With Greg Sherwin and Twain Liu

At the 2018 House of Beautiful Business gathering in Lisbon, I had the pleasure of interviewing Twain Liu—entrepreneur, artist, and AI scientist—as part of the “Bureau of Inquiry” series of deep conversations. Liu has a provocative and enlightening take on how three powerhouses of Western intellectual history are still influencing scientific development today, what it portends for AI, and how one hero—Leonard da Vinci—offers an alternative view. What follows is an adaptation of our rather unconventional conversation.

Greg Sherwin: Aristotle, Descartes, and Leibniz play a very prominent role in your thinking about artificial intelligence, which is not the typical shortlist of influencers you might find in most Silicon Valley AI labs. Even more curiously, you cast them as the intellectual villains of the story rather than as heroes. So let’s start with Aristotle. For someone considered the father of Western philosophy, why — if I may paraphrase your playful choice of words—“Aristot-hole”?

Twain Liu: He’s “Aristot-hole” because he created this giant black hole in reasoning with his zero-sum, divisive “them OR us” dualism. Everything in Aristotle’s world is only a black or white, “True” or “False” literal. Even his syllogism is literal:

All men are mortal.
Aristotle is a man.
Therefore Aristotle is mortal.

Let’s compare that with Turing’s syllogism which is much more imaginative, witty and makes us wonder laterally, “What does Turing mean and why did he say that?!”

Turing believes machines think.
Turing lies with men.
Therefore machines do not think.
Yours in distress,
Alan.

The other thing that’s objectionable about Aristotle is his inconsistency. So, for example, some of his quotes sound noble on the surface, such as “Educating the mind without educating the heart is no education at all.” Yet, when we examine his writings more deeply, we discover he also propounded the idea that emotions and intuitive feelings are “irrational elements” associated with women, and he saw this as a valid reason to remove them from scientific method and learning processes. So there’s no heart or art in Aristotle’s methods. That too is why I call him “Aristot-hole.” He has a hole in his heart and his art.

Sherwin: Descartes elevated ethics as the highest and most perfect form of science. Yet he also advocated the reductionist view of man as a machine, and his reductionist thinking lies at the roots of the dehumanizing bias and discrimination in our modern data-driven algorithms. Is there a way we can reconcile these seemingly incompatible truths within Descartes’ own lens of dualist thinking?

Liu: The main problem with Descartes isn’t just that he wasn’t strong enough to resist Aristotle’s binary reductionism and decoupling of emotions from our mental processes. It’s his conceits of, “I think, therefore I am” (cogito ergo sum) and:

It must certainly be concluded regarding those things which, in external objects, we call by the names of light, color, odor, taste, sound, heat, cold, and of other tactile qualities … ; that we are not aware of their being anything other than various arrangements of the size, figure, and motions of the parts of these objects which make it possible for our nerves to move in various ways, and to excite in our soul all the various feelings which they produce there. (1644 Principles of Philosophy, p. 282).

The subtext is that he, Descartes, as the observer, has the faculties and power of intelligence but the subject being observed doesn’t. To Descartes, the subject has no matter, no meaning, and no value and values—except what Descartes the observer classifies, confers, and infers. Now, this dehumanization and disregard for the subject is just about acceptable when the subject is a stone or a bit of metal. It’s completely conceited, egoistic, and selfish of Descartes when the “size, figure and motions” belongs to another human being or organism with anima (from Latin anima, “breath, spirit, life”)—such as animals, plant life, insects, birds, and so on.

It would have been much more considered if Descartes had written, “Id nos percipimus ergo sumus”—“We sense so we are”—and if he’d recognized that the observer and the subject shape each other’s senses and knowledge in dynamic and symbiotic ways.

Let’s think about our everyday realities. Every person we see, speak with, or interact with imparts some information that then transforms our perspectives, thinking, and behaviors. In Descartes’ paradigm, the thinking processes are unilateral; it’s his emotion-void worldview imposed, top-down, on everyone else.

Sherwin: Silicon Valley is enamored with, if not also blinded by, decades of tremendous success with classical computing and binary thinking. The roots of this thinking are supported by a rather blind faith that the entire universe of things, thoughts, and even feelings can be reduced and fully described by a collection of 1s and 0s. So why is this such a limiting and dangerous idea?

Liu: Well, to start with, nature and physics research show us that our intelligence evolved over 13.8 billion years from quantum stardust—as whole atoms with electrons zipping in different directions, interacting with each other and with protons, neutrons, and force fields to make ever-evolving complex intelligence. In tandem, biochemical and neuroscience research shows that we’re a complexity of intertwined male and female DNA+RNA with the core building block of our intelligence being from eukaryotic cells, formed 2.7 billion years ago.

By comparison, the collection of 0s and 1s [bits] is a man-made mathematical invention of Leibniz’s from the period 1697 to 1701. The idea that our entire universe of information and intelligence which evolved over 13.8 billions years can be, retrospectively, fitted according to one man’s ideas of 0s and 1s is scientifically unsound. There is much more complexity, dimensionality, nuance, and cultural context in human intelligence than we currently have the mathematical and scientific tools to measure and model. Moreover, the idea that the faster processing of 0s and 1s makes the machines more intelligent than us and that we should subjugate and parameterize our language, culture, value, and values to their {0, 1} limitations is something we should, unequivocally, reject.

Sherwin: Gender identity is proving to be a very topical social example of how we humans might not so neatly fit within the confines of binary, dualistic thinking. We now have public conversations about people being non-binary. In a broader context, would you say that it’s the ultimate truth that we all are essentially non-binary?

Liu: That’s easy to reason. What’s in our DNA+RNA that makes up our gender? Thymine, Cytosine, Adenine, Guanine and Uracil. None of them are binary (0 or 1 bits). Again, imposing limited and limiting mathematical definitions of scalar numbers onto natural, dynamic biochemistry in this way is scientifically unsound.

Sherwin: Dualistic thinking isn’t just about binary classifications either. There are many examples where we humans create mental shortcuts to make the world more understandable by placing everything in its own imaginary box with a label. Take a lot of affective computing research, or what we might call “emotional AI.” With a heavy basis on the work of psychologist Paul Ekman, much of it is founded on a taxonomy of discrete human emotions, akin to painting the full range of human emotions with only a handful of primary “colors” such as anger, disgust, fear, etc. What are the pitfalls of reducing everything to, say, ten traits instead of just a 1 vs 0 binary?

Liu: In some ways, Ekman’s approach is today’s version of Descartes (“We are not aware of their being anything other than various arrangements of the size, figure, and motions of the parts”). It assumes the observer has the right to classify and infer the subject’s internal states of emotions just by their external appearance (the shape and angles of their facial features, upturned corners of the mouth indicating a smile, for example) and the subject’s perceptions of their own internal state and natural propensities and interactions of their emotion genes are of no matter. It’s judging a book by its cover, writ large.

Since Eckman, there’s been subsequent research from people like Lisa Feldman Barrett of Northwestern University, who point out that those ideas of different cultures expressing their emotions with the same facial adjustments in their features is simply not the case. Moreover, the classical approaches to emotion study like Ekman’s misses out on providing context around the person’s image and this also affects how the tester in the psychology study rates the expression as negative or positive and each of the six standard emotions of fear, anger, disgust, sadness, happiness and surprise. For example, if we see someone’s face within the context of them being on a rollercoaster, we may infer they’re experiencing fear and surprise whereas, if we only see a picture of their face with enlarged eyes and raised eyebrows, we’d only read that as surprise.

Interestingly, there’s a 2004 study by Lu and Gilmour which found that the American idea of happiness emphasized being upbeat, whereas the Chinese idea of happiness focused on being solemn and reserved. So if Ekman’s emotions-inference-by visual-features-mapping was applied to this scenario, it would miss out that the solemn-faced, non-smiling Chinese person was actually happy!

Sherwin: In the past few years we’ve heard many more cautionary tales of the biases at every level in our various machine learning and computational systems. But it’s also true that humans have always been rife with bias and mental shortcuts, certainly long before the invention of the computer. Thus these systems often just naturally reflect what is already there, inside of us. Do you think the coming age of AI might exacerbate these bias problems, or is there a real chance to finally identify and correct some of them?

Liu: It’s vital to know the differences between machine biases and subjective human biases, and not to conflate or confuse them. Machines biases are to do with mathematical definitions and tools we’ve inherited since the ideas of chance, probability, and likelihood were first written about — by Cardano in 1563, Pascal and de Fermat in 1654, de Moivre from 1633 onwards, Bayes in 1763, and Gauss, Markov, and others.

The thing to note is that all of their ideas were based on non-human phenomena: dice, cards, chess, board games, ideal gases, and so on, which have none of the direct and indirect qualia of experiences that humans have, i.e., our language, consciousness, culture, free will choices, socialization. Machine biases are about quanta. Subjective human biases are about qualia.

Sherwin: Quite frankly, in 2000-plus years of mathematics, we’ve never had any tools that measure subjective human biases in coherent, comprehensive, representative, and considerate ways.

Liu: Now, the lack of appropriate scientific tools for measuring subjectivity matters because there’s a lot of discussion in the AI community about “fairness is subjective” and a lot of white papers from policy-makers who are trying to address the issues of algorithmic biases — which coincide with the absence of female product designers and engineers in the room when industry-wide frameworks and apps like Google Word2Vec and MS Taybot are being made.

It’s become clear there’s systemic biasing by the AI, and the data collected previously skews male. This biasing has been shown in Google Word2Vec (a core framework in Natural Language Processing), in how search engines and adtech bias against different populations (e.g. showing more technical job ads to men rather than to women so reducing their opportunities; surfacing negative examples of African Americans compared with positive examples of white European-looking people (in the types of political and marketing content that’s shown on social platforms), Amazon needing to scrap its millions of dollars recruitment by AI project, loan calculations, and more.

There are definitely opportunities for us to do some first principles scientific invention in the area of subjective human biases and how to measure and model it appropriately for people.

Sherwin: Is quantum computing poised to fix a lot of this? And if so, how?

Liu: The readiness of quantum computing (QC) for prime time processing of data varies, depending on which research publication we read. Certainly, in the last couple of years, Big Tech has started to release some quantum tools to the developer community: MS Quantum Katas, IBM Q Experience, and Alibaba’s 11-Qubit Cloud service are just some examples.

Quantum ideas and methods play a part in the way my systems invention works. I included some Dirac-based notation in the patent filing back in 2013. Quantum has been an area of interest since I was about 14. Sir Roger Penrose’s ‘The Emperor’s New Mind’ was sandwiched between Robert Pirsig’s Zen and the Art of Motorcycle Maintenance and De Bono’s Lateral Thinking on our bookshelf alongside Descartes, Nietzsche, Schopenhauer, Adam Smith, Canetti, and others. Then, in 2003, I read an article in New Scientist entitled “Entangled Photons Dance Across the Blue Danube.”

Since I’m into art and science, those ideas of photonic poetry in motion completely captured my imagination.

Sherwin: One thing your three horsemen of the AI apocalypse hold in common is Western philosophy. How does Eastern philosophy and culture provide an alternative for how we might view the world, as typically represented by the Chinese philosophical construct of the yin and yang?

Liu: The Chinese philosophy of Yin and Yang can be traced back to 2962 BC, according to some researchers, whilst Greek philosophy started with the triumvirate of Socrates, Plato, and Aristotle from around 470 BC onwards. Now, it’s important to note that each of the Greeks had different thinking from each other. For example, Plato’s ideas were closer to the Chinese model — which is more inclusive of women and acknowledges intelligence in animals too — whilst Aristotle’s paradigm excludes them.

Ah, yes, the other reason I call him “Aristot-hole” is because he was a sexist and a racist and this is articulated in a February 2017 article by Stephen Cave of the Leverhulme Institute at Cambridge University:

“At the dawn of Western philosophy, intelligence became identified with the European, educated, male human. It becomes an argument for his right to dominate women, the lower classes, uncivilized peoples and non-human animals. While Plato argued for the supremacy of reason and placed it within a rather ungainly utopia, only one generation later, Aristotle presents the rule of the thinking man as obvious and natural.

Needless to say, more than 2,000 years later, the train of thought that these men set in motion has yet to be derailed. The late Australian philosopher and conservationist Val Plumwood has argued that the giants of Greek philosophy set up a series of linked dualisms that continue to inform our thought. Opposing categories such as intelligent/stupid, rational/emotional and mind/body are linked, implicitly or explicitly, to others such as male/female, civilized/primitive, and human/animal. These dualisms aren’t value-neutral, but fall within a broader dualism, as Aristotle makes clear: that of dominant/subordinate or master/slave. Together, they make relationships of domination, such as patriarchy or slavery, appear to be part of the natural order of things.”

Meanwhile in the opening chapter of I Ching, the Book of Changes there’s this principle:

Chapter 1.有名，万物母。 Naming is the mother of all things created.

Moreover, when we look at the graphic of Yin Yang it’s about the dynamic symbiosis and complement of male and female. There’s none of Aristotle’s divisive hierarchy of 1 = true = valid = rational = male OR 0 = false = invalid = irrational = female.

Sherwin: Let’s turn from your AI villains to one of your obvious heroes: Leonardo da Vinci. What might we learn and apply from this 15th century thinker and artist?

Liu: Da Vinci is wonderful in so many ways! My dad, in his infinite wisdom, showed me “Vitruvian Man” when I was about eight and I got him immediately. Since then, he’s been my inspiration in synch with Yin Yang, art, and the natural sciences for most of the things I think and do.

The main thing that da Vinci teaches us is to be curious and to pay attention to things that aren’t obvious: “The artist sees what others can only glimpse” and “Principles for the Development of a Complete Mind: Study the science of art. Study the art of science. Develop your senses — especially learn how to see. Realize that everything connects to everything else.”

Ours is the opportunity to re-train our brains to think with more perspectives and dimensionality like da Vinci and less like Descartes with his reductionism and unilateral and linear narrowness.

The future of technology and humanity, working in beautiful symbiosis, is ours to shape and paint. So let’s get to work doing it!