes it very complex. To say what we need is algorithmic transparency, im not even sure what that means. Its really not what you need and its going to vary from case to case. In the case of autonomous vehicles, a lot of it could be achieved. You could regulate what we want them optimize for and insist on transparency of the metrics, of the outcomes. If the cars are continuing to kill a lot of people, thats an important metric. If Energy Savings are not going down, and to do what we do with products which is we send them back. We do a recall and the manufacturer is required to get it back up to snuff. We dont require manufacturers to tell us what the process is. Slicing and prove we fixed it, now its doing what is required to do, we dont need to insist on transparency. Rather than algorithms, we will need data transparency. But even that is complex. I dont think regulators are capable of analyzing data. Thats a pretty specific skill set, a deep skill set. Maybe the transparency for that may include, on the samples diverse . Where did the samples come from . Are you representing the community and the people affected by . Whats the vetting process and what are the outcomes . Theres different places where you could insist on transparency and some solutions you dont need transparency. We need to know regular get it right. It will vary by Aircraft Safety pretty well despite, okay fine, but we do it really pretty well. Its going to vary by product and its going to be a nightmare of struggles because the impetus will always be among regulators to do the simple thing and the thing that crushes everything, so long as they get the outcome they need, and we will lose some benefits that we need from that. Its a scary time. I have two more questions and then i will come up for any trepidations or joys or fears that you may have. By the way, i was at google i o, the Developers Conference last week with a bragged most the time about Machine Learning. What struck me most was dictating Machine Learning models that exist that were 100 gigabytes and gotten down to half a a gigabyte. What that means is entire model fits in your phone. Even an inexpensive phone. They showed heartwarming video of a woman in india who is a mother who is illiterate who needed her kids to read to her. She could point the phone at the sign. It would both read the signer translated and read it out loud because entire model existed within this small little bit of minerals, whatever its made out of. And so i think well see more and more and by the way, that reduces fear of privacy violation. Because the old model all your data went to the cloud. Now can happen locally and all the conclusions go to the cloud. Theres a lot of developing happening. You will see more and more and more which is why you want to buy the book. Its going to be in everything. Intel inside would become Machine Learning in sight. Lets go to the internet and ask us and them, and have a discussion. So the net i think of as a connection machine. Thats the relationship of Machine Learning and the net as they both enable or use or export connections. Connections among dated in Machine Learning, connections among people and information in the net. I get into argument all the time with people who tell me that the net is a medium. Like the one he worked in, like what is surrounding us. I argue, no, that its something new. And i get into this fight all the time. The reason it matters because i think we dont want to bring the same presumptions and regulations to the net we had in another world. Its also a power relation, call it a medium and jet set up, assume to set power relationships which dont necessarily exist on the net. Part of what ive been thinking is that if you do a chart of how many people, and government looks at it, is a big circle that says media, inside his broadcast, print and internet. I think thats wrong. Instead all drop a bigger circle and called the internet inside that is a little circle called media. Inside there is one called medications. As of the circles outside that are being drawn in by newtons gravity. Finances, maybe 15 inside. Retail, maybe 25 inside. Crime, 10 inside. And so on and so on. So im coming to see that the internet describes society. In this way. In my field in journalism we keep covering the internet as if it is a technology story. And then come to see its not. Its a story of people, a story of net is our behavior on it. Everything would talk about, its not that technology screwed it up, its that we screwed ourselves. Maybe it amplified. Maybe it made it easier. Maybe it made it worse. I dont know, but the internet is made of people. Im starting to wonder about i need to, i was talking to a communications school, and i thought thats kind of retro. Whole school around media. Why isnt this a school around the internet . How are we studying the internet as society, as the future . Not predicting. The dumbest job title i can think of is futurist. But understanding the contrasts of where were going what would it mean in your view because you basically done that through your books, small pieces loosely joined was about how we have connections that are not as clear as it used to. Everything is miscellaneous. Similar about knowledge, too big to know was our brains are too small for all of this. Now in this when youre looking at how the machines are bringing their own order to this that we dont have. As you look, what you done really is, since the manifesto, trying to understand disney world that is ruled a great measure by internet. How should we be studying it . Hash it would be understanding it . What should we be doing with this in universities to make this more clear . I dont know how disciplines get started historically. I i know its complex and i dont know how you do it, what the forces are that actually get universities to commit. I dont know. I agree that the internet is uniquely important, has not turned up the way some of us thought initially. I attribute a lot of that with my own blindness to privilege. Optimistically, oh, what im doing on the internet, which is a positive and light and goodness, anybody who encounters this will be liberated to find more information to make connections which is crucial to internet. That turned out not to be. Thats an assumption privilege people make about life. It is, for me it support for e same reason, many of the same ways and reasons Machine Learning is, that the architecture of the internet gets repeated and made visible to us, even when were not hang attention. Its a architecture of an ability that enables you to go anywhere you want and, on the net, and contribute your own stuff and make new things. Its at generative connective architecture that at its best enables us to do more of that. At its worst, it does the same thing except what results are all of the evils, the passing around a false information, the bullying and the gathering together of clusters of bullies. Thats also inherent in the internet architecture, but that architecture is one that enables and thus gets us the value connection when way or another, even when we do it negatively. I think that same repetition of architecture, having some defining impact on how we think about the world, is now about, it is already being repeated in the world, thanks to Machine Learning. I take that as being my arc. I didnt think i had an arc but i think probably i do, and thats what almost all my books have lots of flaky pieces on the cover, loosely joined. We will share one microphone. Repeat the question. Im going to come out and we will share one microphone here. Does this wired one work . I see, dont go out. Repeat the question. You will yell loud enough for my old theaters and i will repeated for the recording old ears. Any questions or desires or fears . Yes, please. That gentleman wants to know about chicken sexing, because we all do. Is that a human being that does that or is that a machine . No, its a human being. Machines cant do because the oldstyle machine, you have to do what to look for. We dont know what to tell what to look for. Its the human occupation. It really wellpaying because for some reason a lot of people dont want to do it. It involves gently squeezing a chicken and, a check, and newborn check and peering at insides. It may not be an occupation that is glamorous, but it is pretty wellpaying and is very mysterious. [inaudible] well, you know. If you compare it to awaken other things, people got trained during world war ii as those of the worst to recognize in aircraft, the type at a glance, and a lot of it was a distance. You cant do it, but the way the country was they were told i may not get the example right but a zero has rounded wings and thats how you can tell from of the things that looks like it except the other one is tilt up a little bit, just like identifying birders, when identifying a bird flying pastorally quickly. You can program a computer. You can tell a tell a computers what to look for. If you dont know what to look for, you cant tell the computer. This is a field where Machine Learning might, one assumes that Machine Learning trained on enough instances would be able to do chicken sexing. As far as i know theres no research being done on this. Probably hard to get a grant for, is what im thinking. How do they know they are right . So how did you know you never know if those really a female. Yes. How did he know since its a male, they discarded, never know. This is not the right answer but there are competitions. People race to see how many, how many they can do and how many they can do correctly. The experts get new 1 as checked by other people. Of course you know when you let a mail through and they grow and they are not laying eggs. You know you got that wrong. As far as we know there are astoundingly accurate at enormous speeds. [inaudible] repeat the question in an evolutionary process of the going to lose the ability to make the source the decisions we outsourced to the machines. So in terms of evolution and genetics, i mean, im not a geneticist. I suspect not because evolution works exactly that way. Nevertheless, its an important question and one that is been asked for a dress for a long, long time. Socrates i think argues against reading, against literacy. So this is fifth century. On the grounds that first of all you cant have dialogue with something written, and is all about dialogue, so that come why would you want to just read one side of the story . And he says we will lose our memories. Theres no need for us to exercise and user memories because we write everything down, and that absolutely has happened. The odyssey, the ill get is i think 16,000 lines and it was passed along, before writing, generation to generation. And so there were people who could sit down and recite the iliad for you. If anybody here and recite the 16,000 lines of the iliad, unwilling to listen, but no, weve lost that ability. However, and to socrates thought, is that they could really stupid. But he was wrong about that. I dont think any of us would give up literacy on the ground that the literacy makes us smarter. Because we have greater memory because we write things down. We accumulate knowledge because we write things down. Its a huge win, even though we lost that skill. Its not clear, i dont know if the source of decisions that Machine Learning systems and make, thats a skill that we need and we will become, in this case, socrates will be right. Well just become stupid, incompetent and unable to operate in the world. So i cant make a prediction. Ill tell you my little experience of this, very quicy is, i have terrible sense of direction, always have. I was born that way. I get lost all the time. I now have and rely upon, making universal sign of the cell phone, you know, the routing software gets me everywhere but actually makes me worse at directions because a get somewhere and i know cognizant math at all. I know told me make the next right. Thats the extent of my sense of space, his voice saying turn left in 400 feet. I am pretty sure as made my already terrible sense of direction worse, but i dont care because me and my machine are smarter than i was. That is always been the case. This is the philosopher, andy clark, scottish philosopher, makes a basic point, he makes very complex points as well but a basic but which is we think and work with machines. We think that weve been taught to think it has very lonely life to be thinking your head. But he says weve always thought in the world with tools, whether its the old gr