If we say that we can measure intelligence and judge it with a scale of measurement. We get a quantity back from that measurment. The scale has no extremeties and has no limits in the quantity it measures unless the thing it measures has absolutes.

Is it possible then to concieve that AI can be created that surpasses the humans measured by the same scale.

If it is, doesn't it mean that the AI that has been created has not been determined by the limitations of those humans and its creation has not been determined from the intelligence associated with humans but would have to be created from something not directly attributable to intelligence and something more fundamental, and thus an AI more intelligent will be a consequence of this.

Or does the scale have absolutes and we have a limit in intelligence and maybe the artificial intelligence will be only created by someone that reaches that absolute measure, or any creation of which will be limited within the extremeties of the scale and be limited and not surpass the humans intelligence also confined within it?

  • $\begingroup$ it is a misconception that intelligence is a measurable value. what about dishwashing intelligence or mathematic intelligence? are those co-related? why value one over another? $\endgroup$ – k.c. sayz 'k.c sayz' Nov 24 '17 at 17:37
  • $\begingroup$ @k.c.sayz'k.csayz' The question looks to be opinion based. $\endgroup$ – quintumnia Nov 25 '17 at 18:07
  • $\begingroup$ @k.c.sayz'k.csayz' intelligence does seem to be measurable in relation to other intelligences, in some cases. (For instance, we were easily able to rank our old-school game AIs by having them play against each other.) But I take your overall point. At best, intelligent measurement will always be necessarily subjective. $\endgroup$ – DukeZhou Nov 28 '17 at 1:39
  • $\begingroup$ @DukeZhou is a subjective measure the greater or lesser the grounds on quantity unless the measure you have is a judgement based upon differences another has from yourself based upon the quality of each difference. Can you say that it's very difficult to judge intelligence simply because we have no concise definition of intelligence itself. A task I would say is itself is very important in determining whether we can simulate it in a machine based upon that definition. $\endgroup$ – Bobs Nov 28 '17 at 21:12
  • $\begingroup$ @Bobs I think the Computer Science definition of intelligence is pretty solid in the sense that it's a gauge of performance in a given task or set of tasks, where objectives are defined and results can be gauged. "Strong narrow AI" has been recently used to connote an AI that outperforms a human in a specific task, such a game, or outperforms weaker algorithms, such as in managing air-conditions systems. Artificial General Intelligence (AGI) is much broader, and connotes a hypothetical AI that performs on a human level or better on any given task. $\endgroup$ – DukeZhou Nov 28 '17 at 21:23

Pre-judgements and biases can influence decisions and even skew relayed information causing informational delays since the information road took a detour when emotional entities from humans became involved. A different question might be which is the lesser of the two evils? People tend to frown on anything or anybody considered emotional so AI might win this race BUT what is a human weakness can be the edge needed to save the world and it will need saving if AI was left to make final judgements. Computers are programned and can be updated but they evolve differently then humans do over time. Humans instinctively begin to adapt to changing variables in our environments no matter how noticable or discreet. Humans automatically problem solve using surrounding variables, risk factors, possible outcomes, different perspectives (based on emotion and experience) and cause/effect of a certain choice or lack of. Knowledge appears indefinate and humans more brain capacity to store memories (partly due to emotion). The judgement process for humans is innate and can be accomplished independently where as when a computer will reach a limit and guess who shows up to help... The emotional HUMINT!!! Overall, when making decisions that will effect our world as a whole Human intelligence needs to win everytime or else humanity will lose. Computers do not take into account life struggles, traumatic events or stressors based on new variables and discoveries. Having AI as the ultimate authority over judgements would be equivalent to the Anti christ taking the throne. Computers hold no accountability, fear from laws of the universe/karma, or remorse. Humans learn from bad decisions and outcomes automatically ( or should in theory) computers could generate mass numbers of bad judgements before a human could turn it off or update it making computers very dependent. However while most technology targets individuals with information, AI does not discriminate but neither does this author. The human used their better judgements this round!

| improve this answer | |
  • $\begingroup$ Analyze your answer carefully,revise it and add some paragraphs! $\endgroup$ – quintumnia Nov 25 '17 at 18:06

For the sake of the question, let's assume that such measurement existed, and then let's assume that there's an AI that surpassed that measurement.

If we truly believed in that measurement, then the answer is quite obvious: yes, the intelligence of the AI has not been limited by the intelligence of the creators.

Closer to the real world, we might start arguing that this actually performs some kind of intelligence and not everything that makes up intelligence. This is a case where we don't trust that measurement anymore to mean what we said it would mean.

Also, another debate around it is whether the AI is really intelligent, or if it has been tuned just to skew the results of the measurements. This is the case against the Turing Test, which was meant to be a particular point in that scale. And, again, it is a case where we don't trust the scale anymore: there must be something else that AI needs to do before it is really intelligent, right?

Let's leave intelligence aside because it's a fuzzy concept for us and focus on something different. As humans, could we create something that's taller than we've been able to lift up?

As humans, have we been able to construct anything that can go faster than the fastest of humans ever was?

As humans, have we been able to construct anything that can perform task X any better/faster than any human has been able to do?

The answer to all of that is: yes, we've been able to do it so far. Why would intelligence be any different?

At the moment we define intelligence as a particular set of aspects (hopefully, that can be measured) then it is easy to create machines that do it better. We've done it with games, we've done it with conversation, we've done it image recognition, all within living memory of places where AI has stood up and performed better.

And all the meanwhile (just in the first part of my explanation) as humans we've gone back and said: "well, that's not realllly intelligence, it's actually missing Y". And we repeat.

To answer your question: it's very easy to overcome human capabilities, and I think intelligence will not be the exception. So yes, the limitations of the creators is not a limitation for the creation. We don't yet have such scale because we're not entirely decided on what we want to measure exactly.

| improve this answer | |
  • $\begingroup$ precisely the determination of intelligence is not as easy to put boundaries on and an overall measure of intelligence.The description of intelligence is loose and an approach at replicating it is down to part of human and animal intelligence and the particular "problem" for want of a better word, and there solution algorithms adopt methodical approaches to problem solution human are not as discrete and are grey but we use the word intelligence to there approach. I'm not questioning intelligence is not a property of the mind but it's definition and measurement is not clear cut in object $\endgroup$ – Bobs Feb 1 '18 at 22:08
  • $\begingroup$ @Bobs I agree, thanks! In order to not dismiss your question with a simple "intelligence is not well defined" answer, I have assumed that there is a definition and a clear way to measure it (whatever it is). From there I created a response on what the implications would be. I hope you find it useful! $\endgroup$ – Alpha Feb 2 '18 at 15:45

Is it possible then to conceive that AI can be created that surpasses the humans measured by the same scale?

Most definitely, and it's been done in Chess and Go. These algorithmic intelligences that outperform the strongest humans in a specific task, such as a game, are sometimes referred to as "Strong narrow AI."

The term strong narrow AI can also be applied to algorithms that outperform other algorithms, such as in managing air-conditioning systems.

Both measures are entirely subjective--not based on absolute values, but on comparison of measurable results in a specific task.

  • Regarding intelligence in general, I don't think it's a question of a single scale, but an infinitely expanding set of scales

This is goal of Artificial General Intelligence (AGI), to create an artificial intelligence that performs any task comparably or better than humans. Still very hypothetical, but believed by many to be quite possible.

There is an idea that intelligence is merely a function of the complexity of the system. Thus microbes are less generally intelligent than frogs, and frogs less generally intelligent that humans, even though microbes and frogs can outperform humans in some ways, because here we're talking about the complexity of the brain.

In the same way, given enough computing resources, a sufficiently sophisticated algorithm should be able to exceed the complexity of the human brain, but this is currently out of reach due to hard, physical limitations.

Turing have made the point that we likely don't need to fully understand the human mind to simulate it (the Imitation Game or "Turning Test".) Likewise, it's clear we don't need an absolute measure for intelligence in order to generate it.

Intelligence can be said to emerge in nature at the most basic biological level, expressed in evolutionary processes. Evolutionary Game Theory and Genetic Algorithms extend this into the computing context. Deep Learning can be understood as a form of self-evolution for algorithmic intelligence, and it definitely works, outperforming all other previous AI methods in ways that are measurable.

The implications of the continuing validation of Deep Learning are that humans are just the catalyst, creating intelligences just intelligent enough to get more intelligent without human intervention. This can be understood as a natural, evolutionary process, because the algorithms get stronger by interacting with the environment and other intelligences.

The limitations of human intelligence would seem to indicate that a superintelligence cannot be built by humans, but rather would evolve from the initial state crafted by humans.

| improve this answer | |
  • $\begingroup$ But that intelligence is a function of memory of observation and reference and logic to accumulate the fact from its reference to concur accumulation of further logic from the truth in the memory of that observation and its reference and impact in further events that recognise the logic in an assumption of knowledge of memory its referencing and the truth of its further accumulation depends upon the initial predicate and any statement is a question of subjective truth. That leaves us with faith being more than accumulated knowledge from false truths lol. $\endgroup$ – Bobs May 20 '18 at 22:21

Not the answer you're looking for? Browse other questions tagged or ask your own question.