My Story

Early years

One of my early childhood memories is of my Dad waking me up in the middle of the night (or what seemed like the middle of the night for a 7 year old!) because he wanted to show me something he was working on. He’d been creating a Printed Circuit Board (PCB for short) for a hobby electronics project. The creation process involved dipping the circuit board in a solution of Ferric Chloride - a corrosive chemical - which dissolved all the parts of the copper board that hadn’t been protected with etch-resistant ink. The result was a beautiful pattern, etched in copper, that “emerged” from the murky liquid. That memory stuck with me. I found my Dad’s interest in electronics and physics very inspirational. Though not formally trained in the sciences (he grew up working in Bolton’s weaving industry) he had a passionate interest in both the physical and life sciences. He bought me a kid’s microscope as a Christmas present one year. Another time I received a mini-greenhouse plant growing kit. And although my brother tended to be gifted with more “traditionally boy-themed” electronics and Meccano toys, I always shared those with him too.

Meanwhile, my Mum bought me craft kits, paint, cross-stitch sets, crayons, wool, and fancy coloring pencils, fostering an entirely different side of my personality - my artistic self. My Mum was much more artistic and spiritual than my highly practical, materialist, engineering father. And interestingly, my Mum also worked in the textiles industry - but she designed the artistic patterns and repeating motifs for fabrics that were then made real using the Jacquard looms that my father would operate and repair. The artist and the engineer.

The combination of those two worldviews is a thread that has been running through the rest of my life so far. I’ve felt these two threads have always been with me, always pulling at each other, each trying to be dominant. But instead of the two very different areas of interest creating a tangled mess, I’ve tried to weave them together into a unique tapestry of my own.

Physics and University

Whilst still in High School, I almost chose biology as my major for my last 2 years of study (known as A-levels in the UK system, roughly equivalent to a US-based AP). It was a very last minute decision to switch; I changed my mind after the deadline for student course submission. I remember having to organize an in-person meeting at my high school during summer vacation to ask for special permission to change my courses from biology to physics. I was a little sad, because I did love biology too, but two things had caused me to make the switch. The first was that I was thinking about biology as part of a track towards a medical degree. But I did some work experience in a hospital and I didn't really like it that much. A medical degree seemed to require a lot of training on the clinical side - and I was more interested in research. Secondly, I read a fascinating book by Alastair Rae, called Quantum Physics: Illusion or Reality. This book was my first introduction to the strange world of quantum mechanics.

Soon after choosing a physics path, I had a teacher who furthered my fascination with quantum theory. Mrs. McCann dedicated physics lessons to describing some of the interesting philosophical and paradoxical unknown scientific questions that arose from the pioneering work of Bohr, Schrodinger, Rutherford and Einstein in the 1920s. I think that was the first time in my life that I realized: “Wow, science doesn’t know everything!” And that meant there were things I could discover.

I then spent 8 years studying Physics and Electronics at The University of Birmingham, UK, first completing a Masters Degree (MSci) in Physics with Electronics and then a PhD in Quantum Physics. My Masters project and PhD focused on experiments on tiny devices called Josephson Junctions. JJs (as I affectionately referred to them) are sort of like the quantum equivalent of a transistor. They can switch state like a transistor, but also exhibit some really interesting quantum mechanical properties that I won’t go into here. You can read my thesis if you want to deep dive!

During that time I got to work with some super cool pieces of hardware equipment, with fancy names like the Plasma Reactive Ion Etching System and the Ion Beam Millatron. I worked in a couple of clean rooms - and yes I had to wear those funny clean-suits. I also wrote my name in tiny letters using a Scanning Electron Microscope with a Focused Ion Beam Milling add-on tool, which was fun. And I also hand-made several copper-track patterned PCBs to build custom electronics for all the measurements I had to do. Thanks Dad for showing me how to do that. In fact I was so into circuits that my friends made me a circuit themed birthday cake.

Quantum Computing

Working with Josephson junctions was cool, but I wanted to actually build something from them. Something large scale. Transistors can be put together in large numbers to form microprocessors and computer chips, and it turned out that in theory, you could put Josephson Junctions together in large numbers and make quantum computers! I was simply captivated by the idea of a quantum computer. It sounded so exotic, and I was inexplicably drawn to the idea.

I left the UK and went to work for D-Wave Systems in Burnaby, British Columbia.

I began as an experimental physicist, testing and calibrating the control circuitry that allowed quantum processors to run algorithms. I did that for about a year but my love of design and communication also meant that I’d created a lot of material that described quantum computing and quantum algorithms quite well. So I moved into technical sales and marketing, which involved designing and implementing novel quantum algorithms, which I then used as a way to teach quantum algorithm design on annealing processors. I worked on a novel way of training neural networks using optimization instead of backpropagation, and applied the quantum computer to some other machine learning techniques like binary classification, unsupervised feature learning, and dictionary learning.

MAX-CAT

One application that I wrote as part of this teaching activity was what I believe to be the development of the world’s first quantum computer game (that could be played against a real quantum computer), in 2011. The game was called MAX-CAT (a nerdy joke pertaining to max-sat, a type of mathematical problem that the algorithm was based upon). MAX-CAT was a single-player territory maximization game, where the player places cat cards with “catstats” on a board, indicating the power of the cat itself and the influence it has on its neighbors. The game was inspired by the concept of Schrodinger’s Cat - a thought experiment where a cat inside a box - unobserved - is in principle able to be put into a quantum superposition of both dead and alive states at the same time. During gameplay, the player attempts to place cats strategically into box icons on the board one by one such that they maximize the number of cats that come out alive at the end of the game. As the cat cards are placed on the board, the game code then turns the scores on the cat cards into an Ising spin problem (a mathematical problem), which is then sent to the quantum computer to be solved. The solution returns 0 or 1 for each “box”, signifying that the cat is dead or alive. As the player gets better at the game they “level up” to face more complex board configurations.

The cool thing about MAX-CAT is that you require a quantum computer to calculate the score, so as the game board gets bigger (say beyond 100 fully connected vertices on the board) it is a game you literally CAN’T play classically (even with the most powerful supercomputer). It would take millions or even billions of years to calculate the score using a classical computer, but only a few seconds on a quantum annealing machine.

Jeff Hawkins

Around mid-2009 (a year before I joined D-Wave) I read a book that changed my life. It was called On Intelligence by Jeff Hawkins. This book puts forward a beautiful computational model of the brain, which explains how intelligence can arise, why humans are so much better at certain tasks than computers, and also explains the hierarchy of intelligence in animals in terms of this model. It was the first time in my life that I’d thought about the brain as being something we could describe algorithmically or mathematically, as opposed to it being a wet, squishy, biological bunch of cells.

Given that I was working in quantum computing at that time, and I’d been focusing on Quantum Machine Learning, that book kept coming back into my mind, and I decided to watch everything that Hawkins or Numenta had published online (which was a lot) to try and figure out if there was a way I could use quantum computers to speed up the type of machine learning algorithms that Numenta were interested in. I had no idea my thinking would open up a can of worms that would take my career in an entirely different direction.

Robots and Teleoperation

I thought “I’ll try to figure out how to implement a Numenta-style hierarchical predictive memory, so that I can then look at it and see if quantum computing can help it go faster”. For that I needed some training data. The Numenta algorithms model cortical columns, which deal with both sensory data (vision, audio, touch) and motor data (outputs to move muscles). Finding data for the sensory system was easy (just collect data from a webcam or a microphone) but the motor output system was more tricky. Motors don’t just move themselves (like a webcam collects images) so I needed something to move the motors for me. This was the point at which I discovered teleoperation.

In order to collect data on a robot and teach it to move, you need to move it yourself first, demonstrate what to do, so that the robot can then learn to do it on its own. I had an idea that if you could control a human-like robot with a human teleoperation suit, and collect enough data, you could in theory train that robot to behave just like a person.

I played with this idea in my head for a while, and was eventually convinced that it couldn’t easily be turned into a quantum machine learning application but was a good idea in and of itself (we’ll come back to that assumption later!). So I founded Kindred Systems Inc. in 2014 to explore the application of teleoperation to human-like robots, and the use of machine learning algorithms to train on the resulting data. I built over 50 robots by hand to collect various forms of data for machine learning algorithms. The most interesting of these (I thought) were the QT3 robots. 3D printed, they were fully self-contained, untethered, small wheeled humanoids with wifi and 5G support that could be teleoperated and talked to cloud services on the back end to capture all the data that streamed to and from them. I think they were pretty ahead of their time for 2015!

Kindred continued on, and the company was very successful, but by 2018 had moved away from the original mission of human-like robots, as the teleoperation technology had found interesting niches in the warehousing and eCommerce space. In order for Kindred to remain focused on customer success, the research work that was still being done on human-like robots and AGI was then spun out into what became Sanctuary AI.

Ultra-human-like robots

I co-founded Sanctuary in 2018 with a move towards focusing on ultra-human-like robotics. The originating idea was that if we want to build human-like AI, we need an embodiment that is very human-like itself. And so followed the birth of the synth. Synths were extremely human-like robots with human-like faces. Progress towards making synths a reality helped galvanize the growing engineering team around the concept of human-like AI.

The term Sanctuary was adopted as the name of the company, because the goal was to create a place that was both a Sanctuary for people that believed in a vision of a future where human-like robots can live amongst us and care for us, and a sanctuary for those nascent, emerging AI beings themselves.

Unfortunately, despite many engineering best efforts, ultra-human-like robots with human-like faces still fall into the uncanny valley to some extent, even today. This is because the mechanical and control issues of engineering tiny motors into a robot’s face and making them move exactly like human muscles is challenging. So I directed the team to relax the requirement for the robot to be ultra-human-like. The humanoid form, function, and aesthetic would remain, but the requirement to replicate a likeness to the human face was removed. The subsequent robot designs could still exhibit very human-like motion and abilities under teleoperation, and the engineering team could now focus more on the functionality of the robot’s hands and body, and worry less about how the face looked and moved.

Consciousness

I used to be a proponent of the Strong AI hypothesis (everything the human mind does can be replicated with a classical computer given enough compute resources and time). So, after working on humanoid robots for 10 years, I thought at some point I’d see the spark of life I’ve been looking for. Specifically I thought for sure we’d have figured out whether or not these things are conscious. But we haven’t. When I look back over my story, I realize that I’ve been thinking about consciousness for a long time. Back in the early quantum days, I’d been exposed to the Hameroff-Penrose idea of ORCH-OR (a quantum theory of consciousness). I thought it was curious back then (because I was a big advocate of quantum computers in general - so any potential application of the technology was fascinating to me) but didn’t take it all that seriously. And through Imitation Learning, I’ve been trying together with colleagues to “sideload” consciousness into machines for a very long time.

But I still don’t believe we have conscious robots. And even worse, we don’t even have any ideas for how to engineer consciousness into AGI cognitive architectures. We’re sort of “waiting for it to emerge.” I’ve now shifted my scientific stance. I’m now open-minded to the idea that consciousness might not be something that classical computers or classical information processing can support.

Quantum Brains

I’m now interested in quantum consciousness because it is the closest thing we have to a theory where a type of computation is happening that goes beyond classical computation. There may be other candidates for information processing that could go beyond our known physics - hypercomputing in higher dimensions, or something weird going on with dark energy :) I don’t know. But quantum theory seems like the most well-formulated and well-explored territory as a place to begin looking. So I’m now working on a new startup idea, called Nirvanic, which explores these topics.