Why would combat robots with DNA-based biocomputers tend to go berserk?

Question

In my universe, there exist these canine robots called the Kynigóskylo-Types created to hunt and kill specific species of mutated animals. Kynigóskylos get controlled by DNA-powered bio-computers (which use the four DNA bases for quaternary calculation rather than binary, making them far faster) mimicking the brains of grey wolves (Canis lupus) enabling Kynigóskylos to devise strategies and cooperate in exterminating targets in combat.

However, during their testing phase, the organization responsible for developing the Kynigóskylos discovered a multitude of severe flaws with the Kynigóskylos and the most problematic was a tendency to display animalistic hyper-aggression when under great duress which made them borderline difficult to control.

Thus, said organization chose to use modified Kynigóskylos as training units for soldiers apart of their ranks instead of fielding them.

What caused these robots to go haywire in the first place?

Answer 1

Bad engineering.

Brains are imperfectly understood and complex. Throw in that they were modifying them, and specifically to improve their attacking abilities, and it's easy to see that their efforts threw the aggression off balance, especially when this occurs under great stress. They omitted an influence of self-preservation that curbs it.

It is a subject of great debate among the makers about which part should have been diminished or increased to produce a more stable version.

Answer 2

a tendency to display animalistic hyper aggression when under great duress

DNA, as shown during its usage in carbon based lifeforms (CBL), is prone to mutations, but at least the CBL have a repair mechanism in place.

Stressful situations have a bad impact on the DNA repair mechanism, even worse if those repair mechanisms were not implemented in the mark-1 version.

The more the DNA based computation is used, the more the errors are propagated, and again stressful situations require more calculations. Top this with the selection of mutation favoring those which still do something vs those which simply freeze the subject.

Answer 3

Three Laws? What are those?!?!

As the scientists in charge of making your killbots have discovered, it is incredibly difficult to instill basic rules into an organic computer without going too far. Since "going too far" is a synonym for "absolute lack of free will", which is something that's rather bad in something you're constantly exposing to chaotic situations, they have given up on trying.

As a result, these killbots have no form of the Three Laws of Robotics.

The best that scientists have been able to do is to "raise" them in an environment where they are, shall we say, encouraged to always follow orders without question. Now that I think about it, this is how militaries usually work anyway...

However, this careful training tends to break down when they are exposed to situations of great stress.

Being killbots, they lapse into full Dalek mode when this happens.

EXTERMINATE!!!!!

Answer 4

Old code.

Your bots are supposed to act like gray wolves. Their brains are not wolf brains, but are based on code that is supposed to model wolf brains. The code was not written new for the project but was based on simulated wolves used for an ecological modelling program. These wolf sims did a fine job of modelling how wolves hunt and make decisions.

Some of the wolf model code for the ecological model was taken from an earlier project which modelled how various quadrupeds move at different paces. The code used for that project was in part lifted from an earlier game in which persons in a post apocalyptic wasteland fight zombies, bears, wolves and other foes; the game (now more than 30 years old) won acclaim at the time for the exquisite detail that was put into the lifelike motion of various foes.

Your bots have inherited some aspects of their distant CGI ancestors. In the game, healthy wolves in numbers would attack a lone human, but a single wolf would flee from a healthy human. A single wolf might attack a wounded human. A wolf cornered and unable to flee would attack. It is this latter code your bots are executing.

It would be a simple matter to excise this piece of code. That is it would be if they could find it - not very many people still code in the old language used for the game.

Answer 5

If they are true bio computers then try simple viral infection or similer. If the 'bio computer' involved requires oxygen, nutrients and water etc to maintain functionality and one or more of these inputs is regularly 'ingested' by the robot while in the field as a means of extending its range/patrol time (as opposed to running off a finite onboard store) then there is a risk of the robot ingesting random viruses, organic nano-articles, bacteria & fungal spores etc while on patrol.

In addition the on-board filtration systems are simply not designed to deal with the heavy workload they end up facing. They work perfectly well under laboratory conditions or during uneventful patrols in urban settings BUT once in the 'field' things change. Faced with exposure to harsh/varying climate conditions and other random environmental factors including the physical stresses of combat they break down.

This means the random micro particles mentioned above can reach the bio chip. The particles/virus etc may actually have no ability to damage the computer but it's mere presence/interference is still interpreted by the computer as an attempted 'attack'.

The computers fallback defense programming then kicks in and it immediately goes into 'seek and destroy mode' futility attempting to locate the source of the attack. Since there is in fact no 'attacker' to locate it simply moves on, staying in seek destroy mode until it is either destroyed itself or someone, somehow manages to incapacitate it long enough to 'flush' the support systems and brain clean of the foreign material.

Answer 6

First some half-meandring information on your idea, go to "long story short" for the answer.

Because DNA continuously breaks down and might not always be available, its also relatively slow.

Thought processes are complex (citation needed), and yours are one based on a quarternary calculation system. But its not just one DNA sequence doing the work, its millions doing sequences and parts of calculations that monitor the body, its surroundings, collect data to form idea's on what it should be doing and how, which muscle groups to activate and what sequences, stored data that needs accessing to recognize what is present in the surroundings and to apply previous experiences and current mood and tasks of the creature to come to a conclusion. So you need a robust system that can correct a few errors on-the-fly.

Unfortunately DNA isnt that flexible. It's incapsulated in the ribosomes and communicates by splitting partially at the right spot, generating messenger RNA (I think it was called that) which then transports itself through the ribosomes and... somehow through a process we dont understand and seems to be based on random chance to eventually reach its destination... reaches the right parts of the cell at some point to have the cell perform its tasks. This complicates matters enormously: first something has to get into the cell and to the DNA to pass the information it needs to start its calculations. After the calculations it needs to generate RNA in order to communicate its answer, and hope it moves to the edges of the cell fast enough to get the next Cell's DNA in line to start the right calculations, hoping its the right cell and that the surrounding cells process their messages in a timely fashion as well despite the randomness involved. (Note: I tried looking but neither my textbooks nor the internet actually mentions how a cell processes information, they all focus on the transmitting between neurons. I cant find if the process takes place in the main body or the dendrites, or that it uses electrical or chemical processes to process its information before it sends it on. I have no idea why that gap is there and it drives me crazy).

On top of that, when a cell is dividing the DNA is duplicating and unavailable (unless like neurons it does not divide and relies on a type of stemcell for replacements), and DNA degrades over time with use and divisions. The body does try to repair it but even when it successfully does so that part of the DNA has created faulty results that needed to be corrected, and eventually these cells will have too many DNA faults and break down/be terminated by the body and replaced.

Long story short:

Your creatures use neurons to send the information, but have a data processing system that is great for long-term thoughts and planning but not that handy for quick short-term thinking. So when these creatures activate their bloodlust and get to killing things it takes too long for their brain to tell it to stop again. Under duress the amount of information the DNA has to process in quick succession will cause information loss and information mingling as the RNA will not always arrive at the cell body's edges and be send on in the same order that they were generated.

Answer 7

They occasionally go haywire because they're not designed properly. Their minds are the product of engineers working through several layers of machine learning and tools designed by machine learning tools. That is to say, none of the code that's actually running in their brains was written by a human, or even by a tool that was written by a human. The engineers have neither control nor knowledge of what's happening at the bottom level. The actual operation of their minds is just a little bit outside the capacity of human comprehension, or at the very least, outside the limits of what can be comprehended on this development budget. The result is a mind that is mostly serviceable. Mostly. Perhaps in theory, the flaw could be tracked down and corrected, but it would certainly require much more money and time than the manufacturer is willing to spend.