Ray Kurzweil headshot and his book cover The Singularity is Nearer

The Singularity is Nearer

This generation of artificial intelligence bots have passed the famed Turing test, as we once knew it.

Experts may quibble with the rules, and we’ll continue to move the goal posts. Now, though, generative AI needs to play dumb to trick humans, because they move too fast, can write too convincingly and have too-comprehensive knowledge to be any person.

That’s from the new book from futurist Ray Kurzweil called “The Singularity is Nearer: When We Merge with AI”. It’s a followup to his 2005 book “The Singularity is Near.”

He’s among the best known, and longest-running champions of the kind of digital superintelligence that is called the singularity, which he says is coming — he estimates it by 2045, and has a bet with a friend that by 2029, AI will pass an even more rigorous Turing test he helped establish. His book is a wild romp of optimism and confidence. Anyone digging into the conversation will appreciate it. I recommend it.

Below I share notes for my future reference.

My notes:

  • His prediction is we’ll reach something we’ll call the singularity at or around 2045, he argues
  • Cost of computing chart
  • His 6 epochs
  • His 2002 bet with Mitch Kapor: AI Turing test by 2029
  • “Aided by superhuman AI, we will engineer brain – computer interfaces that vastly expand our neocortexes with layers of virtual neurons. This will unlock new of thought and ultimately expand our intelligence millions–fold: this is the singularity.”
  • Marvin Minsky his MIT teacher: the symbolic approach and the connectionist approach to creating automated solutions to problem
  • “Complexity ceiling” with rules based systems:  longest running expert system is Cyc by Lenat at Cycorp
  • Neocortex origin: 200m years ago a major brain development led to flexible mammal thinking, most of the climate had evolved slowly though until 65m year ago with an asteroid causing an extinction event that required adaptive thinking
  • And that thumb helps too
  • Darwin used Lyell’s geology (rivers creating canyons over time) as a metaphor to argue for biology evolutions
  • Alpha Go and then Alpha Go Zero which was its own teacher
  • The 2017 transformers paper was a major breakthrough
  • Amanda Askell prompted Chat GPT3 with philosopher John Searle’s “Chinese room argument” and it appeared to generate a novel analogy similar to David Chalmer’s metaphor that a cookbook explain the properties of a cake
  • (Author doesn’t explicitly cite the legal challenge of whether AI is learning from the internet or stealing copyright)
  • From few shot learning to zero shot learning
  • His early insight was that computation was the important ingredient for human intelligence (but his mentor Minsky thought algorithms could do it in the 1990s). The difference between connectivist
  • AI’s remaining shortcomings: contextual memory, common sense and social interaction
  • For the purposes of teaching singularity the most important gap is its ability to develop its own software
  • IJ Good: intelligence explosion
  • Ai researchers: FOOM (hard takeoff of AI quality)
  • When Watson beat Ken Jennings at Jeopardy, viewers also saw its second and third most likely guesses and they were often laughably wrong (universal suffrage was the awkward optimal answer to a question with answer European Parliament because it wasn’t really grasping but answering via probabilities)

  • AI effect is us moving the goal posts on what singularity is so author created rigor to the Turing test with Mitch Kapoor in their 2002 “long bet” on the singularity
  • Once mini nanobots can non invasively enter our brains through capillaries and connect through the cloud with better computation performance, we will add to our neocortex like we did 2 million years ago and then continue to scale it up: that’s the author’s singularity
  • What is consciousness?: “The ability to have subjective experiences inside a mind – and not merely to give the outward appearance of doing so. Philosophers call such experiences Qualia.”
  • Cambridge declaration on consciousness
  • Here’s a simplified synopsis of this section: Kurzweil discusses the idea that our experience of consciousness is deeply personal and can’t be fully shared or compared with others. For example, when we see colors like green or red, we can’t be sure if others experience these colors in the same way. Even with advanced technology, we wouldn’t be able to prove if the same brain signals produce the same experiences in different people. This leads to the concept of “zombies,” which philosopher David Chalmers describes as beings that behave like conscious humans but have no inner experience. Kurzweil argues that if such zombies could exist, it suggests that consciousness might not be directly linked to physical brain activity. This idea challenges the ability of science to fully explain consciousness, leading to what Chalmers calls the “hard problem of consciousness.” To address this, Chalmers introduces “panprotopsychism,” the notion that consciousness could be a fundamental force of the universe, similar to gravity or electromagnetism, rather than just a byproduct of physical processes.
  • We should assume that zombies don’t exist for moral reasons — so that means once an AI passes the Turing test that it may require moral rights
  • Conway’s Game of Life (1970): This ‘zero player game’ follows the birth Rule, death rule and survival rule to generate an infinite game: this simple system creates complex objects that cannot be predicted without being computed. Is that like life?
  • “Further, if consciousness can emerge only from the kind of order-and-chaos complexity of class 4 automata, this can be seen as a philosophical argument for why we exist-without such rules, we wouldn’t be here to ponder the question.
  • This opens the door to “compatibilism” —the view that a deterministic world can still be a world with free will.?! We can make free decisions (that is, ones not caused by something else, like another person), even though our decisions are determined by underlying laws of reality. A determined world means that we could theoretically look either forward or backward in time, since everything is determined in either direction. But under rule 110-style rules, the only way we can perfectly see forward is through all the steps actually unfolding. And so, viewed through the lens of panprotopsychism, the emergent processes in our brains aren’t controlling us; they are us. We arise from deeper forces, but our choices cannot be known in advance-so we have free will as long as the processes that give rise to our consciousness are able to be expressed through our actions in the world,” 88
  • Minsky “society of the mind” because our brains are multiple clusters that may have make different decisions and we don’t entirely understand how one decisions bubbles up
  • Ship of Thesus comparison to “You 2” if more and more of your brain was replaced with a digital copy, when do you cease to be you?
  • Hugh Ross: our universe appearing in its current form is about as likely as a Boeing 747 being assembled by a tornado hitting a junkyard
  • But the anthropic principle of selection bias: the universe we inhabit is the only one that could exist for us to do the math
  • Hans Moravec’s paradox: tasks hard to humans are easy to machines and vice versa
  • While brain emulation (WBE)
  • LOAR: law of accelerating returns: he clarifies he believes this only happens in some technologies
  • “Fading negative affect bias” shown in many studies including a 1996 Colorado state Richard Walker
  • Nostalgia (Johannes Hofer) is a coping mechanism . As author puts it “ if the pain of the past did not fade, we would be forever crippled by it.” Research like North Dakota state clay Rutledge
  • Kahneman and Tversky: “availability heuristic” means that because we remember more bad things (becaus ethat’s what news), we think it’s all bad
  • Author runs through a whole lot of data on “the world is getting better” — does not reflect the free and low cost information and electronics that benefit from LOAR
  • Homicide rates since 1300 in Europe, see below
  • Peter Singer’s 1981 “expanding circle” of empathy
  • His chapter on life is getting better is very much a repeat of other stuff: his section on democracy spreading specifically reads as a rehash of old Eurocentric ideas
  • Why doesn’t digital technology show exponential economic growth? Author argues (and told Christine laggards this at an IMF meeting g on October 5,2016) that it’s because “We factor out this growth by putting it in both the numerator and denominator” cheap computation means it looks smaller when it should be bigger
  • Predicts that by late 2020s, we will print clothing and common household goods out from 3D printers for pennies per pound
  • Vertical gardening improves food density
  • He argues that though wealthy may start with life extension, its like cell phones which got cheaper
  • AI will remove tasks but not necessarily professions
  • Why low productivity? “The close relationship between marginal cost, price and consumers willing to pay has been weakened”
  • (((My note: these techno-optimistic books often compare big old expensive computers with an iPhone but they also make way way way more iPhones than those old big computers so some of that value is captured in economic data)
  • Consumer surplus: Tim Worstall in Forbes in 2015; Facebook revenue was $8b but at the time consumer put into the service estimates it was about $230b
  • “As digital technology takes up a larger and larger share of the economy, consumer surplus is rising much faster than GDP suggests”
  • Calls out flagging labor-force participation rate but notes prime age workers and underground economy aided by internet and crypto and iPhone app economy
  • In 2018 Deloitte estimated 5m jobs in app economy
  • 2013 Oxford’s Frey and Osborne famous predictions of Ai jobs
  • Author argues the price competitive of digital technology will take over other industries too like food and clothing by eventually changing them
  • “One of the great challenges of adapting to technological changes is that they tend to bring diffuse benefits to a large population, but concentrated harms to a small group”
  • Covid vaccine development with machine learning
  • Simulated clinical trials by AI and the resistance he expects (bio simulation)
  • Peter Weibel said Hans Moravec felt we’d only ever be “second class robots”
  • Richard Feynman seminal 1959 lecture: plenty of room at the bottom
  • Aubrey de Gray: longevity escape velocity
  • Among his category of nanotobots Robert Freitas respirocyte artificial red blood cells that let you hypothetically hold your breath for four hours
  • Kurzweil takes a very different tack than economist Robert Gordon’s pessimistic view of our productivity
  • Perils he lists: nuclear, biotechnology, nanotechnology (gray poo)
  • Bill Joy’s 2000 “why the future doesn’t need us”

Leave a Reply