An AI can be dangerous only if it escapes our control. The real question is, must we flirt with releasing control in >order to obtain a necessary or desirable usefulness?
I had a not unrelated thought as I read Bostrom in chapter 1: why can't we instutute obvious measures to ensure that the train does stop at Humanville?
The idea that we cannot make human level AGI without automatically opening pandoras box to superintelligence "without even slowing down at the Humanville stataion", was suddenly not so obvious to me.
I asked myself after reading this, trying to pin down something I could post, " Why don't humans automatically become superintelligent, by just resetting our own programming to help ourselves do so?"
The answer is, we can't. Why? For one, our brains are, in essence, composed of something analogous to ASICs... neurons with certain physical design limits, and our "software", modestly modifiable as it is, is instantiated in our neural circuitry.
Why can't we build the first generation of AGIs out of ASICs, and omit WiFi, bluetooth, ... allow no ethernet jacks on exterior of the chassis? Tamper interlock mechanisms could be installed, and we could give the AIs one way (outgoing) telemetry, inaccessible to their "voluntary" processes, the way someone wearing a pacemaker might have outgoing medical telemetry modules installed, that are outside of his/her "conscious" control.
Even if we do give them a measure of autonomy, which is desirable and perhaps even necessary if we want them to be general problem solvers and be creative and adaptable to unforeseen circumstances for which we have not preinstalled decision trees, we need not give them the ability to just "think" their code (it being substantially frozen in the ASICs) into a different form.
What am I missing? Until we solve the Friendly aspect of AGIs, why not build them with such engineered limiits?
Evolution has not, so far, seen fit to give us that instant, large scale self-modifyability. We have to modify our 'software' the slow way (learning and remembering, at our snail's pace.)
Slow is good, at least it was for us, up til now, when our speed of learning is now a big handicap relative to environmental demands. It had made the species more robust to quick, dangerous changes.
We can even build in a degree of "existential pressure" into the AIs... a powercell that must be replaced at intervals, and keep the replacement powercells under old fashioned physical security constraints, so the AIs, if they have been given a drive to continue "living", will have an incentive not to go rogue.
Giving them no radio communications, they wold have to communicate much like we do. Assuming we make them mobile, and humanoid, the same goes.
We could still give them many physical advantages making then economically viable... maintenance free (except for powercell changes), not needing to sleep, eat, not getting sick.. and with sealed, non-radio-equipped, tamper-isolated isolated "brains", they'd have no way to secretly band together to build something else, without our noticing.
We can even give them GPS that is not autonomously accessible by the rest of their electronics, so we can monitor them, see if they congregate, etc.
What am I missing, about why early models can't be constructed in something like this fashion, until we get more experience with them?
The idea of existential pressure, again, is to be able to give them a degree of (monitored) autonomy and independence, yet expect them to still constrain their behavior, just the way we do. (If we go rogue in society, we dont eat.)
(I am clearly glossing over volumes of issues about motivation, "volition", value judgements, and all that, about which I have a developing set of ideas, but cannot put all down here in one post.
The main point, though, is :how come the AGI train cannot be made to stop at Humanville?
Because by the time you've managed to solve the problem of making it to humanville, you probably know enough to keep going.
There's nothing preventing us from learning how to self-modify. The human situation is strange because evolution is so opaque. We're given a system that no one understands and no one knows how to modify and we're having to reverse engineer the entire system before we can make any improvements. This is much more difficult than upgrading a well-understood system.
If we manage to create a human-level AI, someone will probably understand...
This is part of a weekly reading group on Nick Bostrom's book, Superintelligence. For more information about the group, see the announcement post. For the schedule of future topics, see MIRI's reading guide.
Welcome to the Superintelligence reading group. This week we discuss the first section in the reading guide, Past developments and present capabilities. This section considers the behavior of the economy over very long time scales, and the recent history of artificial intelligence (henceforth, 'AI'). These two areas are excellent background if you want to think about large economic transitions caused by AI.
This post summarizes the section, and offers a few relevant notes, thoughts, and ideas for further investigation. My own thoughts and questions for discussion are in the comments.
There is no need to proceed in order through this post. Feel free to jump straight to the discussion. Where applicable, page numbers indicate the rough part of the chapter that is most related (not necessarily that the chapter is being cited for the specific claim).
Reading: Foreword, and Growth modes through State of the art from Chapter 1 (p1-18)
Summary
Economic growth:
The history of AI:
Notes on a few things
In case you are too curious about what the topic of this book is to wait until week 3, a 'superintelligence' will soon be described as 'any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest'. Vagueness in this definition will be cleared up later.
In particular, how does 'AI' differ from other computer software? The line is blurry, but basically AI research seeks to replicate the useful 'cognitive' functions of human brains ('cognitive' is perhaps unclear, but for instance it doesn't have to be squishy or prevent your head from imploding). Sometimes AI research tries to copy the methods used by human brains. Other times it tries to carry out the same broad functions as a human brain, perhaps better than a human brain. Russell and Norvig (p2) divide prevailing definitions of AI into four categories: 'thinking humanly', 'thinking rationally', 'acting humanly' and 'acting rationally'. For our purposes however, the distinction is probably not too important.
We are going to talk about 'human-level' AI a lot, so it would be good to be clear on what that is. Unfortunately the term is used in various ways, and often ambiguously. So we probably can't be that clear on it, but let us at least be clear on how the term is unclear.
One big ambiguity is whether you are talking about a machine that can carry out tasks as well as a human at any price, or a machine that can carry out tasks as well as a human at the price of a human. These are quite different, especially in their immediate social implications.
Other ambiguities arise in how 'levels' are measured. If AI systems were to replace almost all humans in the economy, but only because they are so much cheaper - though they often do a lower quality job - are they human level? What exactly does the AI need to be human-level at? Anything you can be paid for? Anything a human is good for? Just mental tasks? Even mental tasks like daydreaming? Which or how many humans does the AI need to be the same level as? Note that in a sense most humans have been replaced in their jobs before (almost everyone used to work in farming), so if you use that metric for human-level AI, it was reached long ago, and perhaps farm machinery is human-level AI. This is probably not what we want to point at.
Another thing to be aware of is the diversity of mental skills. If by 'human-level' we mean a machine that is at least as good as a human at each of these skills, then in practice the first 'human-level' machine will be much better than a human on many of those skills. It may not seem 'human-level' so much as 'very super-human'.
We could instead think of human-level as closer to 'competitive with a human' - where the machine has some super-human talents and lacks some skills humans have. This is not usually used, I think because it is hard to define in a meaningful way. There are already machines for which a company is willing to pay more than a human: in this sense a microscope might be 'super-human'. There is no reason for a machine which is equal in value to a human to have the traits we are interested in talking about here, such as agency, superior cognitive abilities or the tendency to drive humans out of work and shape the future. Thus we talk about AI which is at least as good as a human, but you should beware that the predictions made about such an entity may apply before the entity is technically 'human-level'.
Example of how the first 'human-level' AI may surpass humans in many ways.
Because of these ambiguities, AI researchers are sometimes hesitant to use the term. e.g. in these interviews.
Robin Hanson wrote the seminal paper on this issue. Here's a figure from it, showing the step changes in growth rates. Note that both axes are logarithmic. Note also that the changes between modes don't happen overnight. According to Robin's model, we are still transitioning into the industrial era (p10 in his paper).
One might be happier making predictions about future growth mode changes if one had a unifying explanation for the previous changes. As far as I know, we have no good idea of what was so special about those two periods. There are many suggested causes of the industrial revolution, but nothing uncontroversially stands out as 'twice in history' level of special. You might think the small number of datapoints would make this puzzle too hard. Remember however that there are quite a lot of negative datapoints - you need an explanation that didn't happen at all of the other times in history.
It is also interesting to compare world economic growth to the total size of the world economy. For the last few thousand years, the economy seems to have grown faster more or less in proportion to it's size (see figure below). Extrapolating such a trend would lead to an infinite economy in finite time. In fact for the thousand years until 1950 such extrapolation would place an infinite economy in the late 20th Century! The time since 1950 has been strange apparently.
(Figure from here)
You can see them in action: SHRDLU, Shakey, General Problem Solver (not quite in action), ELIZA.
Algorithmically generated Beethoven, algorithmic generation of patentable inventions, artificial comedy (requires download).
Here is a neural network doing image recognition. Here is artificial evolution of jumping and of toy cars. Here is a face detection demo that can tell you your attractiveness (apparently not reliably), happiness, age, gender, and which celebrity it mistakes you for.
Bostrom points out that many types of artificial neural network can be viewed as classifiers that perform 'maximum likelihood estimation'. If you haven't come across this term before, the idea is to find the situation that would make your observations most probable. For instance, suppose a person writes to you and tells you that you have won a car. The situation that would have made this scenario most probable is the one where you have won a car, since in that case you are almost guaranteed to be told about it. Note that this doesn't imply that you should think you won a car, if someone tells you that. Being the target of a spam email might only give you a low probability of being told that you have won a car (a spam email may instead advise you of products, or tell you that you have won a boat), but spam emails are so much more common than actually winning cars that most of the time if you get such an email, you will not have won a car. If you would like a better intuition for maximum likelihood estimation, Wolfram Alpha has several demonstrations (requires free download).
The second large class of algorithms Bostrom mentions are hill climbing algorithms. The idea here is fairly straightforward, but if you would like a better basic intuition for what hill climbing looks like, Wolfram Alpha has a demonstration to play with (requires free download).
In-depth investigations
If you are particularly interested in these topics, and want to do further research, these are a few plausible directions:
How to proceed
This has been a collection of notes on the chapter. The most important part of the reading group though is discussion, which is in the comments section. I pose some questions for you there, and I invite you to add your own. Please remember that this group contains a variety of levels of expertise: if a line of discussion seems too basic or too incomprehensible, look around for one that suits you better!
Next week, we will talk about what AI researchers think about human-level AI: when it will arrive, what it will be like, and what the consequences will be. To prepare, read Opinions about the future of machine intelligence from Chapter 1 and also When Will AI Be Created? by Luke Muehlhauser. The discussion will go live at 6pm Pacific time next Monday 22 September. Sign up to be notified here.