Why Quantum?

This post is part of the Quan­tum Physics Se­quence.
Fol­lowup to: Quan­tum Explanations

“Why are you do­ing these posts on quan­tum physics?” the one asked me.

“Quite a num­ber of rea­sons,” I said.

“For one thing,” I said, “the many-wor­lds is­sue is just about the only case I know of where you can bring the prin­ci­ples of Science and Bayesi­anism into di­rect con­flict.” It’s im­por­tant to have differ­ent men­tal buck­ets for “sci­ence” and “ra­tio­nal­ity”, as they are differ­ent con­cepts. Bring­ing the two prin­ci­ples into di­rect con­flict is helpful for illus­trat­ing what sci­ence is and is not, and what ra­tio­nal­ity is and is not. Other­wise you end up trust­ing in what you call “sci­ence”, which won’t be strict enough.

“For an­other thing,” I con­tinued, “part of what goes into be­com­ing a ra­tio­nal­ist, is learn­ing to live into a coun­ter­in­tu­itive world—learn­ing to find things un­der­neath the sur­face that are un­like the world of sur­face forms.” Quan­tum me­chan­ics makes a good in­tro­duc­tion to that, when done cor­rectly with­out the hor­rible con­fu­sion and de­spair. It breaks you of your be­lief in an in­tu­itive uni­verse, coun­ters naive re­al­ism, de­stroys your trust in the way that your cog­ni­tive al­gorithms look from in­side—and then you’re ready to start see­ing your mind as a mind, not as a win­dow onto re­al­ity.

“But you’re writ­ing about physics, with­out be­ing a physi­cist,” the one said, “isn’t that… a lit­tle...”

“Yes,” I said, “it is, and I felt guilty about it. But there were physi­cists talk­ing com­plete non­sense about Oc­cam’s Ra­zor with­out know­ing the prob­a­bil­ity the­ory of it, so my hand was forced. Also the situ­a­tion in teach­ing quan­tum me­chan­ics is re­ally awful—I saw the in­tro­duc­tions to Bayesi­anism and they seemed un­nec­es­sar­ily difficult, but the situ­a­tion in quan­tum me­chan­ics is so much worse.” It re­ally is. I re­mem­ber sit­ting there star­ing at the “lin­ear op­er­a­tors”, try­ing to figure out what the hell they phys­i­cally did to the eigen­vec­tors—try­ing to vi­su­al­ize the ac­tual events that were go­ing on in the phys­i­cal evolu­tion—be­fore it dawned on me that it was just a math trick to ex­tract the av­er­age of the eigen­val­ues. Okay, but… can’t you just tell me that up front? Write it down some­where? Oh, I for­got, the math doesn’t mean any­thing, it just works.

Fur­ther­more,” I added, “know­ing about many wor­lds, helps you vi­su­al­ize prob­a­bil­ities as fre­quen­cies, which is helpful to many points I want to make.”

“And fur­ther­more,” I said, “re­duc­ing time to non-time is a pow­er­ful ex­am­ple of the prin­ci­ple, in re­duc­tion­ism, that you should re­duce some­thing to some­thing other than it­self.”

“And even fur­ther­more,” I said, “I had to break my read­ers of trust in Science, even trust in physi­cists, be­cause it doesn’t seem pos­si­ble to think and trust at the same time.”

“Many-wor­lds is re­ally a very clear and sim­ple prob­lem,” I said, “by com­par­i­son with the challenges you en­counter in AI, which are around a hun­dred times less clear-cut. And many sci­en­tists can’t even get many-wor­lds, in the ab­sence of au­thor­ity.” So you are left with no choice but to as­pire to do bet­ter than the av­er­age sci­en­tist; a hell of a lot bet­ter, in fact. This no­tion is one that you can­not just blurt out to peo­ple with­out show­ing them why it is nec­es­sary.

Another helpful ad­van­tage—I of­ten do things with quite a few differ­ent pur­poses in mind, as you may have re­al­ized by this point—was that you can see var­i­ous com­menters who still haven’t got­ten it, who are still say­ing, “Oh, look, Eliezer is over­con­fi­dent be­cause he be­lieves in many-wor­lds.”

Well, if you can viscer­ally see the ar­gu­ments I have laid forth, then you can see that I am not be­ing care­less in hav­ing an opinion about physics. The bal­ance of ar­gu­ments is over­whelm­ingly tipped; and physi­cists who deny it, are mak­ing spe­cific er­rors of prob­a­bil­ity the­ory (which I have speci­fi­cally laid out, and shown to you) that they might not be ex­pected to know about. It is not just a mat­ter of my form­ing strong opinions at ran­dom.

But would you be­lieve that I had such strong sup­port, if I had not shown it to you in full de­tail? Pon­der this well. For I may have other strong opinions. And it may seem to you that you don’t see any good rea­son to form such strong be­liefs. Ex­cept this is not what you will see; you will see sim­ply that there is no good rea­son for strong be­lief, that there is no strong sup­port one way or the other. For our first-or­der be­liefs are how the world seems to be. And you may think, “Oh, Eliezer is just opinionated—form­ing strong be­liefs in the ab­sence of lop­sided sup­port.” And I will not have the time to do an­other cou­ple of months worth of blog posts.

I am very far from in­fal­lible, but I do not hold strong opinions at ran­dom.

“And yet still fur­ther­more,” I said, “tran­shu­man­ist mailing lists have been ar­gu­ing about is­sues of per­sonal iden­tity for years, and a tremen­dous amount of time has been wasted on it.” Prob­a­bly most who ar­gue, will not bother to read what I have set forth; but if it stops any in­tel­li­gent folk from wast­ing fur­ther time, that too is a benefit.

I am some­times ac­cused of be­ing over­con­fi­dent and opinionated, for tel­ling peo­ple that be­ing com­posed of “the same atoms” has noth­ing to do with their per­sonal con­ti­nu­ity. Or for say­ing that an up­load­ing scan performed to the same re­s­olu­tion as ther­mal noise, ac­tu­ally has less effect on your iden­tity than a sneeze (be­cause your eyes squeeze shut when you sneeze, and that ac­tu­ally al­ters the com­pu­ta­tional state of billions of neu­rons in your vi­sual cor­tex). Yet if you can see your nows braided into causal­ity of the river that never flows; and the synap­tic con­nec­tions com­put­ing your in­ter­nal nar­ra­tive, that re­main the same from one time to an­other, though not a drop of wa­ter is shared; then you can see that I have rea­sons for this strong be­lief as well.

Per­haps the one says to me that the ex­act du­pli­cate con­structed on Mars, is just a copy. And I post a short com­ment say­ing, “Wrong. There is no copy, there are two origi­nals. This is know­able and I know it.” Would you have thought that I might have very strong sup­port, that you might not be see­ing?

I won’t always have the time to write a month of blog posts. While I am enough of a Tra­di­tional Ra­tion­al­ist that I dis­like trust, and will not lightly ask it, I may ask it of you if your life is at stake.

Another one once asked me: “What does quan­tum physics have to do with over­com­ing bias?”

Robin Han­son chose the name “Over­com­ing Bias”; but names are not steel chains. If I’d started my own per­sonal blog for the ma­te­rial I’m now post­ing, I would have called it “Rein­vent­ing Ra­tion­al­ity” or some­thing along those lines—and even that wouldn’t have been the real pur­pose, which would have been harder to ex­plain.

What are these se­ries of posts, re­ally? Raw ma­te­rial for a pop­u­lar book on ra­tio­nal­ity—but maybe a tenth of this ma­te­rial, or less, will make it into the book. One of the rea­sons I write long posts, is so that I can shorten them later with a good con­science, know­ing that I did lay out the full ar­gu­ment some­where. But the whole quan­tum physics se­quence is prob­a­bly not go­ing to make it into the pop­u­lar book at all—and nei­ther will many other posts. So what’s the rest of it for?

Some­times I think wist­fully of how much more I could have ac­com­plished in my teenage years, if I had known a frac­tion of what I know now at age 15. (This is the age at which I was a Tra­di­tional Ra­tion­al­ist, and ded­i­cated and bright as such ones go, but knew not the Way of Bayes.) You can think of these blog posts, per­haps, as a se­ries of let­ters to my past self. Only not ex­actly, be­cause some of what I now write, I did already know then.

It seems to me, look­ing back, that the road which took me to this Way, had a great deal of luck in it. I would like to elimi­nate that el­e­ment of luck for those who come af­ter. So some of what I post, is more for­mal ex­pla­na­tions of mat­ters which Eliezer-15 knew in his bones. And the rest, I only wish I had known.

Per­haps if you prime some­one with enough ma­te­rial as a start­ing point, they can figure out the other 95% on their own, if they go on to study the rele­vant sci­ences at a higher tech­ni­cal level. That’s what I hope.

Eliezer-15 was led far astray by the seem­ing mys­te­ri­ous­ness of quan­tum me­chan­ics. An an­tipro­ject in which he was aided and abet­ted by cer­tain pop­u­lar physi­cists—no­tably Sir Roger Pen­rose; but also all those physi­cists who told him that quan­tum physics was “mys­te­ri­ous” and that it was okay not to un­der­stand it.

This is some­thing I wish I had known, so I ex­plained it to me.

Why not just tell me to ig­nore quan­tum physics? Be­cause I am not go­ing to “just ig­nore” a ques­tion that large. It is not how we work.

If you are con­fronting real sci­en­tific chaos—not just some light mat­ter of an ex­per­i­men­tal anomaly or the search for a bet­ter the­ory, but gen­uine fear and de­spair, as now ex­ists in Ar­tifi­cial In­tel­li­gence—then it is nec­es­sary to be a poly­math. Healthy fields have healthy ways of think­ing; you can­not trust the tra­di­tions of the con­fused field you must re­form, though you must learn them. You never know which way you’ll need to draw upon, on ven­tur­ing out into the un­known. You learn new sci­ences for the same rea­sons that pro­gram­mers learn new pro­gram­ming lan­guages: to change the way you think. If you want to never learn any­thing with­out know­ing in ad­vance how it will ap­ply, you had best stay away from chaos.

If you want to tackle challenges on the or­der of AI, you can’t just learn a bunch of AI stuff.

And fi­nally...


There fi­nally comes a point where you get tired of try­ing to com­mu­ni­cate across vast in­fer­en­tial dis­tances. There comes a point where you get tired of not be­ing able to say things to peo­ple with­out a month of pre­limi­nary ex­pla­na­tion. There comes a point where you want to say some­thing about branch­ing Earths or iden­ti­cal par­ti­cles or braids in the river that never flows, and you can’t.

It is such a tremen­dous re­lief, to fi­nally be able to say all these things. And all the other things, that I have said here; that peo­ple have asked me about for so long, and all I could do was wave my hands. I didn’t have to ex­plain the con­cept of “in­fer­en­tial dis­tance” from scratch, I could just link to it. It is such a re­lief.

I have writ­ten hun­dreds of blog posts here. Think about what it would be like, to carry all that around in­side your head.

If I can do all the long se­quences on Over­com­ing Bias, then maybe af­ter that, it will be pos­si­ble to say most things that I want to say, in just one piece.

Part of The Quan­tum Physics Sequence

(end of se­quence)

Pre­vi­ous post: “Class Pro­ject