Technology is Biology, Biology is Technology/ The Transformative Nature of AlphaFold2/ Love & Loss in The AI Age/ Training AI to think with Analogies

Technology is Biology, Biology is Technology. As much as I wish I came up with the sentence, “Technology becomes Biology, Biology becomes Technology”, both the sentence and the idea that comes with it are in reality taken out of the brilliant book “The Nature of Technology” by W. Brian Arthur, which I had the pleasure to read while on vacation last week. Reading the book led me to two considerations:

• First, it was yet again humbling to see how others came to similar conclusions as to the ones I came to, but much earlier than I did (the book is from 2009) and in a much better and more articulated way. Also, so much has happened in the last 10+ years on the front of technology (see the deep tech approach to innovation) that the book cries for a follow-on book or at least an in-depth analysis of the impact that the last 10 years have had on the framework presented
• Second, it made me reflect and think about the DeepMind release of AlphaFold2, something I mentioned last week in a very different context.

The two considerations are very much interlinked. Starting from the beginning: both the Nature piece from last week on AlphaFold2, but also the articles dedicated to it in this issue (see below) cover the DeepMind event very well from a traditional point of view, or, better said, they look at the first-order implications, i.e. a true milestone for humanity, opening the door to huge advancements in medicine and bio-engineering.

The two considerations above further crystallized in two additional thoughts inspired by the reading the book and also by reflecting on the second- and third-order implications of the protein unfolding algorithm from DeepMind (AlphaFold2).

First: In his book, Arthur describes technologies as morphing more and more into complex adaptive systems (CAS), with a strong recursive nature and even autopoietic properties. I had been implicitly thinking along these lines (hence the emphasis on ecosystem while looking at deep tech) but never articulated the thought with such clarity. One consequence of this evolution toward CAS is that technology is becoming like biology and that conversely, biology is becoming technology. AlphaFold2 will play a very important (but not sufficient) role in biology becoming a technology. But it will also contribute significantly to making technology like biology by accelerating the course toward nature co-design, by massively opening the option space when it comes to proteins and everything that can be done with them.

Second: One of the core tenets of the book is about the combinatorial nature of technology. I do think that since the book had been written, the combinatorial nature has evolved even further and in multiple cases, we are seeing more and more of a convergence of technologies, not only a combination, that is leading to an increased dimensionality of the technologies. It might sound like only a semantic difference, but the impact is in fact foundational, with most of the new groundbreaking technologies arising from the convergence of the three domains of matter and energy (substance), cognition and computing (thinking), and sensing and motion (actuation and interface). A technology like Synthetic Biology (the design and engineering of organisms) would not be possible without the convergence of biology, AI, and high throughput screening. And it is not only about “combining” for instance biology with AI, it is about blending the two to create something completely new (the convergence).

DeepMind Creates ‘Transformative’ Map of Human Proteins Drawn by Artificial Intelligence

Last December, DeepMind garnered international attention with its unprecedented performance at the Critical Assessment of protein Structure Prediction (CASP) competition. DeepMind's protein-folding-prediction model, AlphaFold 2, received a score that was "informally considered to be competitive with results obtained from experimental methods" - methods that typically take years to complete per protein structure (AlphaFold spends just days for its predictions). Researchers compared the leap at CASP14 to the 2012 ImageNet moment - though arguably it surpasses even that milestone in importance. Andrei Lupas, an evolutionary biologist at the Max Planck Institute for Developmental Biology, said AlphaFold's results were a game-changer: "This will change medicine. It will change research. It will change bioengineering. It will change everything."

Now DeepMind has created "the most comprehensive map of human proteins," consisting of AlphaFold's predictions of some 350k protein structures, including ~250k that were previously unknown, according to a New York Times feature. Scientists are likening the potential impact of the data to that of the Human Genome Project: "Most significantly, the release includes predictions for 98 percent of all human proteins, around 20,000 different structures.... it isn't the first public dataset of human proteins, but it is the most comprehensive and accurate." According to the New York Times piece, the map "may accelerate the ability to understand diseases, develop new medicines and repurpose existing drugs. It may also lead to new kinds of biological tools, like an enzyme that efficiently breaks down plastic bottles and converts them into materials that are easily reused and recycled."

News items:

RNA Breakthrough Creates Crops That Can Grow 50% More Potatoes, Rice

A genetic tweak that targets RNA can grow crops that yield significantly more food and show increased drought tolerance, announced scientists from the University of Chicago, Peking University, and Guizhou University.

The Jessica Simulation: Love and Loss in the Age of A.I.

This one really moved me when I read through it. It is a bit of a longer read, but really a fascinating piece. In 2019, while on the road due to wildfire smoke near his home, Bay Area programmer and indie game designer Jason Rohrerstarted working on a side project called Project December. Essentially a service that allows anyone to converse with realistic, short-lived chatbots for $5, the mysterious site was spotted by freelance writer Joshua Barbeau. Using the custom chatbot option, Barbeau fed Project December Spock quotes, finding the bot mimicked the half-Vulcan surprisingly well. Barbeau then had the idea of feeding the service social media posts and text messages from his late fiancee, who had died in 2012. In a beautifully told feature, San Francisco Chronicle's Jason Fagone details their realistic conversations.

An interesting aside. Last December, Protocol explored a fringe-y Microsoft patent that, in part, aimed to do what Project December is already doing - extending a human persona into the digital domain. Using all types of recorded media (including social media posts) as training data, the Microsoft chatbot would be able to text and sound like the real person. And yes, that was a Black Mirror episode back in 2013. Microsoft didn't limit its hypothetical chatbot to the dead: it could be "a friend, a relative, an acquaintance, a celebrity, a fictional character, a historical figure, an entity," and so on.

News items:

Money’s Quantum Time Bomb

Efforts are underway to protect the world’s financial networks from a quantum attack. But some question if the threat to crypto-assets is really as existential as feared.

Is the Future of Travel Underwater?

We don't talk about underwater tourism much here, which is sort of surprising, given the deep tech and life sciences angle, and the fact that the marine world covers 70% of the planet. Underwater tourism includes not just Great Barrier Reef exploration, but what we traditionally equate to touristic fare - (underwater) restaurants, (underwater) hotels, even (underwater) sculptures.

Scuba divining, a relatively expensive and time-consuming passion, has its own excursionist-lite alternative - the Seawalker experience on Green Island in the Great Barrier Reef, which allows people to "submerge while wearing a large glass helmet," minus the diving certification. There are less publically accessible experiences, like DeepFlight - an hour-long, $1,500 personal submarine ride; and even ancient sites that are getting renewed interest: "In Kas, Turkey, numerous underwater archaeological sites are drawing tourists to the area’s sunken cities, amphorae fields and Lycian rock tombs; while off the coast of Haifa, Israel, a Neolithic village attracts divers with its oldest-known coastal defence wall."

News items:

Work in These Sectors? Here's How Drones Can Help Your Bottom Line

Inspection drones are fast-becoming a key acquisition for a diverse array of companies, being used in major industries like insurance, mining, and aggregates, using cutting-edge technologies to drastically reduce the time workers spend gathering and analyzing data

Imagination & Creativity

The Computer Scientist Training AI to Think With Analogies

Artificial Intelligence: A Guide for Thinking Humans author Melanie Mitchell believes that training machines to recognize and process analogies are vital for unlocking AI's potential. "Today’s state-of-the-art neural networks are very good at certain tasks, but... very bad at taking what they've learned in one kind of situation and transferring it to another." According to Mitchell, this is "something that we humans do all the time without even realizing we're doing it. We're swimming in this sea of analogies constantly." Models like GPT-3 can "generate language very, very convincingly" without really understanding anything. But what machines can't do is "take something [they] already know in some way and map it to something new."

Mitchell has an interesting theory on how to integrate knowledge transfer into AI algorithms: provide them with a body analog. "My intuition is that we will not be able to get to humanlike analogy [in AI] without some kind of embodiment. Having a body might be essential because some of these visual problems require you to think of them in three dimensions. And that, for me, has to do with having lived in the world and moved my head around, and understood how things are related spatially."

News items:

I Got an Artificial Intelligence to Write My Novel

First-time fiction author and neuroscientist Erik Hoel feeds the jacket copy of his recent novel into GPT-3 and sees what happens.

Society & Ethics (S&E)

Panpsychism, the Idea That Inanimate Objects Have Consciousness, Gains Steam in Science Communities

As philosopher David Skrbina notes, one of the major frustrations in the scientific community has been formulating a good explanation for what makes us self-aware, and more broadly, why we are conscious at all (see David Deutsch's definition of a good explanation). According to Skrbina, "as far as I can tell, and the latest research I've seen, they have been unable to do this, which suggests that consciousness is either a deeper or a more complex phenomenon than most of our scientists have thought and maybe are willing to admit." Panpsychism - the idea that consciousness is "inextricably linked to all matter and simply grows stronger as a physical object become more complex" - is one idea trying to fill this void.

Here's philosopher Hedda Hassel Mørch summary of how panpsychists think about consciousness: "Panpsychism typically does not take all things to be conscious as a whole, or to have their own unified consciousness. Fundamental particles would have simple, unified consciousness. Sometimes, this simple consciousness 'combines' or unifies into more complex forms. This happens in the human brain—we have unified consciousness as a whole. But it probably doesn't happen in e.g. tables and chairs—these things are mere collections of independently conscious particles."

News items:

Facebook’s Disinformation Problem Is Harder Than It Looks

That Facebook can distribute dangerous amounts of misinformation around the world in the blink of an eye is not a new problem, but the attention stepped up when President Joe Biden told reporters during a White House scrum that Facebook was "killing people" by spreading disinformation.