Last Saturday, I had a great chat on Skype with Samim and Roelof - two very cool guys who are planning a sort of machine learning revolution. I met them on Twitter, where I came across their "Ethical Machines" podcast. This is a remarkable podcast that I highly recommend to anybody who is interested in AI, computers or the future of society. Unlike the regular commentary on mass media, whose portrayal of AI is often a one-dimensional caricature of the question "How far are we from the Skynet in Terminator ?", this podcast gives a nuanced understanding of everything about AI. This is because Samim and Roelof are both active programmers and researchers, who follow and shape the most recent trends in AI. But this is not just a technical podcast. It is also about the ethics and politics of AI, as well as about art and culture.
I will give a small profile of Samim and Roelof. They created Gitxiv, which is a nice mashup of Arxiv and Github - two places on the internet where scientific researchers post preprints of scientific articles and the source-code of the projects. Gitxiv is a place of combining these two and more. As compared to various scientific communities, the machine learning community is remarkably open and sharing, with even big companies making the source-code of their projects available for the general public. Gitxiv is a place to bring all these energies together. They are soon planning to extend this venue to share data sets - an important element for reproducing scientific results today. I feel Samim is an artist at heart, he is interested in AI because it opens a new frontier in creative exploration. He is also a sharp thinker and entrepreneur. He is based in Berlin, which is probably the most interesting place in the world for a person with this combination of qualities. Roelof is a Ph.D student in Sweden, working in the area of natural language processing. But he is also a political activist with a strong understanding of social activism. In my past, I have briefly participated in activism about free software, though this is nothing to write home about. But I understand how incredibly enriching it is to do the basic ground work in social activism and to engage with people. So Roelof has some great perspectives out of his experiences.
Together, Samim and Roelof are a great team and they both share a passion in democratizing AI technology to the masses. I ran into them through making some snarky comments on Twitter, with my typical pessimism about the future of AI. Especially over the past one year, I grew very critical. I wrote many critical articles in this blog about the progress of AI and about the increasing despondency of our society's future with it. This is rather depressing, as I consider myself to be an AI researcher at my heart. There is a classic trope in PhD-Comics about the "bitter post-doc" - I probably fit this bill very well. When I speak with younger researchers and students, I have to consciously work on what I say so as to not depress them completely. But my pessimism about AI has less to do with technology, and more to do with the single-minded "Skynet" narrative that our society is building for it.
Speaking with Samim and Roelof literally lifted my spirits up, at least for a brief while. Samim is a very optimistic guy who sees the silver lining under any cloud. Right in the beginning of our chat, he pointed out that we need more narratives on how we tell the AI story. He pointed out that there are already several positive ones - the ecological perspective, the global consciousness perspective, the young entrepreneurship perspective. Even with my bitter pessimism, I couldn't deny that ! Our chat then became about how we can build on these narratives. We remembered the glorious days of the personal computer revolution in the 1960s and 70s - in some way, we are all fans of the great pioneers like Doug Engelbart, Alan Kay and so on. At the same time, we are conscious of the great consolidation going on in the market, where huge data clusters are condensing under the rule of monopolies. I talked about the consolidation going on in the visual effects and creative industry, damping some of Samim's hopes that this might be a way out. But Samim is too optimistic to ponder on my negatives. Roelof added a very interesting point - about how consolidation is going on in the sphere of academia and universities. He pointed out the creation of the University of Amsterdam - merging two historically separate universities (religious and secular) into one, and then the creation of commercial research centers within the premises of the university, where students have to sign an NDA before stepping in. This is shocking news to me (well not really, what did I expect !?), but we bemoaned how the Netherlands - which historically had a liberal tradition, that saved European culture from death during the aftermath of the printing press revolution - is no longer as much a defender of the free culture as it used to be.
We then talked about how, or if it is even possible, to replicate the personal computer revolution in the sphere of AI and machine learning. What we need desperately today is a "Homebrew Computer Club" - the rag tag band of losers, programmers and nerds, which in the 1970s took on the grand big monopolies of Xerox and IBM. I mentioned that this has to be a popular movement, not limited to the elite set of programmers and researchers, but inclusive of all sections of the society: the young kids at school, old people, and especially, artists and the creators of culture. We need to get those guys and girls who define the "quintessence of cool". AI and machine learning is something everybody should claim ownership on. But how ?
A few weeks ago, I visited the Paris Machine Learning Meetup - hosted by the brilliant blogger Igor Carron and his co-conspirator Frank Bardol. I talked about virtual faces and Leonardo da Vinci - as random a mix up of ideas as you can imagine. I have been thinking of going to this meetup for a long time, mostly because I liked its logo, which I show below.
This logo talks about giving power to the data, but it is really using a revolutionary image of "giving power to the people". This reminded me of the stories from the early days of computing, where Ted Nelson published his legendary book "Computer Lib / Dream Machines", which also uses the same iconography.
This crazy book used to be the bible for the Homebrew Computer Club. I haven't seen this book, but finding and owning a copy of this is one of my life's missions. This book kickstarted the idea of democratizing computer technology to the masses, much before anybody has seen or heard of a personal computer. What we need today is a seller of such dreams about machine learning for the masses. This includes machine learning for your grandma, your dog, and your street artist. We cannot afford to bind machine learning in the prison of researchers, elite programmers and mathematicians. The only hope to save our society from descending into a totalitarian state is by democratizing AI. But how to do that ?
The meetups sprouting in all major cities of the world show a path forward. They widen the audience from researchers, engineers and managers of fat-wallated companies, to something broader. But there is a long way to go before we get your grandma, your dog and your street artist to get interested in machine learning. Alan Kay thought of computer programming as a medium, something as simple to use as a book. But we have not got there yet. Even bonafide computer programmers don't think of programming as a medium, they think of it as a skill to show off. In reality, programming should be as trivial to do as speaking a language: obviously requiring some training, but something that one can do without conscious effort. We have a long way to go till we get there.
Even though I grew up as a quintessential nerd and studied computer science in various universities, I never understood what computers were all about, until I came across this lecture by Michel Serres in 2007. The French research institution INRIA, where I was working as a doctoral student, was celebrating its 40 years and invited the philsopher Michel Serres to be the keynote speaker of the function. His talk was about a point that is so simple that it blew my mind away: computers are not tools for solving problems, but tools for solving people, who in turn will solve the problems. In other words, all the fantastic applications of computers and the internet are just a side-show to something much bigger: like tiny ripples of water on a tsunami. Very few people understand the real impacts of the computing revolution, because they need to imagine this from a perspective of a "changed brain" or a "new self", not from the present self. But there is a catch - before the computing revolution can catch on and make its true impact, it has to engage with the vast majority of the society. It should not be limited to just the elite few programmers and researchers.
It is not Michel Serres who first articulated this vision, but Marshall McLuhan. This greatly inspired early pioneers like Engelbart and Kay. An even early expanse of this vision is from Vannevar Bush, who wrote the essay "As we may think" at the end of the second world war. In physical terms, a computational way of thinking would rewire the human brain, expanding the higher cognitive functions, as well as those dealing with compassion and empathy. In other words, we will become a better species through the practice of computational thinking. I think the early dreams of the computing pioneers have largely failed. We stand today in a desolate moonscape of parched desert, where the vast majority of human population live in a prison of apps and trivial status updates - what Alan Kay once reminisced as Henry Thoreau talking about the implications of the transatlantic telephone cable, "that it will help the vast majority of Americans to know about the latest fashion statement of a European princeling". This pessimism about human nature aside, I think there is a fundamental reason why the early dreams of computing pioneers have failed - the lack of useful applications to engage the user to the full potential of computing. I think this is now beginning to change with machine learning.
In the early days of personal computing, Alan Kay and colleagues have made little children draw on the computer screens and play music, and used this as a basis for establishing the principles of computer programming. As a child drew a picture on the equivalent of a "MS Paint" like program, they romanticized that the child was "programming". By the way, "MS Paint" (And Apple Paint, or any of the other clone demos) is a trivialized corruption of the original ideas behind this demo at Xerox Parc, which indeed had educational value to teach many computing aspects. But despite the best efforts of the pioneers, they did not succeed in inculcating a knowledge (and love) of programming in the masses. Today, we might be in a much better position, because we can rely on large data sets, sensors and machine learning source-code to realize far more engaging applications.
So the time for a "Homebrew AI club" is ripe. A new culture of computing can start today, where every human can be an active participant. This can take full advantage of the connecting power of the internet. But what is stopping it ? I don't know. But it may be time to put off our conspiratorial hats and believe in the full potential of the human species. It is time to get rid of the fears of the NSA, the big brother state, the lousy social networks and see the bigger picture.
I don't know how and when this massive social change will happen. But after talking with Samim and Roelof, and generally brewing some thoughts in my puny head, I have a list of points to ponder on.
Ten commandments about AI (actually, just ten talking points) :
1) The word "homebrew" is brilliant. It is reminiscent of home-brew alcohol, which has a direct benefit on the human user, and which immediately alters the mental states, which is exactly what we need to aim for.
2) We need artists. Anything big like a social revolution will not happen due to a bunch of nerds talking about mathematical equations. We need big mojo people like Che Guvera or Steve Jobs (though less arrogance and ass-holish behavior would be nice). Heck, we need women. It is high time there are more women in computing. Most of all, we need to engage with people who have a creative spark.
3) All of us nerds need to start at home and explain technology to our families, girlfriends and boyfriends. The first thing to start explaining is probably data security and privacy. There are a monumental number of losers on the web who share their private data without even knowing about it. We should first subtract our friends and families from this group.
4) But privacy awareness and computing knowledge are not one and the same thing. There will never be a magic switch that will bring us to a privacy-respecting world. Living in the digital world will always be a battle against adversarial powers, which in the future, will only become more powerful and obscure though the use of data and machine learning. It is highly important that everybody knows how to keep track of their own data and use it for the better. But how do we train n00bs in this ?
5) An important obligation is with helping elderly people. Most elderly people are already clueless with technology - fiddling with the inner lives of TV remotes and email preferences is not for them. But everybody has a right to lead a dignified life online. Before we snark and snigger on the troubles of elderly people with technology, let's imagine how much more awful our own lives will be when we grow older. Technology will screw us million times over then (that is, if we are still alive then, and did not all disappear in the smoke of a nuclear explosion).
6) The first step about doing machine learning is collecting data. What better place to start than collecting data about oneself: one's own friend's circle, one's shopping habits, one's tax bills, one's entertainment preferences etc.. We need open-source software that helps people to collect data and organize it in a nice manner.
7) The second step is to train people to use machine learning: simple regression functions, then more complicated methods like deep-learning.. It is likely that not everybody will understand the maths behind it. But this is not as important as being able to use these methods in a regular and confident
manner -similar to how one uses home appliances everyday. I still remember the awful day when I, as a fresh graduate student in USA, put a tin-foiled sandwich in the microwave and saw it explode in fireworks. To make it all worse, this happened in front of a bunch of school kids, that I was supposed to teach about robotics ! In my entire life until then, as an engineering student in India, I have never used a microwave oven. If you grilled me about what would happen if you put metal in a microwave, I would have pondered over my physics knowledge and answered correctly that it would explode. But when I was just hungry and wanted to eat my sandwich, physics was the last thing on my head. So I completely failed as a functional user of home appliances in the USA, due to my complete lack of training. With respect to using machine learning, I think we need to train people to be first functional users of technology. Obviously, a few will cross over into learning the maths behind it. But even if they don't do that, it is still quite okay.
8) The third step is to expand this training to a full programming language. I think most people get it wrong with teaching programming languages. Most humans don't care about the Turing-completeness of a language or a programming paradigm. They just want to get shit done. But knowing about the basics of computability and information theory will be a must for anybody in the future. So we need to train them how to think computationally - what are the types of algorithms, how do we store data structures, how do we evaluate computational cost etc. We need ways to explain this in a simple manner to everybody. We can do this. After all, driving a car is not trivial. Putting a sandwich in a microwave is also not trivial (as I found out). But with a little bit of training, people do both these tasks quite well.
9) We need a way to swallow our nerd pride. It is not easy. We all have to work on it. When you are superior to your peers in some skill that is essential for anybody's survival, you can very often feel smug about your superiority. I don't know if it is humanly possible to not think otherwise. But being a smug monkey is not the point. Evolving from monkeys to humans is our goal.
10) We need to enlist doctors and the medical profession for our help. Of all the various disciplines, I think clinicians have the most respectful view of human potential. Every day, they see broken people in their clinics, but they try to fix them and raise their potential. In a way, great teachers are also like clinicians. They treat each student separately and help them realize their individual potentials. The medical profession is also very relevant because everybody is concerned about their own bodies and their medical choices. Often these choices are complex and require a fair amount of statistical and computational thinking. So we can develop computing paradigms that teach machine learning for people by using their own personal data for medical choices.
I would like to finish off this blog by talking about narratives, and about the stories we need to tell about AI. I will use two existing pop culture narratives - the movies "Star Wars" and "Lord of the Rings". I am a great fan of both these movies.
The "Return of the Jedi" from the Star Wars movies gives a very nice narrative about how a "primitive" tribe of Ewoks overcame the much greater power of imperial storm troopers. Sure, there were a few Jedi warriors who were helping them, but the Ewoks were the fundamental game changer in the battle, which is one of the most lovely aspects of the movie. In our battle over AI, we need to get the Ewoks - people who value friendship and nature more than technological gadgets.
The "Lord of the Rings" books offer another similar narrative (actually George Lucas was quite inspired by these books). The final battle with the dark lord Sauron is won through an alliance of elves, men, dwarves, as well as hobbits and ents. I particularly like how central the hobbits and the ents are to these battles, as these are tribes that are not technologically superior, but value friendships and nature immensely. Another nice reference is to the wizards (loosely analogous to the AI researchers), who are split between the forces of the light and the dark.
Well, these are just two narratives, not quite complete, but much better than the stupid Skynet narrative about AI. The danger of the Skynet narrative is that it is fatalistic - as if humanity is like a deer caught in the headlights, unable to do anything about AI.
This binary narratives about AI need to stop ! Humans are the agents of their own lives !
Robots are going to get your job.
Robots are going to make out with your boyfriend.
Robots are going to eat your babies.
Robots are going to wipe you clean.
No, Robots are going to make your breakfast.
No, Robots are going to make you immortal.
We also don't need help from a Jesus-like savior like in the Matrix movies (yes, hello, the title of my blog). But we need stories that help us believe in everyone.