Garry Kasparov’s book ‘Deep Thinking’ is brilliant and full of insights for the machine Intelligence seeking world.
Brute Force
Though brute force transformed the human-machine battle, in favor of machines, brute force according to Kasparov is not intelligence, it is abandoning the foundation of human pattern recognition, creativity, and the power of randomness pegged against zillion computations, which may have offered a short term victory to machine-driven extrapolations, but is risky in the long run.
“It’s like taking a boat out into the middle of a lake and only realizing when the boat springs a leak that you don’t know how to swim.”
The brain is a flawed model for Intelligence
“Turing’s dreams of AI were that the human brain is itself a kind of computer and that the goal was to create a machine that successfully imitated human behavior. This concept has been dominant for generations of computer scientists. It’s a tempting analogy — neutrons as switches, cortexes as memory banks, etc. But there is a shortage of biological evidence for this parallel beyond the metaphorical and it is a distraction from what makes human thinking so different from machine thinking.”
“The human mind isn’t a computer; it cannot progress in an orderly fashion. Our minds see tech progress as a straight diagonal line, but it’s usually more of an S shape. Computers can only give you answers — they don’t know how to ask the right questions”
Kasparov highlights the gravity of past success, where success invariable creates biases. His intention it seems is to highlight that it is hard for an error-prone distracted brain to create sustained intelligence and hence augmentation. The beyond brain model of thinking is a prominent opposition and rightly so because conceptually speaking brain is a complex network and understanding the architecture of complexity is more important than the computational synapses.
Optimization and Innovation
Kasparov rightly highlights the informational limitations faced by the machine. The machine can’t see beyond its search “horizon”. An idea, similar to Herbert Simon’s bounded rationality, extended to machines. The machine’s intelligence is limited by the bounds of information available today.
“I am a veritable genius when I’m on the grid, but I am mentally crippled when I’m not.”
Kasparov correctly highlights the societal dependence on informational computation and how over-reliance on processed information is like outsourcing thinking which is serviced by illusory innovations that are an accumulation of many incremental optimizations and not real innovations, citing Max Levchin’s ‘Innovation at the margins’ that looks for small efficiencies instead of taking on more substantial risks in the main areas of business. Kasparov suggests that for disruption, deep thinking, going back to roots is essential for humans as machines lack understanding and purpose.
As compelling as the idea of incremental changes as an antithesis to innovation is, this is how Nature works to generate innovation. Though Kasparov mentions the S curve, he does not cite the Verhulst curve. The S curve is an observed mechanism in nature, which does not need deduction, induction, or abduction, it just works. The S curve explains how incremental changes form a curve that takes an idea to its peak while allowing other ideas to concurrently gain traction and grow. And assuming we can eventually teach machines common sense, we can give them understanding, understanding can give them consciousness, which can eventually give them purpose. To suggest that such a future is impossible is clairvoyance.
Cognitive Opportunity or Cost
Kasparov builds his case on augmentation by suggesting that brute force is not only incremental and not creating real innovation, it has also destroyed societal cognitive capability as it relies on instant fact-finding overusing one’s own captive capabilities and gives into the illusory perfection of machine analysis.
“When you run out of memorized preparation you have to start using your brain… This is why it’s important to use your brain while preparing, not just the engine… Over-reliance on the machine can weaken not enhance your compensation if you take it at its word all the time…Computers, like visiting aliens, don’t have common sense or any context that they aren’t told or cannot build…”
Reward-oriented machines that only care about the outcome and fraying cognitive abilities which amplify errors may make for a societal disaster, but Kasparov assumes here that context is driven by content. What if it was the other way around?
What if there was a world, an alien world, where context was independent or agnostic to content, in such a world, machines won’t need new data to find context. They would be ready to traverse any realm powered by their knowledge of a universal context. Such a world would not only have thinking machines but also would not suffer a cognitive gap, as cognition would become contextual, architectural, and hence unambiguous.
A world where new information may not create new context and errors would be about understanding dynamic context. Where machines could make their own rules about contexts and may not need to be given pre-set constraints. Machines could learn from observation without rules.
Process generating Mechanism
Kasparov concludes by suggesting that a deep-thinking brain with an intelligent machine-driven process makes a superior team that could handle the shocks that come from novelty. Such a team won’t be befuddled by the Berlin defense, where the queens come off the board very early.
I think differently and believe that process is somewhere connected to the outcome probabilistically. A process can deliver both good and bad outcomes probabilistically and just because an expected positive outcome has happened more times while using a process does not give it a certainty of happening again. A process does not come out of thin air, it comes from a process generating mechanism, a framework, which by its way of functioning makes a certain process effective or ineffective. Such a mechanism has more intelligence than the process as it understands when a process would work and when it should fail. Simply speaking, the mechanism would answer which processes to choose and what outcome to expect.
Deep and Shallow Thinking Mechanism
My views contrast with Kasparov in view of the necessity of intelligence to be deep. Nature does not do deep learning, nor does it do deep thinking. Nature does not have the time to compute and calculate. Nature is an intelligent mechanism. If nature was digging deep it would not have experimented so much, so fast. Nature does not think deeply because its mechanisms can do with fewer resources, and extract disorder from order, leaving mysteries for us to figure out.
Computation is not the answer to intelligence. If computation was the answer, we should have solved many of the world’s problems by now, including the climate problem. After the quantum hype is over, we should go back to condensed matter physics. The physics of phase change. Nature is a tinkerer and does not use a sledgehammer to break a nut and thinking that we can bulldoze our way out of our intelligence rut is naive thinking.
Nature is not chess; it is chaotic and hence unsolvable. This is what is happening in the investment space. Brute force creates a make-believe world of top-notch capabilities, that lack common sense, work under a set of pre-set constraints, which are not tested for regimes and environments and hence amplify errors and risks in a domain which is already suffering from a legacy of underperformance. The view that now with machines we can compute faster and understand market patterns yet again like in chess, creates an illusion of intelligence.
The intelligence mechanism is not a brain. Even if it was, it would flip between its computational and non-computational states. The flipping state is a context that proves the absolute wrong and the relativistic probabilistic states correct. Thinking mechanisms can be deep and shallow at the same time, flipping between the two states, probabilistically, like a Markov Chain.
Fast and slow thinking is a borrowed expression from Instant and delayed gratification first laid out by John Rae, in Sociological Theory of Capital, 1834. If Information can flip between relevance and Irrelevance, particles can exhibit quantum and classical states, the disorder can transform into order, the effect can drive cause, intelligence cannot be anything but understanding such mechanisms.
Nature’s Mechanisms
Common sense is common because it is observed across mechanisms and a sense of such commonality can be assumed to be referred to as common sense. Building machines that have common sense begins with the idea of mechanisms.
Neural Network is a mechanism that attempts to simulate a learning environment. Why it lacks common sense or cognitive capabilities could be an indication of its architecture. Is there something missing in the neural network deep thinking that it fails to think?
The design flaw is obvious and starts with thinking about a model outside the human brain. If you align computers to compute, why would they think about anything else but computation? The whole idea of seeking context every time they deal with new data is inefficient, especially if there were simple ways to work on a common universal context for data.
The Future of Intelligence is Linguistics
“The main lesson of thirty-five years of AI research is that the hard problems are easy and the easy problems are hard. The mental abilities of a four-year-old that we take for granted — recognizing a face, lifting a pencil, walking across a room, answering a question — in fact solve some of the hardest engineering problems ever conceived… As the new generation of intelligent devices appears, it will be the stock analysts and petrochemical engineers and parole board members who are in danger of being replaced by machines. The gardeners, receptionists, and cooks are secure in their jobs for decades to come.”
Steven Pinker, ‘The Language Instinct’
Moravec’s Paradox as mentioned above might sound ominous but thinking machines may not have a challenge multitasking their role between gardening or investment management, all because they understand the pun.
Tim Berner Lee famously said that it is hard to build a semantic web because it is hard for machines to understand the human pun. The semantic web also referred to as Web 3.0 was expected to be a web of data that can be processed by machines, hence allowing a faster and more optimal search. The expectation was a meaningful manipulation, a language through which machines could make databases talk. The databases still don’t talk, but some machine reading has already started happening between inter and intra-domain databases. It may still take a while before computers can talk.
Tagging data is assumed to be a pre-step to listening computers. First comes the tagging, then the reading, relating, and the listening. After that comes “I don’t understand”, but that’s all fine if the user is patient and willing to give feedback to the computer about where it is wrong. Something like a parent teaching a child. This is not how we perceive technology today and this is not how Lee imagined the semantic web process, which is more about knowledge navigation than knowledge machines, more about searching for knowledge than about assimilation of knowledge. The future is eventually going to be something like what was immortalized in Startrek IV, The Voyage Home Scene 17 reminds us of an imagined future.
[Plexicorp Factory — Nichol’s office]
SCOTT: Well, this a fine place you have here, Doctor Nichols.
NICHOLS: Thank you. I must say, Professor, your knowledge of engineering is most impressive.
McCOY: Back home, we call him the miracle worker.
NICHOLS: Indeed. …May I offer you something, gentlemen?
SCOTT: Doctor Nichols, I may be able to offer something to you.
NICHOLS: Yes?
SCOTT: I notice you’re still working with polymers.
NICHOLS: Still? What else would I be working with?
SCOTT: Ah, what else indeed? I’ll put it another way. How thick would a piece of your plexiglass need to be, at sixty feet by ten feet to withstand the pressure of eighteen thousand cubic feet of water?
NICHOLS: That’s easy, six inches. We carry stuff that big in stock.
SCOTT: Aye, I’ve noticed. Now suppose, …just suppose, …I was to show you a way to manufacture a wall that would do the same job but be only one inch thick. Would that be worth something to you, eh?
NICHOLS: You’re joking?
McCOY: Perhaps the professor could use your computer.
NICHOLS: Please.
SCOTT: Computer… Computer!
(McCoy hands him the computer mouse which Scott tries to use as a microphone)
SCOTT: Ah! Hello computer?
NICHOLS: Just use the keyboard.
SCOTT: The keyboard. …How quaint.
(Scott rapidly types a formula into the computer that appears on the monitor screen)
NICHOLS: Transparent aluminum?
SCOTT: That’s the ticket, laddie.
NICHOLS: It would take years just to figure out the dynamics of this matrix.
McCOY: Yes, but you’d be rich beyond the dreams of avarice.
SCOTT: So, is it worth something to you? Or should I just punch up ‘clear’.
NICHOLS: No! No! (a female employee comes into the office) …Not now Madeline! …What exactly did you have in mind?
McCOY: Well, a moment alone, please. …Do you realise of course, if we give him the formula, we’re altering the future.
SCOTT: Why? How do we know he didn’t invent the thing!
McCOY: Yeah!
Summary
Societal dependence on information is why machines should eventually become thinking and more intelligent than human beings. Information is a drug, which though bounded is relentless and essential to induce noise into the psychological beings called humans. The perpetual functioning of a society happens because the disequilibrium created by noise and human effort to make sense of it — creates equilibrium.
Machine intelligence may struggle with context now but a common universal context won’t force it to create a new context with every new piece of data it encounters, every time, as believed by Kasparov.
Kasparov believes Intelligence is not a trophy because relying on things without common sense is foolishness. In a world where machines have common sense, intuition, and context, intelligence should know when to seek a trophy and when to lose, in the context of the understanding and purpose of its being. Cognition is a mechanism that should explain why reasoning is computationally light hence resolving Moravec’s Paradox. The Intelligent process comes from a mechanism, a framework, which adapts to the data.
Eventually, the world should wake up to the idea of duality which challenges the existence of absolute and embraces the relative and probabilistic. Augmentation and thinking machines are not conflicting ideas. Intelligence without a mechanism is weak AI. The generality of AI needs a model of the world, which subsumes the functioning of a brain but does not stop there, as it could be both deep and shallow.
Bibliography
[1] Kasparov, G. “Deep Thinking”, 2017
[2] Pal. M, “Human AI”, SSRN, 2017
[3] Simon, H. “Bounded Rationality and Organizational Learning”. Organization Science, 1991
[4] Pal. M, “AlphaBlock”, SSRN, 2017
[5] Francois, J. “Evolution and Tinkering”, Science, 1977
[6] Rae, J. “Sociological Theory of Capital”, 1834
[7] Bengio, Y. Lecun, Y., Hinton, G. “Deep Learning for AI”, Communications of the ACM, July 2021.
[8] Berners-Lee, T. Hendler, J. Lassila, O.”The Semantic Web”. Scientific American. 2001.
