A younger student, Ben Schultzer, emailed me yesterday, to ask about AGI. Since I like writing, I wrote a long answer. I’m reproducing some of it below.
1. Evidently (because of OpenCog), you have at least some belief that symbolic/classic techniques are valuable towards the development of AGI.
Do you think that a hybrid approach is the best way of getting there (compared to, for examples, purely deep learning, purely symbolic, or even simply simulating all 100 billion neurons of the human brain)?
Yes, but … symbolic vs. neural is the wrong perspective. Sparse representations vs. dense representations is the correct distinction to make. Both involve numbers and vectors and probabilities; the difference is the density of the interconnections between “things”. I will expand on this answer in a new blog post on the OpenCog Brainwave blog later today (tomorrow?)
2. What is …
3. Could you give a rough estimate for the minimum amount of computing resources required to establish and run an AGI?
The next problem is the term “AGI”. It’s not what you think it is. Most people seem to think of it as a nice chap you can invite over for dinner, and have a good chatbot with it. It’s not like that. Let me explain.
If you were the only human in the world, your brain would be all there is. But no, there are others, and your brain uses words to communicate with other brains, the network forming a giant global brain. For example, I stub my toe; there’s a chemical reaction that releases assorted signalling molecules in my toe, that tweaks some neurons, that pulse some solitons up my spine to my brain, and I yell out “ouch” which makes air vibrate which makes neurons in your brain know that I stubbed my toe. Natural language is a Star Trek teleporter for acetylcholine and dopamine.
What is the global brain doing? Well, on facebook, its is a quivering blob of jello getting delusional about flat-earth, global warming, BLM and #defund. More stable structures in the global brain are universities and rule-of-law, both having been stable configurations for about 800 years now, and corporations and nation-states, which have been around for about 400+ years. (Sure, Rome is older, but Rome was agrarian.)
Starting in the 1980’s, a significant portion of the computing performed by the global brain has been done in silicon, first by spreadsheets in corporations, but also desktop-publishing in local communities. Then, in the 2000’s, social media completely rewired the global brain. Before 2000, individuals used sound-waves to communicate with neighbors, friends, bosses, and radio waves and television to communicate with mass audiences. After 2000, with facebook and other systems, individual brains are now connected electronically with fiber optics, directly, point-to-point, without intermediaries, without having newspapers in the middle, filtering out who can say what to who. The network topology of the global brain is radically different now. (This is one reason why everything is so “crazy” now — without the filtering of newspaper editors suppressing crackpot stories, the global brain has started to amplify the insanity – flat-earth, chem-trails, magenta-people, trump derangement syndrome… its all because the network topology is different. The Great Firewall of China is an attempt to tamp down and modulate the crazy-making of social media. At least, that’s the benevolent description of it.)
Besides communications, silicon electronics also provides RAM storage for the global brain. For example, Wikipedia, but also blogs and PDF’s. The long-term memory ability of the human brain has dramatically improved — in the 1980’s, it took the typical human a minute or two to recall one fact out of a small number of facts. Now it takes only seconds, thanks to google and duck-duck-go. And you can remember much much more: you can remember all of Wikipedia, for example, which was impossible in the 1980’s, unless you walked to the library to read a book. Which the global brain usually didn’t do, preferring to drink and party.
So, anyway, I estimate that currently, silicon electronics makes up about 0.01% to maybe 0.2% of the processing power of the global brain. It’s still poorly integrated. Cell-phones are OK, augumented reality would be better, and a neural lace: wires directly implanted in the brain would provide the highest bandwidth. But do you really want a permanent facebook party wired to your brain? Hmm. Anyway, by Moore’s law, if we are at 0.01% to 0.2% today, then…
Well, the point of this long story is that you are already a participant in AGI: its just called “corporations” and “communities” and “courts of law” and “police stations”, and that AGI that you are already a part of is just 0.01% in silicon, although that fraction is increasing exponentially.
The point is: AGI is not going to be like this super-smart robot, individual and distinct, the nice chap we can invite over for dinner. It will be part of “us”. It will be intimately connected to us, literally fucking with our brains. If you thought algorithmic propaganda was problematic, well, that’s just the first baby-step on the path to AGI. True AGI is a kind of atom bomb going off, and instead of splitting uranium atoms, it’s splitting memes. You know — those memes that show a picture of the Corona-beer guy, and the caption says “I don’t always vote for Trump, but when I do, I prep for the end of the world.”
5. Do you have any advice for a younger person who hasn’t established their career yet?
If we actually have an algorithmic-propaganda bomb go off, do you think “careers” will still matter?
In the short term, having reliable access to a monetary income is important. Capitalism may take more than a few decades to implode, so having money is still very useful. Also, having a high IQ is also useful for survival. So, high IQ is partly determined by genetics, and partly determined by social milieu — so I recommend Ivy League Universities, and hanging out with the very smartest people possible, the smartest ones who will admit you to their social circle. In the old days, this was called “social climbing”, and there’s a nasty way to do it, and a nice way, where you don’t have to leave wreckage behind you.
Exactly how this will help you when the algorithmic propaganda bomb goes off, I dunno. How much time do we have? I dunno. Will it end wonderfully? I dunno. It will be different, though.
I need to find a footnote plugin for wordpress. Because the above text deserves footnotes. So, instead, a backgrounder post here: called “Endorphin Supply Chains.”
What does that long footnote-post have to do with AGI? Well, AGI slots into this system of global supply chains and manufacturing systems. Existing AI algorithms are used to not only inform/misinform voters on political issues, but also improve manufacturing and shipping. The AGI algorithms will be deployed into this infrastructure, as a particularly powerful controlling feedback loop. The boundaries of AGI are fuzzier than you think, and the future is closer than it appears.