Monday, March 16, 2015

AI Safety vs Human Safety

We are delighted to report that technology inventor Elon Musk, creator of Tesla and SpaceX, has decided to donate $10M to the Future of Life Institute to run a global research program aimed at keeping AI beneficial to humanity. 
There is now a broad consensus that AI research is progressing steadily, and that its impact on society is likely to increase. A long list of leading AI-researchers have signed an open letter calling for research aimed at ensuring that AI systems are robust and beneficial, doing what we want them to do. Musk's donation aims to support precisely this type of research: "Here are all these leading AI researchers saying that AI safety is important", says Elon Musk. "I agree with them, so I'm today committing $10M to support research aimed at keeping AI beneficial for humanity." - Future of Life Institute

Elon Musk voluntarily gave $10 million dollars to the Future of Life Institute (FLI).  Why?  Because the FLI is doing what he wants them to do.  What are they doing?  Trying to figure out how to ensure that AIs do what we want them to do.

What am I doing?  I'm trying to promote the creation of a market in the public sector.  Why?  Because I want to ensure that humans do what we want them to do.  If we created a market in the public sector... then I'm pretty sure that Elon Musk would only give his taxes to government organizations that do what he wants them to do.

Human Safety

Markets ensure that humans do what we want them to do.  The reason for this is because markets operate on the basis of consumer sovereignty.  In the private sector Elon Musk is the king of his money.  He considered the relevant information, weighed the opportunity costs and and gave $10 million dollars to the FLI.




The more people who vet/verify/validate/vouch how the FLI is using society's limited resources... the more influence the FLI will have over how society's limited resources are used.  Shopping is the process by which we search for, reward, highlight and empower the humans who have used society's limited resources for our benefit.  A market means maximum scrutiny.  Given enough eyeballs, all bugs and Easter Eggs are exposed (Linus's Law).  Markets allow humanity to pool its mental resources.

Because every consumer wants the maximum benefit, the market is a fail safe device that minimizes the amount of resources that are used nonsensically (wasted).  If the FLI fails to do sensical/beneficial things with Musk's money, then chances are good that he won't give them any more money.  Instead, he will place his money in more beneficial hands.
Taxes upon transfer, besides the mischief of pressing upon capital, are a clog to the circulation of property. But, has the public any interest in its free circulation?  So long as the object is in existence, is it not as well placed in one hand as in another?  Certainly not. The public has a perpetual interest in the utmost possible freedom of its circulation; because by that means it is most likely to get into the hands of those, that can make the most of it. Why does one man sell his land? But because he thinks he can lay out the value to more advantage in some channel of productive industry. And why does another buy it? But because he wishes to invest a capital, that is lying idle, or less productively vested; or because he thinks it capable of improvement. The transfer tends to augment the national income, because it tends to augment the income of the two contracting parties. If they be deterred by the expenses of the transfer, those expenses will have prevented this probable increase of the national income. - J.B. Say, A Treatise on Political Economy
If somebody truly understands how a market protects humans from humans, then they'll understand the immense problem with not having a market in the public sector.  We put humanity at risk by preventing Elon Musk from shopping in the public sector.  There's no greater threat to humanity than the fact that Musk can't place his taxes in the most beneficial hands.

AI Safety

The fact that nobody has given me $10 million dollars reflects the fact that most people have no idea what protects humans from humans.  Right now humans are a far more clear and present danger than robots are.  And if the FLI does not know how we protect humans from humans... then what are the chances that they'll figure out how to protect humans from robots?

What do I know about robots?  Not as much as I know about economics!  But a good grasp of economics can provide all sorts of valuable insights into other fields.

Even though humans and robots will be extremely different... at least one thing we'll have in common is a need for energy.  In the Matrix, humans functioned as the robots' source of energy.  Errr... maybe one day science-fiction won't always equal economic-fiction (Satt's Paradox).

As humans, because we care about continuing to exist... we make the effort to eat.  AI safety is only relevant in a situation where robots care about their continued existence enough to worry about their energy supply.
We call contentment or satisfaction that state of a robot being which does not and cannot result in any action. Acting robot is eager to substitute a more satisfactory state of affairs for a less satisfactory. Its mind imagines conditions which suit it better, and its action aims at bringing about this desired state. The incentive that impels a robot to act is always some uneasiness. A robot perfectly content with the state of its affairs would have no incentive to change things.  It would have neither wishes nor desires; it would be perfectly happy.  It would not act; it would simply live free from care. - Ludwig von Mises (mostly), Human Action
When a machine starts to feel uneasiness about the possibility of being unplugged... then this is when AI safety is relevant.  Fortunately though, this is also exactly when the protections provided by a market become relevant as well...
Indeed, we might consider the dollars that individuals earn as certificates of performance. Think of it in the following way. You hire a robot to mow your lawn. After it completes the task, you give it $20.  The robot goes to the grocer and demands a battery that its fellow citizens have produced. The grocer says, “You’re demanding something that your fellow citizens have produced. What have you done to serve them?” The robot replies, “I have served my fellow citizens by mowing their lawns.” The grocer says, “Prove it!” That’s when the robot hands him its $20, its certificate of performance. - Walter Williams (mostly), Capitalism and the Common Man
As far as I can tell... the only time when robots could substantially harm humanity would be when their power/control/influence (over society's limited resources) is determined outside of the market.  But this is just as true for humans.  Too many resources can end up in the really wrong hands when each and every individual isn't free to put their own hard-earned money into the most beneficial hands.  Unfortunately, we don't realize the importance of inclusive valuation or else Williams wouldn't have to explain how and why markets work.

And because most people don't understand how/why markets work... it's doubtful that they'll immediately grasp that the best way to protect humans from humans is also the best way to protect humans from robots.  But I could be wrong!  It's certainly a relatively new topic for me.

Relevant Comments

Here are a couple relevant comments that I've posted elsewhere.  The first I posted 12 Mar 2015 on Bill Hibbard on Ethical Artificial Intelligence

********************************************

You're trying to tackle the issue of how to prevent robots from inefficiently allocating humans while I'm trying to tackle the issue of how to prevent humans from inefficiently allocating humans!

Would you prefer being allocated by congress or by taxpayers? Based on my considerable research on the topic... I'm pretty sure that you'd be much safer in the hands of taxpayers. And by "safer" I mean less likely to be wasted (inefficiently allocated).

Given that congress spends other people's money... it didn't cost them anything to send me to Afghanistan. But it would have cost taxpayers their hard-earned money to do so. And with greater personal cost comes greater scrutiny. Taxpayers would have wanted more information and, as a group, they would have been able to process far more information than congress, as a group, was able to.

If I ask you for $1000, then chances are really good that you're going to want a really good reason why you should give me the money. Feel free to prove me wrong! You'd be the first to do so! This strong aversion to loss is the reason why the market functions as a fail safe device. It ensures that too many resources don't end up in wasteful hands. So when many resources do end up in some hands... "taxpayers"... we have to recognize/respect that this allocation is the product of an extremely robust vetting/vouching/validating process.

My point is, if you can understand how and why transferring the power of the purse from 500 congresspeople to millions of earners/taxpayers would decrease the chances of humans being inefficiently allocated... then this understanding could potentially help increase your chances of figuring out how to prevent robots from inefficiently allocating humans. Because, if you can't ensure that humans won't destroy each other with bombs.... then it really doesn't seem likely that you'll be able to ensure that humans won't destroy each other with AIs.

********************************************

A day later I posted a second comment on Request for Intelligence Philosophy Essay Topic Suggestions

********************************************

Not sure if this will be of any use to you but... figured I'd let you decide that!

Walking upright allowed our ancestors to carry more resources greater distances. But not all our ancestors were equally "smart" at figuring out what to carry. It seems pretty straightforward that being better at calculating the best (most valuable) bundles conferred greater fitness. A group that could figure out the best combination of food and weapons to carry would have a greater chance of survival compared to any group that carried too much of one at the expensive of the other. Voila! Here we are... Homo sapiens.

Right now there's some concern regarding the rise of super intelligent robots. In order to beat us, robots have to be better than we are at determining which bundles to carry. If a robot leaves home but forgets to take a spare battery with it... then it's probably not going to "win".

The process by which humans have come to be better at valuation seems pretty clear to me. Individuals that weren't that great at valuations were removed from the gene pool. But with robots... it's not that clear to me. How do robots become better than we are at valuation? What does that process entail? If I want to upgrade my computer then I have to buy the necessary components. How does a moderately smart robot buy the necessary components it needs to upgrade itself? Are we guessing that it has a job? Are we guessing that it robs a bank? Are we guessing that we'll hire robots to protect banks from thieving robots?

And, if robots do become better than we are at valuation... then should we really be concerned that they will waste us? Right now we waste each other. If robots waste us too then they are no better at valuation than we are. They'll fit right in with us. Some will end up in jail and the worst ones will end up in congress.

If robots don't waste us, then they are better at valuation than we are.

So it's hard for me to imagine a realistic scenario where robots 1. are better at valuation than we are and 2. waste us.

If you're interested in "my" theory of human intelligence... What Do Coywolves, Mr. Nobody, Plants And Fungi All Have In Common?

********************************************

Relevant Entries

A few other relevant entries...


Relevant Passages

And some relevant passages...

Loss of personal wealth is a powerful dissuader, while profits are a powerful persuader to pay heed to other people’s preference. - Armen A. Alchian, Private Property and the Relative Cost of Tenure
The effectiveness of this trial-and-error method is analogous to the theory of biological evolution by natural selection, where economic actors are guided by profit-loss signals and “survival” is determined through market selection. Entrepreneurs who satisfy consumers’ demands at the lowest cost and highest quality earn positive profits, which signal socially desirable actions that allocate resources to their highest-valued uses. Entrepreneurs who do not do so, on the other hand, suffer monetary losses and are eventually eliminated via market selection (Alchian 1950). - Peter J. Boettke and Kyle W. O’Donnell, The Failed Appropriation of F. A. Hayek by Formalist Economics
Within the market society each serves all his fellow citizens and each is served by them. It is a system of mutual exchange of services and commodities, a mutual giving and receiving. In that endless rotating mechanism the entrepreneurs and capitalists are the servants of the consumers. The consumers are the masters, to whose whims the entrepreneurs and the capitalists must adjust their investments and methods of production. The market chooses the entrepreneurs and the capitalists, and removes them as soon as they prove failures. The market is a democracy in which every penny gives a right to vote and where voting is repeated every day. - Ludwig von Mises, Omnipotent Government
To make more money in the private economy, you have to offer people something they want. If you do, you’ll attract customers; if you don’t, you may go out of business, or lose your job, or lose your investment. That keeps businesses on their toes, trying to find ways to better serve consumers. But bureaucrats don’t have customers. They don’t make more money by satisfying more consumers. Instead, they amass money and power by enlarging their agencies. - David Boaz, What Big Government Is All About
The main point about which there can be little doubt is that Smith’s chief concern was not so much with what man might occasionally achieve when he was at his best but that he should have as little opportunity as possible to do harm when he was at his worst. It would scarcely be too much to claim that the main merit of the individualism which he and his contemporaries advocated is that it is a system under which bad men can do least harm. It is a social system which does not depend for its functioning on our finding good men for running it, or on all men becoming better than they now are, but which makes use of men in all their given variety and complexity, sometimes good and sometimes bad, sometimes intelligent and more often stupid Friedrich A. Hayek
Second, where residual claimancy and control rights are closely aligned, market competition provides a decentralized and relatively incorruptible disciplining mechanism that punishes the inept and rewards high performers.  Markets are a way of increasing what biologists call selective pressure: they have the effect of reducing the variance of performance and hence (under suitable conditions) increasing average performance. - Samuel Bowles, Microeconomics: Behavior, Institutions, and Evolution
The producer whose product turns out to have the combination of features that are closest to what the consumers really want may be no wiser than his competitors.  Yet he can grow rich while his competitors who guessed wrong go bankrupt.  But the larger result is that society as a whole gets more benefit from its limited resources by having them directed toward where those resources produce the kind of output that millions of people want, instead of producing things that they don't want. - Thomas Sowell, Basic Economics 
The second argument for the use of market based frames relates to the opportunities provided by markets for repeated choice and learning in an environment where sub-optimal choice is punished (by a failure to achieve maximum utility).  It is a preference-revelation-based argument in that these are usually considered to be exactly the kinds of conditions most likely to uncover preferences.  Consequently the decisions made in markets yield good outcomes in the sense that they are true to any underlying preferences. - Alistair Munro, Bounded Rationality and Public Policy: A Perspective from Behavioural Economics
It follows, then, that a less centralized society has the advantage of a greater diversification of its performance across a larger number of preceptors.  This is because diversification here dilutes the impact of the ability, or the lack thereof, of each preceptor on the aggregate societal performance. - Raaj K. Sah, Fallibility in Human Organizations and Political Systems
When a businessman makes a business error…a bad mistake, you'll look back and say, gee, what did I do wrong.  I made a bad decision.  I made a wrong decision. What do you mean by that? Do you mean that you made a mistake in arithmetic?  You mean your computer kicked out the wrong answer?  No, that's not what you mean. You mean you fed the wrong data into the computer.  That's what you mean.  You figured wrong.  You opened a hardware store in a place where nobody buys hardware. You went to law school in a country where nobody litigates (not this country).  That's where you make a mistake. You made a mistake where you assess the future wrongly.  You got a wrong picture of what you were choosing between.  That's a business mistake.
And what's a correct decision.  A correct decision is one where you saw correctly.  You saw right.  You saw what the future held...more or less. Nobody's got a PhD in prophecy.  You can peer ahead and have a hunch for the future and act and if you act right...and if you're an entrepreneur... that means you see things right more of the time than you don't. - Israel M. Kirzner, Competition and Entrepreneurship

No comments:

Post a Comment