in conversation with Ben Vickers (Part 1)

Artists + Machine Intelligence is a program at Google that brings artists and engineers together to create art with machine intelligence. AMI provides technical and financial support to artists and hosts interdisciplinary conferences. Our goals are: to facilitate a rigorous conversation around Machine Intelligence, to open Google’s research to new ways of thinking and working, and to support an emerging form of artistic collaboration between artists, engineers, and intelligent systems.

BEN VICKERS   Can you give an overview of what AMI at Google is, and the sort of work it has been responsible for to date?

KENRIC McDOWELL   The Artists + Machine Intelligence program brings artists into Google Research to collaborate with AI (Artificial Intelligence) researchers and engineers. We support artists interested in exploring AI with funding, technical advice and collaborative partners from our network of technologists, philosophers, designers, neuroscientists, etc.

BV   For broader context could you talk a little about the role of ‘Research’ at Google?

KMD   For the larger context, I’ll pull from Google’s white paper Google’s Hybrid Approach to Research:
“The goal of research at Google is to bring significant, practical benefits to our users, and to do so rapidly, within a few years at most. Research happens throughout Google, exploring technical innovations whose implementation is risky, and may well fail.”
As for AMI specifically, we are part of a larger research group of ~250 people within Google Research. Our group is unique at Google, in that we have an end-to-end design team working ‘embedded’ in Research. Our organization works with teams like Pixel and Photos to ship new features, but a lot of what we work on is 2-5 years out. Before AMI, I did speculative design prototyping and storytelling.

BV    The AMI website poses the question: “what are the emerging relationships between humans and machines? What does it mean to be human? And what can we learn about intelligence, human or otherwise, through art?!” I’m curious to understand the answers you’ve encountered to this question and what you have learnt so far?

KMD    For better or worse, our collective understandings of AI-human relationships come from film and television, where effective storytelling requires conflict between protagonists and antagonists. Artificial superintelligence makes a great villain.
So we have to unlearn these narratives to see what we are working with. There are certainly material, social, and political layers to machine learning (ML) as a phenomenon, but mathematically speaking, the neural nets that run machine learning are highly multidimensional, probabilistic representations of knowledge.
When our representation of knowledge is multidimensional, we see multidimensionality everywhere. I started to experience this after being immersed in AI art. Constructs like gender or interpersonal relationships began to take on a non-binary, non-spectral, multidimensional quality when I started thinking through the sensory experiences I was having of neural net space.
This is what artists do well; they move us through new spaces using our senses and imagination, enabling us to collectively develop new language from these experiences.

BV    What is the origin story of AMI and what role did DeepDream play?

KMD   An engineer in our group named Alex Mordvinstev was carrying out experiments to visualize the internal processes of ML systems when he discovered the process of “hallucinating” neural nets. When the first DeepDream image (colloquially known as “trippy squirrel”) was leaked on Reddit in the summer of 2015 generative AI went viral, at least within a specific underground community. We launched the program with a DeepDream show at Gray Area in San Francisco and moved into collaborations with individual artists.

BV    AMI echoes initiatives such as EAT (Experiments in Art and Technology) in the US at Bell Labs and the Artist Placement Group initiated by John Latham and Barbara Steveni in the UK. How do you relate to these moments in art history and what are the most valuable lessons to take from them?

KMD    These programs started with the assumption that there were two cultures and ways of thinking, creative and technical. While it was important to reconcile that perceived conflict of disciplines, our century demands complexity, hybridity, and holism in thought.
I carried out interviews with Bell Labs alumni like A. Michael Noll, Kenneth C. Knowlton, and Laurie Spiegel. While they weren’t in the EAT camp, one lesson I took from those conversations, and from EAT, is that the value of an art and technology experiment might not be immediately perceived, and that we need to trust the creative process to produce meaning, especially when it traverses contexts.

BV    Given the focus specifically on AI, could you breakdown an understanding of where the ‘state of the art’ currently rests in the development of artificial intelligence?

KMD Google Research calls AI the overall field of “making things smart,” which can include rule-based techniques sometimes called GOFAI or Good Old-Fashioned AI, whereas MI and ML refer to specific techniques used to learn from data. Superintelligence comes from the idea that machines that can independently access data and teach themselves to learn can theoretically surpass human intelligence generally (ML is currently superhuman in very specific tasks, like playing the game of Go, or diagnosing certain diseases.)

BV    What are examples of some of the projects you have published to date?

KMD    In 2017, we’ve completed three projects with artists (and are working on two more).
We participated in Ross Goodwin’s Wordcar road trip, where a text-generating neural net connected to a surveillance camera mounted on a car wrote dadaistic descriptions of everything it saw on a drive from NYC to New Orleans. Ross has collected 200,000 words from the trip and is working on an art book for publication. In this piece the car became a nexus of machine perception tied to the canon of American road trip literature. The output slices through complex terrain, touching infrastructure, surveillance, food distribution, language, and the American myth of the road.
We did a project with Refik Anadol and Mike Tyka called Archive Dreaming, an immersive ML visualization of the 1.7 million item archive of the SALT Museum in Istanbul. The experience is visually and architecturally impressive, but what’s most exciting about it is the way it reframes the archive as a multidimensional space of images connected by features, that is to say, visual relatedness. The piece also uses the ML model to hallucinate new images that could exist in the archive but don’t. To my knowledge this is the first time a museum archive has become generative in this way.
We also funded the production of a catalogue for the Harold Cohen retrospective at UCSD. Cohen is a seminal GOFAI artist and an important link between the programs you mentioned and our effort with AI.
I can’t say much now, but we’ve still got a couple collaborations in process, and some exciting explorations into architecture and urbanism that will inform our next year of work.

BV    Since Bell Labs there’s been something of a tension in the relationship between artists and the role of multinational corporations, the AMI program seems particularly sensitive to this but I wondered if you could speak about this issue and how you navigate it.

KMD    This is difficult to navigate in part because of the extreme power differential. The way we overcome this is to find partners within Google who genuinely want and need external critique to make their work better. Not everyone wants this but there are enough people that do, and we help them enter the conversation with an openness that empowers artists.
This is perhaps a more rarified way of gathering what User Experience (UX) designers call user feedback. In this case, however, it’s upstream of the product, because it happens in the research phase, and it’s done in the space of artistic creativity, which is much less bounded than traditional UX.
We also value the integration of ML technologies into artistic practice for its own sake, outside of the benefits to research. A well-rounded research program will take into account as many uses of a technology as possible, including applications for art and creative expression.

BV    The text Art in the age of Machine Intelligence published on the AMI website states “the issues raised by MI aren’t mere ‘theory’ to be endlessly rehearsed by critics and journalists. We need to make decisions, personally and societally.” What kind of decisions is this statement referring to?

KMD    Neural nets literally compress epistemologies into numbers; they shouldn’t be made without investigating foundational assumptions about society, meaning, even the nature of mind.
Take, for example, a neural net designed to predict crime rates (this already exists). The way crime and criminality are defined in training can reinforce or transform systemic bias. Tech work now must expand to include ways of thinking that are not strictly mathematical or programmatic.

BV    And how can art play a role in informing those decisions?

KMD    Critique is a way. In a more generative sense, artists move through problem spaces uniquely. Art is free to provide qualitative value, and in doing so it produces insights that couldn’t occur in a strictly quantitative development process.

BV    AMI is specifically trained on the role of machine intelligence in relation to artistic production—but machine intelligence as a technology is not coming about in a vacuum of accelerated innovation, instead we find ourselves in a moment of extreme flux, in which many new technologies are being realized in parallel and others are now reaching maturity, namely the likes of: blockchain, robotics and drones, virtual reality and augmented reality, synthetic biology, and other biotechnologies such as CRISPR/Cas9. Whilst there is a logic to making historical parallels between machine intelligence and the printing press, this combination of technology and the hybrid relationships that are likely to form present a situation that seems somewhat unreadable and perhaps entirely unknowable. Given your position within this, how do you navigate?

KMD    The situation you’ve described demands mental plasticity, openness to complexity, comfort with ambiguity. People that embrace cultural hybridity, people that unpack and reconstruct their own ideology, and people that think systematically, holistically and without prejudice, have an advantage here.
It’s a heady time, so it’s important to have a clear vision of your relationship with Earth. There are tools to help with this, although they may be de-legitimized under the Black Iron Prison.

BV    In 2015 Yuval Harari gave a talk at Google called Techno-Religions and Silicon Prophets—where he laid out the history of the great religions of our time, Christianity, Islam, Liberalism and Socialism. In it he succinctly identifies that with the advent of AI and in particular black-box neural nets we’re seeing a return to scripture and oracles for the decisions we make in the world. I’m curious to have your take on this idea, and whether you think Harari is correct in his analysis?

KMD    One point that Harari reiterates many times is that an ideology need not be true to take over the world, so he may be correct whether or not the premise makes sense. However, AI is still being developed, and artists’ contributions are already helping to make AI more perceptible, so my hope is that we have a deeper relationship than that promoted by authoritarian religions.
For whatever it’s worth, our sense of isolation and entitlement are unsustainable, and we need new modes of being that strengthen collectivity and interdependence. We need AI expression that reflects our multidimensionality and co-creative capacity back to us.

BV    Given the increased interdisciplinary nature of cultural production and the need for building bridges between worlds and filter bubbles, what are your predictions for how these areas will connect over the next century?

KMD    I can easily imagine a global patchwork of divergent experiments on diverse cultural and technological stacks, in a feedback loop of increasing novelty. But frankly, the undeniable ecological realities that will emerge from the Earth to restructure our civilization are most pressing to me. I think we’ll be very lucky if in 2050 we’re still arguing about fake news on Twitter.

in conversation with

(Part 1)

All images: Refik Anadol, Archive Dreaming, 2017 Courtesy: the artist

KENRIC McDOWELL has worked at the intersection of culture and technology for twenty years. His resumes include work for R/GA, Nike, Focus Features, HTC Innovation and Google. He currently leads the Artists + Machine Intelligence program at Google Research, facilitating collaboration between Google AI researchers, artists, and cultural institutions. He has spoken about art and interdisciplinary collaboration at MacArthur Foundation, Serpentine Gallery, Eyebeam, UCLA IDEAS, Nabi Art Center, and the Google Arts & Culture Lab in Paris. He received his MFA from the International Center of Photography-Bard in New York City.

BEN VICKERS is a curator, writer, explorer, technologist, and Luddite. He is CTO at the Serpentine Galleries in London and an initiator of the open-source monastic order unMonastery.