One of the hosts, Carina Namih, announcing a talk at the Cutting Edge Stage, CogX, London 11–12 June

The Cutting Edge of AI — Part 2

Notes from day two of Exponential View talks and panels at CogX Festival, 11–12 June, London

Exponential View team ran one of the five main stages at this year’s CogX festival in London. We gathered some of the world-leading researchers, investors, founders and scientists to discuss the cutting edge of AI, compute, deep tech and human transformation — many of them are EV readers (note: if you’re not on the list yet, subscribe here).

I previously shared notes and the agenda from the first day on the Cutting Edge stage here. Below, I summarize key points from four talks from the second day. Head all the way to the bottom for videos and full agenda. All the talks are worth watching, really!

Building safe and accountable learned systems

Sarah Gold, founder of Projects by IF defined eight characteristics which are essential for safe and accountable machine intelligence, or learned systems, focusing on the real problems of real people and societies.

1. Legible and understandable decision-making to the people who use services powered by a learned system. Most of the existent services are hard to read, and don’t reveal what data they collect, how and why. Fixing this becomes particularly challenging in the learned systems, which are non-deterministic — there’s an infinite number of interfaces someone might experience.
2. Individuals should be able to override the system: trust is a two-way relationship, and won’t work if we enable the systems to make decisions on our behalf, without any feedback.
3. Recovery needs to be graceful. Learned systems will go wrong, but users need to be able to say “stop” or go back to the previous state in the decision-making process, without causing a total failure of the system.
4. Safe and accessible systems must have parts of it that are open. We need to have transparency built into the systems and course-correct before anything major goes wrong.
5. We need to be able to verify who has access to what data and why.
6. Take into account that trust is not an individualistic action. To enhance trust, users need representation by individuals, groups or organizations that will mediate the relationship.
7. System’s intent must be clear and public. Clarity on whose interests are being represented must be transparent.
8. Regular testing is a must. Any system requires testing and standards — we need to define what the testing eco-system will entail.

Philosophical framework for modern machine learning

Zavain Dar, Principal at Lux Capital, spoke about the philosophical evolution that’s pushing the advancements in AI. Zavain dedicated a whole issue of Exponential View to this topic, so head here for a more thorough thought-piece.

• Most scientists, mathematicians, physicists and philosophers who were early AI practitioners emerged from Kepler’s line of empiricism, which rests on three main postulates:

1. Metaphysical ground truth exists, whether or not we unearth it.
2. Language and abstractions sit at the correct layer of specificity to reduce the rules to a finite set of statements.
3. These rules might exist, but not necessarily graspable by the human brain.

• Applied to the field of AI, these postulates exist as:

1. Intelligence is something to be unearthed, exists independent of humans.
2. Intelligence can be reduced to a finite set of rules from syntax and semantics of code.
3. Intelligence sits within the grasp of humans explicit understanding and discovery — give enough talent, researchers, enough computers, and server and they’ll explicate the exact rules of intelligence.

• Modern machine learning has made these three assumption irrelevant; with growth in digitized data, availability of compute, maturation of computing paradigms, there’s a philosophical shift in how we view (computer) science. The New Radical Empiricism postulates:

1. Don’t assume there’s ground truth, but employ data to predict and model forward.
2. We’re no longer limited by the language or the machinery.
3. Grokability is irrelevant — ML is a discipline of using machines to push beyond the limits of human solvability.

The cutting edge of quantum computing

Microsoft’s Director of Quantum Business Development, Dr Julie Love, dedicated her talk to the opportunities quantum computing will open in the coming years, and the way to get there.

• The basic building blocks of computing haven’t changed over the last 4000 years, leaving classes of problems unsolvable. For example, despite the advancements with conventional building blocks, problems such as RSA-2048 key remain a problem conventional computers would solve in billion years; quantum computer of the right size and high enough quality of qubits, is a problem that can be solved in about 100 seconds.
• Quantum computing will be applied to niche applications, and exist as a co-processor for the classical workloads. Real-world problems quantum computers could help us solve: nitrogen fixation (a process of over 100 years old, currently consumes 3% of world’s natural gas), carbon capture, materials science (currently we lose 15% of electricity due to the losses in our grids), machine learning (faster, higher accuracy with less data; solving hard optimization problems).
• Quantum-inspired optimization: just by learning how we would use quantum computing to solve problems, we can map the problems into the conventional architectures. Even though they run slower than they would on quantum computers, they can still outperform the best in class — 4000x faster.
• The heart of the solution is quantum hardware. In order to be able to solve real problems, we need quantum hardware that scales 1000s, millions, billions of qubits. We need: scalable quantum foundation, qubits resilient to noises from the environment, cryogenic systems that can scale, new inventions at the quantum classical interface, powerful classical computers to control the quantum hardware itself, new platforms for correcting and identifying errors, scalable software stack, full integration with cloud and algorithms for real-world integration.

Beast machines — understanding consciousness in the AI age

Anil Seth, Professor of cognitive and computational neuroscience at the University of Sussex and co-director of the Sackler Centre for Consciousness Science, spoke about the essence of consciousness, and what it would mean for a machine to become conscious.

• According to Descartes, humans have a rational soul, an immaterial essence that exists apart from the world and its reality. He considered the non-human animals to be beast machines, made of flesh and blood, but lacking the physical soul to experience anything and occupy a special place in the mind of God. Machine soul of AI arises not in spite of, but because of our nature as beast machines.
• Consciousness is the presence of any kind of subjective experience, a sense of what is it like to be a certain creature. Consciousness has much more to do with being alive than with being smart. Brains didn’t evolve to play Go or to do machine learning. Their fundamental imperative is to keep organisms alive. Consciousness is the only thing that really matters, thus we must ask: is there a track from artificial intelligence to artificial consciousness?
• Perception is important in this, as its the best guesses of whats going on in the world, and inside a body. Perception is about control and regulation, surviving. The brain is continually predicting the future state of the organism, all physiological quantities that have to be preserved. Predictive perception is fundamentally shaped by the need to control and regulate the interior of the body. It is life, not intelligence that is a necessary condition for consciousness.
• Consciousness may not be a matter of information processing that could be simulated; it could be that a conscious machine must be a living machine.

Day 2 agenda and videos

Haiyan Zhang, Innovation Director, Microsoft Research Cambridge—The Power of Machine Learning and Design Thinking (video)

Vishal Chatrath, CEO, PROWLER.io — The Future of Decision-Making (video)

Dr. Yuan (Alan) Qi, VP and Chief Data Scientist, Ant Financial Services Group — Bringing Inclusive Financial Services to the World (video)

Sarah Gold, Founder, Projects by IF— Safety and Accountability in Learned Systems (video)

Alex Caccia, CEO, Animal Dynamics Engineering Swarms — Hard vs. Easy Problems in Machine Automation (video)

Sabine Hauert, Assistant Professor, University of Bristol, President and Co-Founder, Robohub.org — Engineering Swarms (video)

David Pinn, VP of Strategy, Brain Corp — Robotic Solutions That Scale (video)

Panel “Autonomous Machines” hosted by Kenneth Cukier, Senior Editor, The Economist, with Alex Caccia, CEO, Animal Dynamics, Sabine Hauert, Assistant Professor, University of Bristol, President and Co-Founder, Robohub.org, David Pinn, VP Strategy, Braincorp (video)

Sean Gourley, Founder, PrimerAI in conversation with Azeem Azhar, Founder of Exponential View — Closing the Intelligence Gap (video)

Jan Erik Solem, CEO, Mappilary — Mapping the Real World (video)

Georg Polzer, CEO, Teralytics — Changing How the World Moves (video)

Panel “Mapping the World to Build Smart Cities” hosted by Dr Larissa Suzuki, Honorary Research Associate, University College London, Jan Erik Solem, CEO, Mappilary, Georg Polzer, CEO, Teralytics (video)

Zavain Dar, Principal, Lux Capital — Machine Learning and New Radical Empirism (video)

Panel “Investing in Very Deep Tech” hosted by Azeem Azhar, Founder of Exponential View, Zavain Dar, Principal, Lux Capital, Leila Zegna, Founding Partner, Kindred Capital, Carina Namih, Partner, Episode1 Ventures (video)

Matt Jones, Principal Designer, Google AI — Centaurs or Butlers? Designing for Human Relationships with Non-Human Intelligences (video)

Vivian Chan, Co-Founder and CEO, Sparrho — Democratising Science with AI (video)

Chris Gibson, CEO, Recursion Pharmaceuticals — Radical Empiricism and Technology Enabling Translational Biology at Scale (video)

Shiva Amiri, Director of Data Science, Zymergen — Going Beyond The Bounds of Human Intuition (video)

James Field, Founder, LabGenius — Harnessing Evolution with AI (video)

Panel “Biochemistry Meets AI” hosted by Leila Zegna, Founding Partner, Kindred Capital with Chris Gibson, CEO, Recursion Pharmaceuticals, Shiva Amiri, Director of Data Science, Zymergen, James Field, Founder, LabGenius (video)

Kamil Tamiola, CEO, Peptone — Directed Protein Evolution — an AI perspective (video)

Noor Shaker, CEO, GTN — Generative Tensorial Networks: A Quantum Leap in Drug Discovery (video)

Dr. Julie Love, Director of Quantum Computing Business Development, Microsoft — Empowering the Quantum Revolution (video)

Dave Palmer, Director of Technology, Darktrace — The Future Impact of AI on Cyber Crime (video)

Robert Hercock, Chief Research Scientist, BT Security Research Practice — How To Defend Against Automated Cyber Attacks (video)

Mariarosaria Taddeo, Deputy Director, Digital Ethics Lab, University of Oxford. Research Fellow, Oxford Internet Institute, University of Oxford — The Next Wave of Interstate Cyber Conflicts (video)

Panel “Cyber Security in the Age of AI” hosted by Stephanie Hare, Researcher, with Dave Palmer, Director of Technology, Darktrace, Robert Hercock, Chief Research Scientist, BT Security Research Practice, Mariarosaria Taddeo, Deputy Director, Digital Ethics Lab, University of Oxford. Research Fellow, Oxford Internet Institute, University of Oxford (video)

Karina Vold, Philosopher & Research Fellow, Leverhulme Centre for the Future of Intelligence, University of Cambridge — Ethics of Brain-Computer Interfaces (video)

Anil Seth, Professor of Cognitive and Computational Neuroscience and Co-Director, Sackler Centre for Consciousness Science, University of Sussex — Being a Beast Machine: Consciousness, AI, and Life (video)

If you enjoyed reading this summary or watching the talks, you’re going to love Azeem Azhar’s Exponential View — the #1 newsletter about exponential technologies and their impact on the society.

Subscribe → www.exvw.co