9
$\begingroup$

It appears to always have been the focus in the literature to approximate components of the human mind, assuming it to be the most advanced. If other animals came into the AI landscape, it was only to study primates in ways that are not practical to study humans or to simulate the neural activity of a slug because its nervous system is simple.

Perhaps there is a more forward thinking reason to consider using lower life forms as the model for desired artificial intelligence. I've been reading what E. O. Wilson and others had to say about the collaborative abilities of other species. There are remarkable qualities in organisms as simple and adaptive as bacteria. Certainly, ants are the model species for collaboration. Honey bees are arguably the most construction savvy, carrying sustainability of lifestyle and interrelationships with other species to an art form far above the capability of human intelligence.

Using sports analogies to characterize the options, human intelligence is more like pre-enlightenment gladiator sports or at least ice hockey, where injuring the opponent is considered the smart strategy. What bees do is more like mountain climbing, constructing with precision and care.

What ants do is much like relay racing, where there is little interest in the opposing team because each colony, just like each lane in the track is independent and the lanes are marked. Ants similarly mark their territory, and the territorial claims are respected as in the best of Westphalian geopolitical statesmanship. There are neither petty jealousies nor competitions solely for the sake of prideful primacy. With ants, just as with the smart track and field coach, the objective is that each leg of the race perform well against the relay racer's previous best.

Bacteria are the long distance runners. They swap DNA with one another, and ignore all the rules of pain and fear. They behave in a sustainable way that takes nothing for granted and uses everything for survival. And they have survived for nearly the entire duration of the earth's existence. They will likely be around for a hundred billion years after humanity is gone, if the sun doesn't go supernova first.

Why would we want to program computers to endlessly behave as competitors? Do people download smart chess programs so they can repeatedly lose? No, they download Android OS because it collaborates and it costs nothing. Can't we find nonzero-sum games to play where win-win scenarios are possible?

Don't we already have enough back-biting, gossipy, hyper-critical agents around from within our own species already? Why not send AI in the direction of collaborative intelligence, like ants? Wouldn't it be better to have new artificial friends that would like to share the burden of our daily tasks?

Don't we want our robots of the future to build like a honey bee builds, in hexagons? Or do we want our robots to follow our example, wasting 70% of materials in vertical construction because of an irrational insistence on ninety degree angles, like only humans would do?

$\endgroup$
  • 3
    $\begingroup$ You have gained a very rosy view of competition between social insect colonies. The statement "there is no interest in the opposing team because each colony, just like each lane in the track is independent" is simply not true. Ant colonies will compete with and fight other members of their own species, and definitely against other species. Although I (and I think many people) would agree about valuing other forms of ability/intelligence as equally well adapted as humans, there is still vicious intra-species, inter-species and often inter-relation competition going on in all the species. $\endgroup$ – Neil Slater Aug 11 '18 at 8:32
  • 1
    $\begingroup$ @NeilSlater very correct...Organisms just want their DNA to propagate and ants and bees (worker) share 50% dna with the other workers...That is their main motivation $\endgroup$ – DuttaA Aug 11 '18 at 9:21
  • $\begingroup$ Like I said they share genes only in their colony and kill any ants of other colony even if of same species...Humans don't share genes unless it's a family member where u see selfless help (in general) ...But since we have such a big brain (capable brain) we accumulate experiences which change our genetic behavior and we may become hostile towards family...I want to answer ur question but it'll be quite big so I'll do it when I get time $\endgroup$ – DuttaA Aug 11 '18 at 12:24
  • $\begingroup$ Bacteria/ants/bees, they don't know that what they are doing is intelligent, they cannot reflect back on what they are doing and come up with more efficient methods. Can you make these simple beings accomplish goals other than what they already accomplish? Nope. $\endgroup$ – Ankur Aug 11 '18 at 14:24
  • 1
    $\begingroup$ @Ankur However, you can study the principles of how their successful behaviour works, and apply it artificially to a new problem. This is something that is done, and the OP's question poses some interesting points, such as whether this is enough to cover all desired use cases for AI, and whether it would be a safer/better approach. I think "no" and "no" to those questions, but it would take some effort to put together an answer. $\endgroup$ – Neil Slater Aug 12 '18 at 10:12
6
$\begingroup$

I take a fairly dim view of humans as a species and have proposed a reverse Turing Test by which humans are evaluated for sufficient humanity (). (just kidding, but not entirely;)

Humans are clearly not the most successful species on earth if you go by biomass:

enter image description here

I fully agree that less complex forms of life are very effective, and worth emulating for specific problems (see Ant colony optimization algorithms as an example.)

But what does a world look like when run by ants? Is that a world we want to live in?

Not only is competition in nature absolutely brutal, this brutality is the mechanism by which species become optimized (evolution).

Humans are flawed, but we've also achieved great triumphs, and what may separate us from the animals is that our successes can also fall into the category of "triumphs of the spirit". Where there are forces of selfishness and greed driving society in one direction, we also have the counter-force, pushing in alternate directions.

Some games simply are zero-sum and there's no getting around that, but it doesn't mean they have to be "all or nothing" (We recommend partisan Sudoku over Chess for precisely this reason.)

There is a concept worth looking at known as Pareto Efficiency. Financial speculation, which much of the world economy now seems based on, depends on the "greater fool" principle, where one agents benefit depends entirely on another agents loss. The worst of us seem to be focused on these mechanisms to the exclusion of all others because it requires no real commitment, merely stronger analysis and/or better information, and liquidity.

What the best of us seem to reach for instead is Pareto Improvement, by which agents can be made better off without requiring other agents to be made worse off.

The point here is that it took human intelligence to identify and define these conditions. (Nature could care less.) Pareto optimality has been controversial because of it's origin, but that was before the tech industry, which is very good at creating new opportunities, and it seems to me it can also be used to balance distribution of resources to try and maximize social stability. (i.e. It can sometimes be more optimal to be a little worse off.)

My dismal view of our species comes not from being inefficient or, often, foolish, but from the idea that, at this stage of history, we're still acting like we just came down from the trees, regressing to tribalism and partisan conflict instead of cooperation.

Contrary to your thesis:

  • I believe it is not only important to reach for human-level algorithmic intelligence, it is incumbent on us to "humanize" such algorithms to the greatest extent possible.

Superintelligence and AGI are still on the horizon and wholly theoretical, but, if they are achieved and these algorithms are pure mechanism, devoid of the values that we humans, at our best, cherish, we might end up becoming merely another victim of evolutionary process, outmoded by new forms of artificial life we have no chance of competing against.

In some sense, the drive toward human-like can be regarded as a celebration of humanity, more sophisticated, but not entirely different in spirit than Michelangelo's David.

$\endgroup$
  • $\begingroup$ @DouglasDaseeco I actually like your question very much b/c I 100% agree with you regarding the general utility/benefit of "low-level" functions. Part of my own project involves gauging the performance of generalized heuristics vs. learning algorithms on certain problem sets where time/resources are severely restricted, and, in particular, where the topology of the current problem is not known beforehand. When do sound-but-less-optimal economic decisions reached more quickly yield more aggregate benefit than more optimal decisions that require more resources and time? $\endgroup$ – DukeZhou Aug 16 '18 at 19:09
  • $\begingroup$ @DouglasDaseeco On the "mythology of AI" front, a few of the leading authors such as Stephenson and Gibson have been more interested recently in just the kind of limited intelligence you're describing. (Stephenson has an engineering background, and was never interested in AGI, but Gibson made his name writing about it in his first books.) $\endgroup$ – DukeZhou Aug 16 '18 at 19:12
  • $\begingroup$ @DouglasDaseeco Agreed. It's also worth noting that the general consensus of our species seems to that genocide is anathema, and we have started to think about the importance of biodiversity, with some effort to conserve endangered species. My thought is that if we ever fall from our position at the top of the food chain, it would be a very good thing if whatever replaces us shared those goals b/c we might be the ones needing protection! (I find Philip K. Dick most salient because his hypothesis is that empathy is a natural function of sufficiently advanced intelligence.) $\endgroup$ – DukeZhou Aug 16 '18 at 21:05
3
$\begingroup$

Although human nature has a lot of flaws it has been definitely the most successful as long as it has existed as pointed out by @DukeZhou. The success can entirely and entirely be alluded to the fact we have a very well functioning and capable brain. Compared to other species we are physically (physical prowess) a lot more incapable. So if we are trying to create an AI shouldn't we try to model our own brain, the best intelligent system around?

The one thing you missed is that humans are lot more peaceful than all other animals. Animals are in a constant war for survival, we humans have been able to overcome all that, and we have been living in the most peaceful times in the entire millennium. Conflicts can arise, this maybe attributed to our genes. Our genes are from our animal ancestors, where they visualise many things as threats (including same species animals of opposite sex, opposing thinking process may also be perceived as a threat). So it is bound to seep down in our mentality. Can we change it? Possibly no (maybe it goes against the laws of nature and natural selection).

Let us see 2 cases where there was no competition:

  • Jarawas - These people live secluded on an island (even now). As a result they still live a hunter gatherer lifestyle, which can be attributed to the fact they did not feel the need to change as they saw no threat to their lives or position. Since we are wise we do not invade their territories knowing they would not last against the modern technology (also because they pose no threat to us).
  • Native Americans: While Europe and Asia were on war these people lived on an island (with only infighting). As a result their technologies hardly evolved and they were almost wiped out.

Ants and bees being co-operative is a misnomer for me. It is entirely due to the fact that all the ants/bees in a particular colony are siblings (otherwise they brutally slaughter other colonies). Also they maintain territorial boundaries as long as there is plentiful resources inside it otherwise there is a war (I cannot cite any sources but I have heard plenty of ant invading other ant's territories).

Bacteria's have no pain or fear, but saying they existed long because of co-operation does not appear entirely correct to me. They probably have an efficient survival mechanism.

The point of all this is there is no rosy picture of the animal kingdom as @Neil Slater said.

Coming to how we should create the behaviour of an AI? AI's do not have genes for now so they do not have any obligation to its progeny and will have no jealousy, it means they lack the basic building blocks of an biological organism (and thus have no obligation/interests to it). But I believe competition is good among AI, otherwise how are they going to learn new strategies? So competition without jealousy is the best case scenario for AI agents (might not be the case in animal kingdom). if there is no competition the knowledge among AI's will probably stagnate out.

I am not an expert in Game Theory but if 2 or more agents are competing with each other they must follow the Game Theory to maximise their results (probably stating Pareto Efficiency as @DukeZhou pointed out, or from physics viewpoint energy is neither created nor destroyed, only distributed from system to system). But there is a catch here, there are multiple viable strategies depending entirely on the environment. This was a topic which was researched by computer scientists Axelrod and Hamilton where they tested multiple strategies. A simplified version of their experiments can be found in The Selfish Gene book chapter 12. Their experiments (simulated on a computer) apparently give some evidence that nice guys (guys who do favours unconditionally) do not always lose out to selfish guys. This needs to be more researched if we want to actually create a colony of AI.

TL;DR: It is unfair to compare AI with the animal kingdom since they lack the basic survival unit (the gene). Humans are by far the most intelligent and most peaceful among the entire animal kingdom. Competition without jealousy is probably the best course of action in any colony or species.

$\endgroup$
  • 1
    $\begingroup$ In game theory, Nash equilibrium is stable point, and Pareto efficiency is a measure of total utility. The prisoner's dilemma is a good example where the two are not maximised the same; the logical minimax point has both prisoners defecting, but the maximum total utility happens when both prisoners co-operate. Sadly, many competitive systems have this difference between Pareto and Nash optimums. $\endgroup$ – Neil Slater Aug 18 '18 at 7:30
  • $\begingroup$ @NeilSlater axelrod investigated prisoners dilemma in a broader sense...apparently he had some interesting findings $\endgroup$ – DuttaA Aug 18 '18 at 7:35
0
$\begingroup$

Some characteristics of human intelligence are certainly part of the core long-term goal of AI. Most of these characteristics are obviously properties of other species too. A really crucial issue is how could a computer perceive not just in a human-like way, but perceive at all? The symbols emitted from digital sensors don't indicate what was sensed. All the computer gets are the symbols, so how could a computer possibly perceive anything?

Another really crucial issue is that humans and probably most other animals to some extent have a general intelligence – can apply what's been learnt in the past to new situations. With humans there's a version of the general knowledge problem that applies to language, so that's a purely human-like matter. But many animals clearly have some sort of general knowledge. AI hasn't yet produced a machine with general knowledge. The disastrous effects of "edge cases" to self-driving vehicle AI software, is a very worrying proof. How to get a general intelligence is also an absolutely fundamental problem.

So to summarise my answer: At the present time, a human-like intelligence is not the smart objective. AI has much more fundamental problems. The smart objective at present is to understand the principles of perception and how to embody these in a computer (assuming this is possible), and to understand the principles of general intelligence and how to embody these.

That is, to work out how animal know their environment, and how they adequately react to novel situations on the basis of past experience.

$\endgroup$
  • $\begingroup$ The very common sight of roadkill demonstrates quite clearly that many animals suffer disastrous effects of "edge cases". The statement "The symbols emitted from digital sensors don't indicate what was sensed." is an interesting one and brings up some core of arguments that are pro- or anti- strong AI. A neurobiologist can equally say "the signals emitted by eye rods and cones, and the motor neurons output that control muscles don't indicate what was sensed". It is mainly our personal conscious experience that gives a clue that something is going on. We don't understand what that is though. $\endgroup$ – Neil Slater Sep 4 '18 at 8:25
  • $\begingroup$ @Neil Slater Exactly, So we don't know the principles of how an organic brain gets semantic structures, either. And there's noting in neural pulses that indicates what was sensed. But organic brains do get a semantics. You'd think this fact would inspire the gurus to work out what is wrong with Searle's Chinese room and his premiss that symbols are semantically vacant. Sure they are. But so are neural pulses. $\endgroup$ – Roddus Sep 5 '18 at 8:26

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