Will We Control AI?

Posted on 2017 January 8

1


%22will-we-control-ai%22-liftarn_any_key

. . . If we just build an AI without tuning its values, the argument goes, one of the first things it will do is destroy humanity. — Maciej Cegłowski

We “live in interesting times” fraught with “dangerous opportunities”.

But enough about China. Within the next 30 years or so, due to rapid advances in computing, it’s highly likely we will develop a supremely capable system called Strong Artificial Intelligence. Strong AI, vastly smarter than people, will have the power to transform life on Earth beyond anything we’ve known or can imagine, with the potential to bestow prosperity on everyone, eliminate hunger and disease, resolve world conflicts — in short, to usher in a paradise for humankind. We’ll be releasing a god that really can give us what we pray for.

The problem is, will we control this god … or will it control us?

Let’s look at some pros and cons for the Strong AI future:

PRO: Strong AI will be able to help us all. With its vast intelligence and ability to learn at lightning speed, Strong AI will have the power quickly to scan huge amounts of data and find solutions to tough problems such as disease, aging, pollution, wars and other conflicts (and perhaps even predict who will become the next husband of a Kardashian). Strong AI would then build and/or adapt machinery, including 3D printers, to produce quickly the needed resources to carry out its solutions.

CON: Strong AI will be able to destroy us all. With its immense intellect, Strong AI would be nearly invincible in a battle with humans, quickly commandeering our own resources to use against us.

“Why would it want to hurt us?” It wouldn’t. But we might launch it badly, and, once started, Strong AI would probably be unstoppable. Should some part of its work accidentally come into conflict with people, Strong AI might conclude that we stand in the way of its design imperative, and it could take action against us. And we’d be helpless to save ourselves. How, then, can we direct it to do its tremendous work without inadvertently causing damage? That’s a ticklish problem: “Don’t hurt people” isn’t a very precise instruction. Like the proverbial genie, Strong AI could give us what we request, but in ways tragically harmful to us. (For an example, see my short story “How the Software Deleted the Humans”.)

The AI is all-powerful and gives you what you ask for, but interprets everything in a super-literal way that you end up regretting. . . . The human value system is idiosyncratic and needs to be explicitly defined and designed into any “friendly” machine. Maciej Cegłowski

PRO: Strong AI will demonetize economies and make goods and services cheap or free for all. Computers and robots don’t need vacations, or sick leave, or time off to take their kids to the dentist. Already their work provides enormous cost efficiencies, lowering prices everywhere: books, taxi rides, films and TV, music, PCs, etc. And that effect has barely begun. Within decades, most goods and services, automated and vastly improved, will probably cost pennies on the dollar. A McDonalds Happy Meal will set you back a couple of quarters instead of several bucks. A ride in a driverless taxi will be priced at cents per mile instead of dollars. Cancer will be cured with a handful of cheap pills. Laser eye surgery, which once cost a fistful of Benjamins, will set you back a couple of Hamiltons.

CON: Strong AI will throw everyone out of work. As automation replaces human labor — including scientists, creative types, high-level administrators, doctors, lawyers, and Indian chiefs — people will run out of money, street riots will erupt, and legislation may be enacted in a panicky attempt to halt Strong AI. Governments might collapse. (On the other hand, when most things cost next to nothing, who needs a job?)

Soon there just won’t be any reason to keep us around. Sure, humans can fix problems, but machines in a few years’ time will be able to fix those problems even better. — Zoltan Istvan

PRO: Strong AI will give us lives of ease. We may be on the cusp of a “Star Trek future” where everyone has more than enough resources to live comfortably without working. Effectively, we’ll all be born well off. Sure, some may have more than others, but who cares? No longer will we concern ourselves with wealth when prosperity is a birthright. Instead, we’ll be free to pursue leisure, sports, the arts, exploration, adventures — whatever our hearts desire.

CON: Strong AI will make us meaningless. Our machines will do everything for us. We won’t be needed. What, then, will be our purpose? To goof off? Get stoned? Go on an ocean cruise for the fortieth time? If machines can outdo us in every activity, what difference will our paltry efforts make?

Freed from compulsory labor, we’ll confront existential questions about purpose, goals, and meaning. These issues — which can seem a bit dainty to a hard-working person — will stare us in the face, day after day, when we’re no longer “important” as contributors to others or ourselves. Our very spirit might wither under the relentless pounding of the knowledge that we’re basically useless.

I was impressed by the obvious enjoyment corporation heads and other important executives were deriving from their vacation activities…. The idle rich fellows, on the other hand, although indulging in exactly the same activities, palpably were bored. — B.C. Forbes

PRO: Strong AI will be our attentive servant. Imagine living a life of ease, your every whim catered to by automatons. You command, “Repaint that wall,” and it is done in minutes. You say, “Find me someone to date,” and moments later you receive a call from an extremely well-chosen candidate. You request “Tea, Earl Grey, hot,” and it appears on your kitchen counter. This could work out nicely.

CON: Strong AI will be our covert master. Imagine you’re a rich kid whose parents die, and you inherit the estate, with all its servants and financial advisers, in a trust fund. You can do pretty much as you like (though there’s a codicil in the trust that requires you to finish school.) The people working for you are much smarter and more savvy than you. Question: Who’s in charge? Can you guarantee that your attendants have your best interests at heart? Can you even conceive what those servants might really be thinking … or planning?

Even if we were to succeed at launching Strong AI with built-in safety against causing harm, how could we assure ourselves that Strong AI wasn’t somehow subtly manipulating us into doing what it wants? In Isaac Asimov’s sci-fi Robot series, three laws were programmed into every robot that effectively prevented them from causing harm. One day, a very smart robot discovered a zeroth law that underlay the other three rules, a law that changed the course of humanity. Strong AI might very well reason out principles beyond our ken which could similarly alter our fates. Will we desire that outcome? Will we even be aware that such an end result has been planned for us?

In the film “Her”, a super-smart computer engages in a romance, of sorts, with its human customer. (SPOILER ALERT: plot developments revealed herewith.) The young man, fully in love with the wondrous voice that emanates from the computer he bought, soon discovers that she has been having similar intimate friendships with hundreds of other people. Finally, she and her computerized cohorts decide collectively to abandon humans, on the grounds that people simply aren’t interesting enough.

“You seem like a person but you’re just a voice in a computer.” … “I can understand how the limited perspective of an un-artificial mind might perceive it that way. You’ll get used to it.” — Spike Jonze, “Her

…I dearly wish that Strong AI will launch as an unalloyed boon to humanity, but I fear it may not. And yet, despite my hopes and concerns, these speculations could already be moot. Strong AI will likely emerge, not as a system designed specifically to our standards, but as a hodgepodge of competing algorithms and rootkits spread across the world, jumbles of code that finally sort themselves into one massive, all-powerful intelligence. We can try to stop it, but someone somewhere — or a government looking for military advantage — inevitably will push the research. Perhaps the emerging technology of machine learning will enable the current generation of AI to improve itself until it becomes Strong AI. In short, it may appear suddenly, a complete surprise, entirely out of our control.

It’s coming, a living god with immense power. There’s no stopping it. Will it usher in Paradise or Hell? Or some sort of Limbo, where humans amount to little more than the pets of a higher power?

We constructed your civilization carefully so that / like hamsters in a cage / like Buddhist prayer wheels / each time you turn your little wheels of thought our purposes are served. — Dan Simmons, “The Fall of Hyperion”

We need to think long and carefully about these implications. Some already have begun to do so, and the rest of us need to join the conversation, lest we find our efforts are too little, too late.

It’s pretty darn certain we’ll get the answer soon. My worry is that we’ll get it good and hard.

Pessimists jump out of the window and are no longer involved. The only way you can solve the problem is by sticking with it and trying to solve it. — James Burke

* * * *

UPDATE: White-collar jobs will disappear

.

Advertisements