We've seen random natural selection significantly improve human intelligence in as few as tens of generations.
"Random natural selection" is almost a contradiction in terms. Yes, we've seen dramatic boosts in Ashkenazi intelligence on that timescale, but that's due to very non-random selection pressure.
Mutations occur randomly and environmental pressure perform selection on them.
Demis Hassabis gives a great presentation on the state of Deepmind's work as of April 20, 2016. Skip to 23:12 for the statement of the goal of creating a rat-level AI -- "An AI that can do everything a rat can do," in his words. From his tone, it sounds like this is more a short-term, not a long-term goal.
I don't think Hassabis is prone to making unrealistic plans or stating overly bold predictions. I strongly encourage you to scan through Deepmind's publication list to get a sense of how quickly they're making progress. (In fact, I encourage you to bookmark that page, because it seems like they add a new paper about twice a month.) The outfit seems to be systematically knocking down all the "Holy Grail" milestones on the way to GAI, and this is just Deepmind. The papers they've put out in just the last year or so concern successful one-shot learning, continuous control, actor-critic architectures, novel memory architectures, policy learning, and bootstrapped gradient learning, and these are just the most stand-out achievements. There's even a paper co-authored by Stuart Armstrong concerning Friendliness concepts on that list.
If we really do have a genuinely rat-level AI within the next couple of years, I think that would justify radically moving forward expectations of AI development timetables. Speaking very naively, if we can go from "sub-nematode" to "mammal that can solve puzzles" in that timeframe, I would view it as a form of proof that "general" intelligence does not require some mysterious ingredient that we haven't discovered yet.
I've read Age of Em. IIRC, Robin argues that it will be likely be difficult to get a lot of progress from evolutionary algorithms applied to emulations because brains are fragile to random changes, and we don't understand brains well enough to select usefully nonrandom changes, so all changes we make are likely to be negative.
But brains actually seem to be surprisingly resilient, given that many people with brain damage or deformed brains are still functional, including such dramatic changes as removing a hemisphere of the brain, and there are already known ways to improve brain performance in narrow domains with electric stimulation (source), which seems similar. So it seems fairly likely to me that getting significant improvements from evolutionary algorithms is possible.
Also, I talked with Robin about this, and he didn't actually seem very confident about his prediction that evolutionary algorithms would not be used to increase the intelligence of emulations significantly, but he did think that such enhancements would not have a dramatic effect on em society.
Hanson makes so many assumptions that defy intuition. He's talking a civilization with the capacity to support trillions of individuals, in which these individuals are largely entirely disposable and can be duplicated at a moment's notice, and he doesn't think evolutionary pressures are going to come into play? We've seen random natural selection significantly improve human intelligence in as few as tens of generations. With Ems, you could probably cook up tailor-made superintelligences in a weekend using nothing but the right selection pressures. Or, at least, I see no reason to be confident in the converse proposition.
He claims we don't know enough about the brain to select usefully nonrandom changes, yet assumes that we'll know enough to emulate them to high fidelity. This is roughly like saying that I can perfectly replicate a working car but I somehow don't understand anything about how it works. What about the fact that we already know some useful nonrandom changes that we could make, such as the increased dendritic branching observable in specific intelligence-associated alleles?
It doesn't matter. Deepmind is planning to have a rat-level AI before the end of 2017 and Demis doesn't tend to make overly optimistic predictions. How many doublings is a rat away from a human?
Other good trigger-action-plans might involve immediately standing up and hopping or splashing cold water on your face. Changing your physical state is a fantastic shortcut to changing your mental state.
I have been kicking around the idea of analogizing chemical reaction energy channels to human behavior.
I have always loved this metaphor because the "channel" is a literal channel in the abstract 3D space drawn by two "intermolecular distance" axes and one "energy" axis. AB must encounter C with a certain range of energies and in a certain range of intermolecular distances in order to oscillate its way through the energy channel and out the other side to dissociate and reform into A + BC.
The analogy would be that certain psychological conditions must be met as you undertake a task if you expect to make it through the "channel" to your desired end goal. For example, if you come in with "too much energy" or "not enough energy" you'll slide right out of the channel.
Overall it's probably one of those metaphors that's more amusing than it is useful.
At the moment I do have people and soon I will have cards for all from 0 to 99. I don't yet have actions or objects. How did you come up with lists for actions and objects?
I started with this list and customized it with names that had meaning for me.
This page ( http://mt.artofmemory.com/wiki/Person-Action-Object_(PAO)_System ) has a lot of links that could give you other ideas.
Mnemonic peg systems are almost perfectly arbitrary so they provide a decent case study for how you form memories in the absence of any governing structure or natural chunking. It's an "IQ-proof" task.
I don't think it's useful to try to adopt mnemonic pegs without any chunking. I personally started having pegs from 0 to 9 and 00 to 99. I filled them with images of woman that I put into 11 categories with 10 people and sorted every pile by the alphabet.
After thinking about it I realized that the system I'm using doesn't have zero structure, but it has less than, say, multiplication tables.
I use the phonetic two digit Person-Action-Object system. So 23 for example corresponds to "Nemo nomming (a) gnome." (n=2 and m=3). 48 is "Raph raving (at a) ref." (r=4, f/v=8). 56 is "Luigi leashing a leach." (5=l, 6=sh/j/ch). This allows you to chunk 234856 as the vivid image "Nemo raving at a leach."
I'm still not quick enough with the system to say whether it's really useful. Part of the purpose of the exercise is to force myself to memorize something difficult just for the sake of getting a sense of what it is to memorize something, since I've avoided that activity for my entire life.
Presumably this is the kind of thing that is fixable by widespread genetic technologies. Once you can identify and/or fix problematic zygotes, the problem disappears.
Or just straight-up medical cures and treatments.
Also, I have my own medical problems that cause me suffering, but I would still rather be alive than not.
Just one point of data: I kept a spreadsheet when I lost 59 pounds in 96 days. I had values for my personal base burn as a function of current weight and per task (usually a rower and hiking), and a daily deficit of 2000 calories correlated fairly well with a daily loss of .55 pounds (in round numbers; I don't want to sound like the proverbial economist with a sense of humor. I also went over some and under some, used nutritional labels and activity estimates that rounded to the nearest 10, &c.)
I was not scientifically rigorous so grain of salt, but over three months or so, I anecdotally found that 3500 calories of deficit correlated very well with a pound of loss. After that I stayed pretty constant while I was paying attention, 1-2 years. Bit North of there now, but I don't do much counting anymore.
This is a good point. While "calories in != calories out" within a broad range of caloric inputs, because humans have some built-in ability to absorb fluctuations in food intake centered around each person's metabolic setpoints, you can definitely get some play at the extrema of the caloric intake/expenditure axis.
In the opposite direction from your example, if someone has a hard time gaining weight, they may find that eating 3000 cal/day has no effect but eating a carefully measured 6000 cal/day definitely moves the needle upward.
The problem in general is that maintaining a caloric deficit of 2000 cal/day for weeks is going to be impossible to achieve for most people, and likewise maintaining a caloric excess of thousands of calories per day is a full-time job (ask any bodybuilder).
I recently committed to learning a mnemonic peg system in an effort to improve my memory. I've learned a few things from the experience.
Mnemonic peg systems are almost perfectly arbitrary so they provide a decent case study for how you form memories in the absence of any governing structure or natural chunking. It's an "IQ-proof" task.
In the past I have systematically overestimated my ability to retain information after a single exposure and simultaneously underestimated my ability to retain information after training. Specifically, in a single training session I will see a particular pairing and think to myself, "Okay, I've got it," and then manage to forget it a couple of hours later. However, after a few days of repeated exposure, it eventually sinks in to the extent that I can recall it instantly. Forcing yourself to memorize unstructured/arbitrary associations concretely shows you both how bad you are at retaining information by default and how amazingly good you can get at retaining and retrieving information if you just keep at it.
I have also realized the importance of "sitting with the mistake". If I fail to recall an association on a given flashcard, or come up with the wrong association, I will sit with that card for 30 seconds or so, visualizing in as much detail as possible all the connections implicit in the association, before moving on to the next card. In the past, with previous attempts to use flashcards, I would simply see the right answer, think "Oops," and move on. Using flashcards without explicitly focusing on your errors is like trying to train an artificial neural network while skipping the propagation of the error signal.
I suspect a lot of "smart" people get through life avoiding the slog of rote memorization as much as possible. I personally managed to get through school without ever really memorizing the multiplication tables because I could always mentally calculate small numbers "well enough". It has been very instructive to force myself to memorize something essentially arbitrary.
= f037147d6e6c911a85753b9abdedda8d){kind=logo}
= 27d567bc2d48d9f40e9ccc20ea370b15)
= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Obviously, but "natural selection" is the non-random part of evolution. Using it as a byword for evolution as a whole is bad terminology.
Fair enough. My lazy use of terminology aside, I'm pretty sure you could "breed" an Em via replication-with-random-variation followed by selection according to performance-based criteria.