Artificial Super Intelligence - are we ready for it?

Many say that we won't experience the worst of the effects associated with global warming for another 50 or even 100 years.

But what if we face a crisis, that very few people seem to know about today, that will manifest itself almost overnight in a matter of only 20-30 years?

Welcome to the world of true artificial intelligence.

The two parts associated with this linked piece are really, really long, but in my opinion well worth it. The author puts all of this in layman's terms so it's long, but a relatively easy read.

http://waitbutwhy.com/2015/01/artifi...olution-1.html

I'm also throwing in a long read on Fermi's paradox, which relates to the question of why we haven't heard from any extra-terrestrial species. Fermi's paradox also involves AI because even an ET species that isn't interested in galaxial travel, could easily send AI voyageurs to spy on or greet others.

http://waitbutwhy.com/2014/05/fermi-paradox.html


My opinion? This stuff concerns me. Humankind seems best at reacting and countering crises that we can see coming. Things like former ice ages, imperialism and industrialization, we could see coming gradually and adjust our lives and behavioral patterns. That's why although I'm concerned with global warming, I do believe that we'll have some interesting answers to it, including big geo-engineering responses. We'll adapt. Also, it might not be all bad. Just as the last Ice Age must have been devastating, humankind also seemed to use it as an advantage by crossing over the ice bridge within the Bering Sea. I'm guessing there will be some unforeseen benefits to global warming.

Crises that come upon us all at once, however, I believe are a much bigger issues for humankind. So far, we dodged a huge bullet with the invention of atomic weaponry. But I think there was a lot of luck at play.

Will we be so lucky when Artificial Super Intelligence arrives? As the author of the linked pieces asserts, I'm not too sure.

Quote:

---

Originally Posted by Odelay

Many say that we won't experience the worst of the effects associated with global warming for another 50 or even 100 years.

But what if we face a crisis, that very few people seem to know about today, that will manifest itself almost overnight in a matter of only 20-30 years?

Welcome to the world of true artificial intelligence.
My opinion? This stuff concerns me. Humankind seems best at reacting and countering crises that we can see coming. Things like former ice ages, imperialism and industrialization, we could see coming gradually and adjust our lives and behavioral patterns. That's why although I'm concerned with global warming, I do believe that we'll have some interesting answers to it, including big geo-engineering responses. We'll adapt. Also, it might not be all bad. Just as the last Ice Age must have been devastating, humankind also seemed to use it as an advantage by crossing over the ice bridge within the Bering Sea. I'm guessing there will be some unfor
eseen benefits to global warming.

Crises that come upon us all at once, however, I believe are a much bigger issues for humankind. So far, we dodged a huge bullet with the invention of atomic weaponry. But I think there was a lot of luck at play.

Will we be so lucky when Artificial Super Intelligence arrives? As the author of the linked pieces asserts, I'm not too sure.

---

Odelay I was going to thank you for these links, but when I began to read the first one, I became irritated at simple errors in a rather freewheeling discussion. For example, at the very start he writes;
Imagine taking a time machine back to 1750—a time when the world was in a permanent power outage, long-distance communication meant either yelling loudly or firing a cannon in the air, and all transportation ran on hay.
I can't take this seriously, not even as a light-hearted comment, given the huge importance of the sea in transportation, long before 1750 and including internal waterways, seas and oceans- even if in the case of internal waterways canal boats were at one time drawn by horses. A rather more eloquent and dare I say more informed historical framework takes the world in 1400 and is in Eric Wolf's Europe and the People Without History (1982) and the first volume of Quentin Skinner's The Foundations of Modern Political Thought(197-eight) which takes it cue from the emerging city states of Italy in the 12th century as a point of departure for a discussion for modern ideas in politics.

May I refer you to the Centre for the Study of Existential Risk in Cambridge (UK) where you will find a good set of discussions on AI and associated issues, as well as links to the Future of Humanity Institute in Oxford and some US institutions.
http://cser.org/about/our-mission/

This is a vast topic, whether it is about the impact alogrithms have had on our daily life, or on the broader issue of humans being replaced by robots with feelings...

Thanks, Odelay, for the articles. I perused the first one over my morning cappuccino at the coffee shop. It was a long one with many points. And thank you Stavros for the link.

I think it may be possible to make a distinction between intelligence, knowledge and processing power (speed and memory), although the three bleed together and support one another. We have been quite successful in building machines with stupendous speed and considerable memory, but not (I think) so successful at building intelligent machines or knowledgable ones (if knowledge is more than just data storage).

Take chess playing programs for example. At first AI researchers attempted to reduce the heuristics, strategies and insights that chess masters have gained through study and experience into computer code. But this sort of attack never led to much success. But today there are algorithms now that consistently outperform masters of the game. What changed? The speed and memory of machines changed. AI researchers realized they win chess games by brute force rather than cunning and intelligence. For the most part modern chess playing algorithms simply run through the tree of all possible moves of the game for more moves ahead than any human could manage.

An analogy: Suppose one would like to solve the cubic equation 2 x^3 - 55 x^2+13 x + 378 = 0. Two methods come to mind. Use a change of variables to transform it into a more manageable form. Factor and backtrack to the roots. If you can get it to work this will provide you with a general procedure for solving all cubic equations. This is what Niccolo Tartaglia did back in the sixteenth century (there’s an interesting history here). The second method is to just start picking numbers 0,1,-1,2,-2,3,-3 etc. and plugging them into the formula so see if they work. A computer can do this super-fast and will come up the answer 27 in the time it takes you to lift your eyes from the enter button to the computer screen. The advantage of Tartaglia’s approach is now he has a general expression for the roots of a cubic which will find you those roots even when they’re irrational. The second approach will only find integer roots and it gives you know way of expressing them generally. The first approach is clever, The second is brute force. Of course the programmers could just code Tartaglia’s formula into a algorithm and then the computer will beat Tartaglia every time. But the point here is: there is no general formula (yet) for playing a good game of chess. Humans stand a better chance using the heuristic strategies they’ve gain through experience. Machines stand a better chance using brute force searches.

Another example is language translation. Several decades ago it was predicted that computers would be translating languages with nuance and ease. This isn’t so. Again, the best algorithms simply utilize brute force. Google translate searches a huge data base of previously translated text for words and phrases that appear in the text to be translated and simply replaces those phrases with the found counterparts in the target language. No grammar. No computerized Chomsky language modules. Just brute force searches. It translates languages better than a chimp can, but it’s not as intelligent as a chimp.

It’s not that I don’t believe machines can think. We’re machines, and we think. But I don’t think (I’m no expert mind you) there are any artificial general intelligence algorithms in existence that come close to what a chimp does.

Nor am I oblivious to the threat of artificial intelligence. In the wrong hands immense populations and be tracked, surveilled and controlled.

The social and economic repercussions of sophisticated software are and will continue to be enormous. Because brute force calculation can simulate intelligent behavior, there is a temptation to replace intelligent agents with algorithms. They replace workers on the factory floor. They sell airline tickets and even fly the airliners. They diagnose disease. They tell us if the DNA of the defendant matches the blood and semen found at the crime scene. They break these very words into bits an route them in different directions around the globe and reconstruct them on your screen. Power doesn’t have to be intelligent to be dangerous.

The article speculates that the machines of the future will be super-intelligent. I question the concept:

I don’t know, but I think it’s possible that intelligence is not hierarchical. Like a traffic light. It can turn green and it’ll be legal to proceed, but it can’t turn more green and provide more assurance of your right to proceed.
Another analogy can be found in the theory of computation. There are a lot of different kinds of automata, but there are none more “powerful” than a Universal Turing Machine (UTM). There is no computational task, in principle, that cannot be done with a UTM. The hardware in which it is realized might be slow or fast. It might be electronic or organic. But computationally speaker it is the top of the mountain. There is nothing higher.
Intelligence may be like that. There may be intelligent agents who think faster than us and have more memory than us. There may be intelligent agents (at least in principle) who thick on geologic time scales. But perhaps there’s no intelligent way to make sense of the question, “Which of us is more intelligent?”

Suppose, however, super-intelligence IS a possibility and in a few short decades there will super-intelligent AI’s will be inhabiting the cloud. I wonder if we’ll recognize them. What sort of culture would they have? How sophisticated would their languages be? Would be recognize their discourse as discourse, or would it just seem like a scatter of random bits? What sort of arts would engage them? Does a chimp see the image of a woman when it examines the Mona Lisa? Should we gaze at a swirl of changing characters on our screens would we recognize it as a super-intelligent agent’s expression of impermanent beauty and balance? Would we even get that super-intelligent art is art? Or is this just a reductio absurdum leading us back to the conclusion that we are ultimately the same?

Sorry for the long rambling response.

Quote:

---

Originally Posted by trish

The social and economic repercussions of sophisticated software are and will continue to be enormous. Because brute force calculation can simulate intelligent behavior, there is a temptation to replace intelligent agents with algorithms. They replace workers on the factory floor. They sell airline tickets and even fly the airliners.

---

By coincidence the BBC a few nights ago showed a documentary on the 2009 crash of Air France 447 which left Rio de Janeiro and crashed into the Atlantic about 3 hours after take-off. It was an extraordinary example of the confusion that took place in the cock-pit when the computers shut down because they could not make sense of the pilot error which resulted in the plane going nose-up into a storm and stalling. But the co-pilot had manual control of the plane because the computers shut down. It raised the question, would this plane have crashed if it had been left solely to the computer? The initial problem was that the storm outside freezed up the pitot tubes which give speeds readings which could not then be read or understood, but a drop in altitude would have melted the ice and re-booted that info and at 38,000 feet that is enough to re-order the plane's trajectory, but the co-pilot kept his hand on the manual control that was sending the plane ever upward until it just could go no further and dropped like a stone in less than four minutes.

By contrast, I have more than once used an automated supermarket checkout and without doing anything wrong have been unable to proceed because of an unexpected item in bagging area...something tells me we have a long way to go before reconciling the wonders of technology with the not always wonderful people who design it.

I'll look at just one of your ramblings - the Universal Turing Machine. A UTM can compute anything that is computable! And no more. Turing showed himself that some seemingly simple tasks such as predicting when a process will end - the halting problem - can not be solved. This may appear rather abstract but actually it is rather important to us as humans to have an understanding when something may stop or change.

AI works in two ways - brute force (OK for playing chess but for not getting through life in general) or by optimising some function (we call it a cost function). Two ways of optimisation - go down hills (reduce some error between what you observe and what you desire) or random jumps in the dark (evolutionary computing). A fully extensive search could need infinite resources and infinite time. Really the idea behind NP-complete problems (like the Travelling Salesman). So brute force and brute "logic" don't always work - so you have to accept answers that are "good enough". There is the rub.

On a more practical side - we may be able to sequence DNA at speeds that seem incredible or believe that we can recognise faces better than humans can, but we don't have an AI computer that can identify a chair!

We still have little idea how our brains convert signals into symbols.

Quote:

---

Originally Posted by trish

Another analogy can be found in the theory of computation. There are a lot of different kinds of automata, but there are none more “powerful” than a Universal Turing Machine (UTM). There is no computational task, in principle, that cannot be done with a UTM. The hardware in which it is realized might be slow or fast. It might be electronic or organic. But computationally speaker it is the top of the mountain. There is nothing higher.

---

Quote:

---

Originally Posted by martin48

I'll look at just one of your ramblings - the Universal Turing Machine. A UTM can compute anything that is computable! And no more..

---

I didn’t mean to imply anything different. What one automaton can compute, so can a UTM. A mathematician often speaks in terms of partial functions (from finite ordinals to finite ordinals). An computing machine is said to compute a partial function f if you can program it to output f(n) whenever it is given an input n from the domain of f (and will not output anything if n isn’t in the domain of f). Before the discovery of the UTM one might have thought that given any computer, there would always be another one that computes more functions. A computer designer might have aspired to build machines which could compute more and more functions. But the UTM is the best one can do in that regard. Once you build a machine equivalent to a UTM you’ve crossed the threshold where this specific aspiration come to an end. No other machine can be “smarter” than the one you just built. Now it’s just a matter of giving it knowledge (programming it), giving it more readily available memory and making it faster. My speculation is that “intelligence” may be analogous to this. Humans may have reached the stage where no other being can be “smarter.” Other agents might be more knowledgeable, might have more ready memory and be faster on their feet; but in some basic sense we may be as “intelligent” as it gets. There’s a depressing thought for you. Good news: there’s no reason to believe this, it’s just a ill expressed possibility.

Quote:

---

AI works in two ways - brute force (OK for playing chess but for not getting through life in general) or by optimising some function (we call it a cost function). Two ways of optimisation - go down hills (reduce some error between what you observe and what you desire) or random jumps in the dark (evolutionary computing). A fully extensive search could need infinite resources and infinite time. Really the idea behind NP-complete problems (like the Travelling Salesman). So brute force and brute "logic" don't always work - so you have to accept answers that are "good enough". There is the rub.

---

True. If one needed the roots of a fifth degree polynomial, brute force search would almost surely fail (unless by dumb luck at least one of them in the search range). But of course there are numerical methods for approximating those roots to within any desired degree of accuracy. Machines are still way better than us at this because of their brute speed. So smart programming, brute speed, lots of memory and a willingness to “accept answers that are ‘good enough’” are important ingredients for good computational problem solving.

Quote:

---

On a more practical side - we may be able to sequence DNA at speeds that seem incredible or believe that we can recognise faces better than humans can, but we don't have an AI computer that can identify a chair!

We still have little idea how our brains convert signals into symbols.
.

---

Totally agree.

Moreover, although they are sometimes modeled as neural nets, brains are not digital machines. They are hybrid digital/analog. I’m not entirely convinced that we can’t–in principle–do more than a UTM (one can mathematically construct–in the abstract–hybrid machines that realize non-computable functions).

Quote:

---

Originally Posted by Stavros

Odelay I was going to thank you for these links, but when I began to read the first one, I became irritated at simple errors in a rather freewheeling discussion. For example, at the very start he writes;
Imagine taking a time machine back to 1750—a time when the world was in a permanent power outage, long-distance communication meant either yelling loudly or firing a cannon in the air, and all transportation ran on hay.
I can't take this seriously, not even as a light-hearted comment, given the huge importance of the sea in transportation, long before 1750 and including internal waterways, seas and oceans- even if in the case of internal waterways canal boats were at one time drawn by horses. A rather more eloquent and dare I say more informed historical framework takes the world in 1400 and is in Eric Wolf's Europe and the People Without History (1982) and the first volume of Quentin Skinner's The Foundations of Modern Political Thought(197-eight) which takes it cue from the emerging city states of Italy in the 12th century as a point of departure for a discussion for modern ideas in politics.

May I refer you to the Centre for the Study of Existential Risk in Cambridge (UK) where you will find a good set of discussions on AI and associated issues, as well as links to the Future of Humanity Institute in Oxford and some US institutions.
http://cser.org/about/our-mission/

This is a vast topic, whether it is about the impact alogrithms have had on our daily life, or on the broader issue of humans being replaced by robots with feelings...

---

Understood about not reading further. The author is clearly shooting for a broad audience that isn’t all that well informed on the latest thinking on AI, and he sort of caught me in that net.

Thanks for the link. To be fair to the author, one of the experts he sites extensively is Nick Bostrom out of Oxford, who happens to be the second named AI thought leader in the AI part of the site you linked to. To be sure, an article that focuses on just Bostrom’s views on AI would give a lay person plenty to think about.

Stavros, you are absolutely correct about the impact of algorithms on our current and future daily lives. Even if AGI isn’t achieved for 300 years, just the development of more and more sophisticated ANI (artificial narrow intelligence) has huge, serious repercussions for the 8 billion people currently residing on the planet. I work in Information Technology and I can foresee even System Administrators being replaced by ANI as troubleshooting badly performing systems will be conducive to large scale trial and error efforts by ANI. Technology replacing technology workers.

I’m not too optimistic about what the capital barons will do with yet another financial windfall when they replace cheap labor with no labor.

Odelay, perhaps the key question in the short to medium term is: would you fly from LA to San Francisco in an aeroplane flown by a computer rather than a human being?

Quote:

---

Originally Posted by Stavros

Odelay, perhaps the key question in the short to medium term is: would you fly from LA to San Francisco in an aeroplane flown by a computer rather than a human being?

---

I'm not a good one to be asked this question as my father worked within the airline industry. As a result, I would say yes. Way back when, I had a time or two within the cockpits of commercial airliners and watched as the pilot would engage the autopilot of aircraft of the late 60's.

I am not an aeronautical engineer, but I can only imagine the improvements in this technology in the intervening 45+ years. With the dramatic decrease of airline accidents since that time, it's hard to argue that autopilot technologies didn't play a role in the increase in flight safety.

And now, after this most recent accident, people are going to seriously ponder whether a computer should be able to take control away from a human. By the way, I liked your question in an earlier post about whether a computer might have figured out the best course of action in a 2009 crash had it been given a chance.

Quote:

---

Originally Posted by Stavros

...would you fly from LA to San Francisco in an aeroplane flown by a computer rather than a human being?

---

Depends on the computer's state of mind. Has it been drinking? Is it depressed? Is it preoccupied with it's deteriorating marriage? Did it get enough sleep since the last flight? Perhaps most importantly, does it nurse a growing hatred of human beings? Especially the one's who whine about the discomforts of air travel?