Houston, We Have An AI Problem (Part I)

0 b.png

When we try to pick out anything by itself, we find it hitched to everything else in the Universe.
~ John Muir

Intro 🎬

First things first. So this essay has nothing whatsoever to do with the venerable NASA space mission control—in case you had started making inferences from the title—though it’s got everything to do with my attempt at doing justice to conveying the gist of a riveting and brilliantly engaging new book which takes an eminently sensible approach to tackle, head on, a crucial problem in a sphere different from space mission control: that of Artificial Intelligence (aka AI).

With a drum-roll, I usher in

Human Compatible: Artificial Intelligence and the Problem of Control by Stuart Russell (Viking)

Yep, that book sure has got me excited lately, sending my mind burrowing down a bunch of kaleidoscopic tunnels of contemplation. To understand better why this book by my esteemed friend Stuart Russell (over at the University of California Berkeley) packs a wallop—surely he doesn’t need an introduction, does he?—consider the dilemma that yours truly faced when he decided to do justice to conveying his new book’s gist, here on our digs: How best to assemble the raft of ideas—plank by blank—on which to set the message sailing (after first putting the gist of Human Compatible in a bottle, so to say)?

Did I really say that? I believe so. So why don’t you grab a cup of coffee and settle down to check one engineer and computer scientist-turned-software practitioner’s take on what makes the message of Human Compatible so riveting?

Oh, and as the smart reader—that’s you—may have already figured out (from the title of this essay), this is but the first installment in what I conceive as a series of essays.


Any one who says that with the means at present at our disposal and with our present knowledge we can utilize atomic energy is talking moonshine.
~ Ernest Rutherford, the late Nobel Prize-winner, former head of the Cavendish Laboratory.

1. Surprises Have A Way Of Surprising Us🎈

As a long time devotee of the classic, foundational work on artificial intelligence—Artificial Intelligence: A Modern Approach, published by Prentice Hall—which is the standard textbook used in universities all over the world (as well as by practitioners who value its no-nonsense, crystal-clear exposition), and of which Russell (UC Berkeley) is, along with Peter Norvig (Google), the coauthor, I  was thrilled by the publication of his (i.e. Russell’s) new popular science book that is Human Compatible.

I got the stellar reading material (in Human Compatible) that I was expecting; no surprises there. What did surprise me was the rapidity with which I absorbed, and then was compelled to embrace, its message of making predictions about the vast unknown that stretches before us and into the future. For example, take the case of how—and this is only one of the many new things I learned from reading Human Compatible recently—the late Nobel Prize-winner and former head of the Cavendish Laboratory (Ernest Rutherford) had, back in the day, solemnly opined that

Any one who says that with the means at present at our disposal and with our present knowledge we can utilize atomic energy is talking moonshine.

Yeah, brah. That was Rutherford’s pontification back on September 11, 1933 (at the annual meeting in Leicester of the British Association for the Advancement of Science), that statement being duly reported in the Times of London the next morning—again, read all the juicy details for yourself in Human Compatible!

Anyhow, and as Russell memorably notes in that section, the Times of London reported the next morning that

Leo Szilard…, a Hungarian physicist who had recently fled from Nazi Germany, was staying at the Imperial Hotel on Russell Square in London. He read the Times’ report at breakfast. Mulling over what he had read, he went for a walk and invented the neutron-induced nuclear chain reaction. The problem of liberating nuclear energy went from impossible to essentially solved in less than twenty-four hours. Szilard filed a secret patent for a nuclear reactor the following year. The first patent for a nuclear weapon was issued in France in 1939.

Um, is that a bit sobering or what?

I mean, here we are, inhabiting the same planet that has witnessed the aforementioned moment of history, with us all on the verge of the third decade of the 21st century, and yet many continue to operate like the proverbial ostrich which buries its head in the sand on sensing danger approaching—Okay, so I’m referring here to all the otherwise smart naysayers who will reflexively shush-shush any talk of the control problem.

Apparently,  the specter of an out-of-control super intelligence hasn’t quite hit home for them, yet.


Than to be sure they do; for certainties
Either are past remedies, or, timely knowing,
The remedy then born—discover to me
What both you spur and stop.
~ William Shakespeare

2. How Uncertain, Really, Is The Future? đŸŽȘ

Yeah, we can surf uncertainty until the cows come home—and the deniers of the control problem are probably blithely surfing on the beaches right now—but you’ve got to draw the proverbial line in the sand at some point in time (and physically in the sand, too, I suppose.)

And it’s high time. That time is now; we, humanity as a collective whole, are not going to get second chances at getting this right (should we fail to address the control problem successfully the first time.)

(“Hey Akram,” I hear some of you mumbling, “What’s up with that Edinburgh sweatshirt hoodie our eyes spy everywhere around here?” Hold your horses; yo, everything will become clear, in good time.)

First, though,  and circling back to the earlier point about how I ended up solving my dilemma of identifying the best way to assemble a gist of the raft of ideas that are floating around, richly so, in the pages of Human Compatible: I hit upon the idea of building a collage in which some things (a book each) vary, while everything else (chiefly the hoodie) remains unchanged.

Second, if you could now please scroll down a bit to the next picture, what you might notice—hey Akram, it’s your Edinburgh hoodie again—is that while the garb hasn’t gone anywhere, we do spy a different book now standing upright in the foreground!

You got it. Oh, the symbolism. But please don’t read much into it at all; it’s merely serving an ornamental purpose, spiffing up our pictures.


Intuitively, we think that rational decision-making means exhaustively enumerating our options, weighing each one carefully, and then selecting the best. But in practice, when the clock—or the ticker—is ticking, few aspects of decision-making (or of thinking more generally) are as important as this one: when to stop.
~ Brian Christian and Tom Griffiths (in Algorithms to Live By: The Computer Science of Human Decisions)

3. Making (Good) Decisions Under Uncertainty 🐍

You tell me? I mean, how good, for example, are stock investors and money managers at predicting financial market trends? Exactly. It’s like looking into the crystal ball yet again (with different expectations, or maybe even the same ones). The bottom line is—and paraphrasing Yogi Berra from memory here—making predictions is hard, especially about the future!

But all is not lost. In fact, being an optimist, I have tremendous respect for the powerful mathematical machinery—starting with the pathbreaking work by Turing prize-winner Judea Pearl—which researchers have assembled to cope with non-determinism.

I invite you to check out the gripping  details (about ideas for coping with non-determinism) in the book itself, but try this on for size now:

Uncertainty has been a central concern in AI since the 1980s; indeed the phrase “modern AI” often refers to the revolution that took place when uncertainty was finally recognized as a ubiquitous issue in real-world decision making. Yet uncertainty in the objective of the AI system was simply ignored.
~ Human Compatible: Artificial Intelligence and the Problem of Control by Stuart Russell (Viking)

Yep, my sentiments exactly, too. Or, as the music band Toto memorably put it (in one of their erudite songs), “There are no guarantees; there are no alibis“. Oh yes.


Computers are getting smarter all the time. Scientists tell us that soon they will be able to talk to us. (And by “they” I mean “computers.” I doubt scientists will ever be able to talk to us.)
~ Dave Barry (the modern-day Voltaire, shall we say)

4. Those Machines Keep Getting Smarter 🏄

A related book—it’s How Smart Machines Think by Sean Gerrish, published by The MIT Press—does a good job of tracing the trajectory of the path taken by humans to make machines increasingly smart. Gerrish, too, is of the clear opinion that we humans, as a race, will keep building smart machines to meet and exceed our own abilities. Interestingly, he can’t help but remark that building machines in our own image is a uniquely human enterprise that’ll fuel this drive to keep on building.

And especially inasmuch as the aforementioned strand of thought relates to the book at hand—Human Compatible—I love how Russell captures this uniquely human endeavor in terms which, dare I say, will be instantly recognizable to the layperson. Thus, after pointing out that our strategy for building a particular intelligent agents will depend on the nature of the problem we face, Russell dives into a delightful consideration of the three-pronged strategy, starting with a thoughtful consideration of the environment that the agent will operate in (“chessboards” versus “crowded freeways” versus “mobile phones.”

For the other two, equally enlightening prongs (in addition to a consideration of the agent’s operating environment that we broached above)—namely the agent’s objective and the observations and actions that connect the agent to the environment—you simply have to check the good stuff yourself (in the pages of Human Compatible).


Beam me up, Scotty, there’s no intelligent life down here!
~ Captain Kirk (issuing command to his chief engineer, Scotty, signaling that he needs to be transported back to the Starship Enterprise)

5. The Control Problem, Redux 🍄

Oh goodness, where do I even begin? For starters, I’ve written a bit about the control problem myself, so let’s perhaps have you check some of that first:

For seconds, you’ll want to circle back to the pages of Human Compatible for a mind-expanding journey of sorts,  one that’ll likely have you looking at the control problem with fresh eyes—Whoah, I caught myself here using the word “likely”, you see; there again you witness the notion of likelihood, aka “probability, underwritten as it is into the quest for understanding the very fabric of our lives. As Pierre-Simon Laplace memorably remarked, many moons ago

Life’s most important questions are, for the most part, nothing but probability problems.

I’ll add that it is Judea Pearl—Turing Award-winner and one of my intellectual heroes—who perhaps places Russell’s book in its rightfully crucial place (more presciently than anyone else I can think of) by noting how

Human Compatible made me a convert to Russell’s concerns with our ability to control our upcoming creation—super-intelligent machines. Unlike outside alarmists and futurists, Russell is a leading authority on AI. His new book will educate the public about AI more than any book I can think of, and is a delightful and uplifting read.

Enough said. Now we make room for… Rumi!


A bystander said, “There’s no one with intelligence
in our town except that man over there
playing with the children,
       the one riding the stick-horse. He has keen, fiery insight and vast dignity
like the night sky, but he conceals it
in the madness of child’s play.”
~ Jelaluddin Rumi (in the translation by Coleman Barks entitled The Essential Rumi — Published by HarperOne)

6. Is There Anybody Out There? 🔭

It’s time for a little digression—regular readers of this blog (Programming Digressions of course) are already familiar with the unique phenomenon that is digressing. Anyhow, my original intent in placing a picture of (my copy of) the quintessential translation of the mind-blowing collection of Jelaluddin Rumi’s poems was to investigate whether you were still awake; as a writer, I have this odd longing for my readers to actually be in a wakeful state while they read the essays around here.

Anyhow, as I broadened my search for stuff to put in this section (entitled “Is There Anybody Out There?” as it is) I realized an uncannily relevant snippet from a Rumi poem, the one quoted atop this section, of course. The quest for intelligence—inside ourselves and outside, in the machines that we make in our image—is leading us to interesting places, isn’t it?

So there you have it; I’m not making this up. Honest.


Beguiling thus the wonder,
The wondrous nearer drew;
Hands bustled at the moorings—
The crowd respectful grew.
Ascended from our vision
To countenances new!
A difference, a daisy,
Is all the rest I knew!
~ Emily Dickinson (In her poem “XXXII: GONE”, from The Complete Poems of Emily Dickinson)

7. How The Journey long Began 🔬

Simply put—and if you’ll be kind enough to turn you gaze upward a bit—pictured here, standing right next to my Edinburgh hoodie, is the canonical AI text that made me fall in love with the field of AI. What more can I say?

Put another way, this is my way of recognizing where it all began.

We keep coming back to the fundamentals, to our moorings.


8. An Enthralling Sojourn 🚂

As we get near the end of this essay, we’re going to take another quick break to broaden our horizons and witness how yet another amazing book has shed scintillating light on some aspects of AI. This happens to be something about which I’ve written a bit. I can’t, in fact, think of anything better than to share those pointers that I suspect you will enjoy following:

Better still, compare and contrast: Find out for yourself how the paths of these two splendid narratives—that of Human Compatible and that of Plato And The Nerd—intertwine and intersect, all at your own pace these holidays, at your leisure.

Good? Got enough reading material for your holidays now?


The notion of understanding human intelligence is self-referential because the process doing the understanding must itself be intelligent. This notion therefore is likely vulnerable to the sort of incompleteness that Gödel found in formal languages, Hawking applied to physics, and Wolpert found in Laplace’s determinism.
~ Edward Ashford (in The MIT Press book entitled Plato and the Nerd: The Creative Partnership of Humans and Technology)

9. Hard Things Are Easy, The Easy Things Hard 🐙

Here’s another fine book—the excellent Artificial Intelligence: A Guide for Thinking Humans by Melanie Mitchell, published by Farrar, Straus and Giroux—albeit one that seems to hang on to a flawed premise: As best as I’ve been able to fathom, concerns for the AI control problem (in Mitchell’s book) have regrettably been relegated to the pile of “here lies much ado about nothing.”

Other than that, Mitchell’s is rock solid, though nowhere near as gracefully written as Russell’s Human Compatible. Man, he writes with uncommon grace. Plus Russell sets the record straight by tracing the history of how—partly due to the inability of logic to handle uncertain information—Good Old-Fashioned AI (aka “GOFAI”) came to connote a pejorative term, with this culminating in the bizarre happenstance that many AI researchers working today in the area of deep learning don’t know much at all about the rich heritage of logic. Go figure.

Russell rightfully goes on to quote the brilliant Demis Hassabis, the CEO of Google DeepMind, when he (Hassabis) observed in connection with the perceived, future arc of the deep learning enterprise which he leads—and I’m paraphrasing here—that he (Hassabis) would like his future systems to build up to this symbolic level of reasoning, which stand on the pillars of maths, language, and logic.

The bottom line, thus, emerges like so: The need for the capacity for representation and reasoning—possibly incorporated into probabilistic reasoning systems, perhaps into deep learning systems, or even into some yet-to-be-invented amalgam design—is not going away anywhere, anytime soon. Yeah, brah.


Distress not yourself if you cannot at first understand the deeper mysteries of Spaceland. By degrees they will dawn upon you.
~ Edwin A. Abbott (in Flatland: A Romance of Many Dimensions)

10. Which Do We Keep, The Shallow Or The Deep? đŸ‘ș

If you feel like your brain is turning to mush right now—having faithfully shared your torturous journey to its end with jours truly—may I offer some light relief? Here we go…

For one thing, the book featured in this section, The Mind Is Flat: The Remarkable Shallowness of the Improvising Brain by Nick Chater (published by the Yale University Press) is a riotous trip that will take you through some really interesting terrain, including a no-holds-barred grapple with the assumption that, below a mental “surface” of conscious awareness, lies a richly complex set of internal beliefs, values, and desires which govern our thoughts, ideas, and actions; and to know this depth, according to Chater, is to know ourselves.

Oh my! Wait till you see what lies in store for you.

But I digress.

So there. And once you’ve regaled yourself in the originality of the narrative—nearly every bit as original as Human Compatible, though tackling a different issue—The Mind Is Flat will beguile you in no time flat.

Now that you’re back, hopefully refreshed after that serving of light relief, I have some news for you: The first installment in this series of essays is drawing to an end. Hey, all good things come to an end, so perk up—as we solemnly bid our farewells till we are reunited in the forthcoming installments in the series—because I’m here to share some stuff that I’ve been meaning to check out, stuff that just might be what you’re looking for to stay happily engaged over the holiday:

Hmm… Maybe good things come in pairs.

And now we’re done. Oh wait, I’m being told there’s more
 So you just keep scrolling down a bit, right past the gaggle of rowdy hoodies coming right up!

control-collage b.png

Stay Tuned, Won’t You? đŸ“ș

“Wait, you got even more Edinburgh hoodies for us, Akram? You’ve got to be kidding!”

No I’m not, even as I invite you to check a book each from my modest library—standing upright—which I have surreptitiously slipped in along with each (cloned?) hoodie in the oh-so-symmetric collage above.

Oh yeah, they will be featured in what we’ve got lined up for you in the next installment in this series of essays that have veritably been inspired by Russell’s Human Compatible; it’s not a coincidence, then, that the aforesaid book is strategically placed at the epicenter of the collage. Surrounding Human Compatible are, starting with the top left—and going clockwise from there—are the following fine specimens of writing at its best, with good stuff each on the subjects of:

  1. Bayesian programming, done with elegant math
  2. Probabilistic programming, demystified
  3. Core concepts in artificial intelligence, pithily explained
  4. How (and whey) smart machines are getting smarter (and smarter!)
  5. Plato meets the modern-day nerd, oh my!
  6. Bayesian reasoning (done right)
  7. Slicing and dicing the world of (conceptual) modeling
  8. In-depth interviews with AI aficionados, who else?

Listen, wintertime is here, so go ahead and please make time for friends and family: Happy Holidays, you all!

PS: That motor-powered gondola in the picture below, jetting merrily on the waters overlooking those majestic spires rising from the Eastern, architectural wonder on the shore, beckons you, doesn’t it? Ahem, minor correction: I’m being informed that gondolas are indigent to a certain Western, quaintly Venetian city…

(“Hey, Akram,” they now say to me in hushed tones, as if to spare my sensibilities, “Just so you know—dude, aren’t you educated or something?—real gondola are human-powered, by ye olde gondoliers, and not by electric motors!“)

Folks, it’s a wrap (See you in 2020).

boat b.png

Your Comment Here!

This site uses Akismet to reduce spam. Learn how your comment data is processed.