I’ve had several special opportunities in the last few weeks to think a bit more about Artificial Intelligence (AI) and its future import for us remaining humans. Below I’m using my old-fashioned neurons to draw some non-obvious links.
The cause for reflection is the unexpected parallel between two events I’ve been involved in recently: (1) an interview of Elon Musk which I conducted for a conference in DC; and (2) the grand opening in London of a special art exhibit at the British Library which my wife and I are co-sponsoring. They each have an AI angle and I believe their small lessons demonstrate something intriguingly hopeful about a future of machine superintelligence…
Let’s take my experience as a snapshot case study, first with Elon Musk setting up the theory and hypothesis on the perils of “Strong AI” where artificial general intelligence could lead to Superintelligence. The “glorious future” in my title is of course an ironic reference to an idealized paradise of robotic perfection.
My 90-minute conversation with Elon onstage at the recent AFCEA Intelligence Symposium was wide-ranging and covered technology areas he’s currently leading work in – space, transportation, energy, innovation in general – but he wanted to lead off with Artificial Intelligence. He began by reiterating some of the arguments he and others like Stephen Hawking and Bill Gates have been making on the potential dangers in “summoning the demon” of Strong AI. (You can read “AI or Not AI?” for quick background, and I recommend Nick Bostrom’s recent book Superintelligence: Paths, Dangers, Strategies.)
I had a string of questions on the AI topic, focused on implications which he and others haven’t addressed yet, at least publicly. I won’t give Elon’s answers – the session was off the record – but they were thoughtful and quite compelling.
(Maybe you should register to attend the next AFCEA event, so you don’t miss out again.)
Here’s one of my questions, on an under-examined implication of the current AI debate:
Shepherd: “AI thinkers like Vernor Vinge talk about the likelihood of a “Soft takeoff” of superhuman intelligence, when we might not even notice and would simply be adapting along; vs a Hard takeoff, which would be a much more dramatic explosion – akin to the introduction of Humans into the animal kingdom. Arguably, watching for indicators of that type of takeoff (soft or especially hard) should be in the job-jar of the Intelligence Community. Your thoughts?”
If your appetite is whetted and you do want to tap the world’s greatest AI experts for the current state of their work, you can’t do better than looking at their own presentations at a secluded conference earlier this year in Puerto Rico hosted by the Future of Life Institute, a group which studies the challenges in ensuring the safety of AI systems, in an effort to counter dystopian developments. I found it difficult to dismiss the anxiety evident among several of these brilliant folks. UC-Berkeley’s Stuart Russell tweaked any skeptics who pooh-pooh AI worries, with this slide bearing a reminder from the early days of atomic research:
But let’s not get too AI-gloomy.
Instead, let’s turn to the parallel story from the arts, born at the very human Burning Man festival. I’m a fan of California artist David Normal, an innovative painter and installation-artist long active in Burning-Man circles. I like his work precisely because it demonstrates the incredibly complex, densely-layered inventiveness of a highly literate creator. One Los Angeles art critic’s review of an earlier San Francisco exhibition by Normal captures what caught my eye:
…the work is inspired by a set of influences as disparate and random as the content of the scenes themselves. The artist’s zoomorphic forms are inspired by Northern Renaissance masters, for instance, and his muscular figures and contorted poses are reminiscent of early Mannerism, as if Timothy Leary had come in and rearranged Michelangelo’s Last Judgment. There are numerous other archaic referents to be found in this curious puzzle–Normal’s Chemical Imbalance, for example, is composed around the form of the Kaballah, and contains quotes from Grünewald’s Isenheim Altarpiece, Bernini’s Ecstasy of St. Teresa, and an Escher lithograph. It is this dizzying and complex fusion that Normal calls Crazyology–taken from a Charlie Parker song, the term is more than just a title of the exhibit, it is intended as a description of the mishmash of influences in the artist’s work. Normal explains: “The list of art techniques and philosophies that exploited and exulted in the irrational is a long and distinguished history that I would sum up as Crazyology….When people ask me what my style is, I say, Crazyology, and that way I have my own term for my work, and I can also refer to all the great crazy stuff that has inspired me – Surrealism, Punk, Dada, Pop Art, Psychedelic Art, etc.”
As fans of that kind of bushy-dendrite complexity, this summer my wife Kathryn and I have joined the Eccles Centre for American Studies and the Burning Man Arts Foundation in funding a major exhibition at the British Library through November 2015, entitled Crossroads of Curiosity. (Here’s information if you’re planning a London visit, which I highly recommend, and here’s a link to more photos and videos of the exhibit opening with Burning Man founder Larry Harvey and British Library Chief Executive Roly Keating.)
At face value, the new installation is stunning but straightforward: four 8-foot-by-20-foot massive lightboxed murals incorporating rich imagery from the Library’s digitised collections, mounted on the BL’s grand Piazza. It is described in the catalogue as “a series of dramatic tableaux featuring provocative juxtapositions of vastly different times, places, and peoples.”
The pieces were first shown last fall at Burning Man 2014, where they were centrally arrayed on the ordinal compass points around the 105-foot-tall Man himself in the desert at Black Rock City (photo at right during their installation there).
But the story of how they came to be – their inspiration – is as interesting an element of the art as the visual images themselves. You see, there’s a man-plus-machine creation story…
A Mechanical Curator Lights Up the Desert
Several years ago the Digital Research Team and British Library Labs under the mighty Mahendra Mahey teamed up with Microsoft Research/Azure (while I was working there) to use their cloud infrastructure to work on a collection of 65,000 high-quality scanned books which were digitized through a partnership with Microsoft and the British Library a decade earlier. What to do with them? One idea from the BL Labs Technical Lead, Ben O’Steen was to set loose an algorithmic bot programmed to recognize pages which had an image or illustration, and digitally clip each one to save separately. The “Mechanical Curator” was born, and the resulting collection was eye-opening. Earlier the group at the BL had decided to do something generous: release the images for free public domain use. As spelled out in late 2013, their intent was to spark the creative imagination of others:
“We have released over a million images onto Flickr Commons for anyone to use, remix and repurpose. These images were taken from the pages of 17th, 18th and 19th century books digitised by Microsoft who then generously gifted the scanned images to us, allowing us to release them back into the Public Domain.
“The images themselves cover a startling mix of subjects: There are maps, geological diagrams, beautiful illustrations, comical satire, illuminated and decorative letters, colourful illustrations, landscapes, wall-paintings and so much more that even we are not aware of.
“Which brings me to the point of this release. We are looking for new, inventive ways to navigate, find and display these ‘unseen illustrations’. The images were plucked from the pages as part of the ‘Mechanical Curator’, a creation of the British Library Labs project. Each image is individually addressible, online, and Flickr provies an API to access it and the image’s associated description.
“We may know which book, volume and page an image was drawn from, but we know nothing about a given image.”
So the algorithms weren’t semantic, or “knowing.” They weren’t capable of characterizing the images, much less “understanding” them. The algorithm never raised an eyebrow, squinted its eye at an illustration and murmured “Now, that’s promising…”
In fact, it’s not surprising that the bot was assiduous but dumb. The state of computer-vision image-recognition is improving thanks to recursive deep-learning algorithms but not yet beyond childlike abilities. See “On Welsh Corgis, Computer Vision, and the Power of Deep Learning” on Microsoft’s AI research and “Welcome to the AI Conspiracy” on Google, Yahoo and Baidu. Adult-level semantic understanding is certainly not approachable today, even in the single focused domain of image recognition.
But the technical spadework was done with the BL collection, for others to build on. Ben O’Steen, the project’s technical lead at BL Labs, made all his code freely available on GitHub. for the million-plus JPEG2000 image files and associated OCR XML metadata [see here for technical info].
Most importantly, the collection was now available to the eyes of the global crowd, via Flickr. I met a student of information science volunteering work at BL Labs, Wendy Durham, who captured the back-story:
In December 2013, BL Labs and Digital Research Teams released 1 million algorithmically snipped images from 65,000 digitised books on to the Flickr Commons. Since then, the British Library Flickr Commons photostream has amassed a staggering 260 million views.
“Just a week after the release, one of those viewers was David Normal, a collage artist from California, interested in 19th century illustration for his work. Following a Facebook posting about the British Library Flickr Commons launch from the guitarist of the punk band ‘Flipper,’ David was thrilled to discover the serendipitous size, quality and relevance of the photostream content with his plans for a project he had organised with the annual arts festival ‘Burning Man’ in the Nevada desert, USA…
“After initially creating four 3 ft. by 8 ft. paintings, David transformed them into four 8 ft. by 20 ft., double sided and illuminated light boxes that were built around the base of the Burning Man statue, forming the centrepiece of the event. Over 70,000 people came to the event and saw the work…”
That telling leaves out an important step, though: What images to pluck from the photostream? And how to arrange them in a way that would create “art”? Here’s where the human element enters. Insert magic here, one might say.
The Spark of Human Creativity
Normal described that creative element in an important lecture in the British Library’s Chaucer Room last fall; I’ll excerpt the details which begin about 6 minutes into the video.
[While planning for 2014 Burning Man with BM director Larry Harvey] “he and I batted this back and forth and came up with the idea of a crossroads of curiosity, that would expand the idea of a cabinet of curiosities, out from just the rectilinear presentation of objects in a case, to be being sets of dramatic tableaux, that are a collection of human dramas and human phenomena.”
“So I was looking for 19th century illustrations, and it was only shortly after that that I learned of the British Library’s release into the Flickr Commons of over a million images. So that was really a kind of stupendous thing for me… Once I had that in hand, I was extremely inspired, because I had just an endless amount of collage material to work with to create these pieces.”
“Working with the British Library’s collection, I began to make the collages, and what I would do is just go searching through the database, just looking at one image after another, about the same way as if I was going to go through a book looking at every illustration in the book, and then I would mark the images that I liked. So it was a very entrancing process. I would sit there for hours, just clicking one image after another, sometimes going through a hundred or two hundred images without seeing anything that was of interest to me – and at other times being in kind of a vein, of things that I liked; and I would favorite all these out.
“I collected something like a total of 3,000 images. And then I would begin to select the ones that I liked best.
“People always ask me, ‘Why did you select the certain things that you liked?’ and a lot of times the things would just go together. I mean, I believe that the machine-gun and the skunk were just side by side with each other in the Flickr favorites page, so it was just sort of obvious, the work had already been done for me. But the Afghan warrior here, he was close by, and so he ends up – I decided to put him on, riding the skunk… the background ended up getting filtered out, and so in the process of making the collages I experimented quite a bit… I did so many different versions. Some elements didn’t end up in the final piece. For instance, I had a troubador with an elephant head…and while that might have been good, he just didn’t seem to fit in [Audience laughter]. So these became the raw material from which I painted, the collages actually became sketches.”
By the way, at about the 10-minute mark David goes into some technical discussion of the innovative lightbox materials and procedures he invented for this project and earlier lightbox exhibits; if you’re a fan of advanced electrical/fabrication techniques you’ll enjoy it. My wife and I own one of his earlier small lightbox prototypes “The Human Tree,” which hangs in our Music Room (left).
Like many artists since the dawn of Impressionism (and every -ism since), David Normal isn’t quite sure how to explain in words his train of creative thinking – because it’s not a logical procession. Again from his lecture:
“At Burning Man, people came around and were fascinated by the work, they spent a lot of time trying to decipher what the meaning of these strange images was. I took to giving what I called ‘Docent Tours,’ where I would walk people around from image to image, and explain the imagery and what it meant to me. The reason I call it a docent tour is because I actually don’t really have much better of an idea of what the imagery means than anybody else [audience laughter]. I mean, I made it, but it wasn’t like it was made with some sort of, you know, didactic purpose, that it’s supposed to explain something to somebody. Rather, the images are very much expresssions of [pauses] – oh, of something that’s just energetic – how shall we say – It’s not a specific – emotional… You know it’s not specifically confined to something that I can put into words. But I do put into words! And I’ll explain this to you soon, when we go and look at these prints I brought with me….”
Certainly the essence of the creative arts is ineffable, and it is impossible to capture or even describe fully the remarkable human inspiration and boundary-jumping leaps which trace the advance of human intellectual history. Some analysts today drive a debate on whether human-level AI will demonstrate the “same” human-like spark. (See for example “Creativity: The Last Human Stronghold?” by the thoughtful technologist and AI observer Israel Beniaminy on attempts to program machine creativity in poetry, art, math, and science.)
But it’s difficult to believe in a human-challenging degree of “intelligence” when the British Library’s Mechanical Curator never came up with the idea of putting the Afghan warrior riding on the skunk; the algorithm would never ponder a collage with an elephant-headed troubadour – much less realize that it “just didn’t fit in.”
The computers never suggested looking to Burning Man for inspiration, or turning paintings into lightboxes, or having a huge party to open the exhibit with Burners from across Europe, Taiko drummers, atonal chant singers and the London’s ultrahip DJ Yoda, all dancing on a Library Piazza bearing banners celebrating the 800th Anniversary of the Magna Carta housed inside, symbolizing man’s insatiable determination to win and protect his liberties.
To make the connection of my two themes explicit: Elon Musk came up with the unconventional idea behind his startup SolarCity while driving to the desert for Burning Man in 2004. His inspiration wasn’t a self-optimizing algorithmic feedback calculation but instead the very human notions of radical self-expression and radical self-reliance reinforced at Burning Man. And even earlier, in 2002 no recursively learning business algorithm running (then or now) would EVER have recommended that a businessman should create a scrappy rocket startup SpaceX to take on Lockheed, Boeing, and the challenge of Mars settlement.
In short, no programmer knows how to encode an AI with audacity. I doubt we ever will. Remember those old Apple advertisements with their call to “Think Different”? That’s humanity’s saving grace. We alone are able to create new and independent meaning from the complexity of the world (and “big data”) around us.
Will Superintelligence Supersede Humans?
I follow in detail the latest AI research and advances, and speaking as someone who has written about “autonomous killer robots” for years, I’m glad that the debate is reaching the mainstream, and can recommend reading several recent pieces:
- “Benefits and Risks of AI” by Tom Dietterich, President of the American Association for Artificial Intelligence (AAAI) and Eric Horvitz, former AAAI President (and a friend from Microsoft Research)
- This month’s excellent re/code piece “Deep-Learning AI Is Taking Over Tech. What Is It?“
- “Autonomous Weapons: an Open Letter from AI & Robotics Researchers” released today at the International Joint Conference on Artificial Intelligence in Buenos Aires.
- An excellent “Conversation with Jaron Lanier: The Myth of AI” (VR pioneer and another former Microsoft colleague)
My bottom line is the same feeling I have about a world with nuclear weapons; it’s not that I am skeptical of the dangers – it’s just that I remain optimistic of our ability to avoid or overcome the dangers.
I’m less interested in what the machines will do on their own, because I see evidence across history – ancient and continuing – of humanity’s ability to invent and then control by further invention. We’ve done it with the most awesome and fearful technology imaginable – the power of atomic and nuclear fission. Our ability to invent new technical, political, and social means of dominating our own Golems should stand us in good stead even with AI.
Call it insight, or the ineffable human spark, whatever it is we have it – and there’s enduring power in thinking the unexpected. David Normal saw a serendipitous art in siting his exhibit where it is: as he explains, “It’s actually placed literally (no pun intended) over 5 stories of books that are housed underneath the piazza – likely the very books from which the collage material was derived.”
What computer would get that joke?