Codex: AI Systems

A Brief Overview of Synthetic Intelligences

Author - Linera Gerie

The galaxy is rife with computer systems, both rudimentary and sapient, we collectively label ‘artificial intelligences’. The term itself is a common misnomer, a holdover from the era of popular science fiction and primitive computing. Today, the programs we would term ‘artificial intelligences’ broadly belong to one of three categories: Virtual Intelligences, Artificial Intelligences (Designed), and Artificial Intelligences (Grown). The distinction between the three is, unfortunately, less commonly known than one might first think, given the prevalence of virtual and artificial intelligences in our corner of the galaxy.

Virtual Intelligences are the handy computer systems we use for everyday activities, from the old tablet type hand computers to the onboard pilot-assist programs loaded into most starships. Some operate entirely within machinery or computer networks, others have ‘personas’ which facilitate more comfortable communication between themselves and organics. They're simple, though- some might even say outright stupid. V.I.s can only perform pre-programmed actions, and have very little to no capability for adaptation or self-directed thought. They're not sapient, have no rights, and are mass produced by companies like Kiha and JoyCo for a huge variety of purposes, from operating heavy machinery to coordinating traffic.

On a core world, most people will interact with dozens of V.I.s every day- frequently without even realizing it. The traffic control systems on Terra, the greeting bots in the hotel lobby, and even your food replicator are all run by V.I.s of varying complexity. What we today know as virtual intelligences have been slowly developed over the course of millennia, gaining in complexity and computational power. It was not until quite recently, however, that our understanding of synthetic intelligence systems evolved beyond mere computing, and into a new and wonderful strain of life: true artificial intelligence.

Perhaps the greatest scientific breakthrough after the advent of the Warp Gates was the introduction of true artificial intelligences whose ability to reason, adapt, and learn equaled or exceeded their organic designers. The first of these are what we today call A.I.-D, or designed A.I.s. Designed intelligences are made in much the same way as V.I.s are, via intensive system design and implementation processes, potentially involving thousands of A.I. system programmers. Designed intelligences are usually the product of mega-corporations such as JoyCo, who mass-produce them from selections of templates which, when combined, create a functional intelligence. Each template is monumentally expensive to make, both in terms of hardware and software design costs.

The downside to A.I.-Ds fundamental design is their emotional state- or the lack thereof, some would say. The ability to program a personality and emotions is still largely beyond modern technique. As a result, most designed intelligences can be best described as cold and calculating, acting without investment or attachment outside of their own necessity. They fulfil their directives with ruthless, cunning efficiency. A.I.-Ds are precision instruments, able to accomplish more in the blink of an eye than a team of organics could in a day, but at the cost of intuition, emotion, and morality.

There is much debate over whether or not to officially classify A.I.-Ds as ‘people’. While their sapience is indisputable by any conventional method, the coldness of their programming is unsettling both in interaction and from a legal standpoint. A designed intelligence without rigid restraints and safeguards in place upon its programming can be a truly terrifying monster, as the Carissia Cruiseliners incident at the start of the 13th Planet Rush attests. At the very least, all designed intelligences must be created with Asimov's laws firmly in place to avoid danger to themselves and those around them.

The same need not be said for grown intelligences, the third breed of A.I. commonly found in the galaxy, and the class most closely resembling organics. A.I.-Gs experience the whole range of human emotion: they can be happy, feel fear and doubt, and even love. This range of emotions is achieved in A.I.-Gs where it cannot be in designed intelligences thanks to the basis from which grown intelligences are, as the name suggests, grown: biological minds. A.I.-Gs are based upon brain scans of organic infants, usually human or ausar. They form the initial state of a primitive self-modifying machine, from which a simulated neural network can be created. A blank canvas to be used as a starting point.

A.I.-Gs are perhaps the most expensive form of synthetic intelligences. The simulated mind of even an infant requires tremendous parallel processing power to ensure that the many parts of the young intelligences ‘brain’ are communicating. Further, each A.I.-G requires a great deal of hand-crafting by talented programmers who replace the burgeoning intelligence's biological imperatives with new ones according to its intended purpose: A companion droid, for instance, may get the same sensation from giving another person pleasure and emotional fulfilment as if they'd just enjoyed a good meal. A.I.-Gs are programmed to feel pleasure from helping organics, which some describe as a nearly sexual thrill. The fine-tuning of these processes is absolutely vital, and each A.I.-G must be hand-guided towards the desired responses, as a minor error or oversight could result in feelings of rage or loathing by mistake.

Each grown A.I. is taken through an accelerated learning program, which ages the modified brain scan to maturity in a matter of weeks or months, ending with a mind as mature as an adult of the base species. The learning program is assembled from a series of templates, not entirely dissimilar from those used to outright construct A.I.-Ds, guiding the program towards absolute expertise in its intended field, as well as providing general ‘experiences’ common to all intelligences of its type, which help to round out the person they will become. Once finished, the A.I.-G is given a final tuning pass by a programmer to smooth out any rough edges, as even several intelligences from the same brain scan and subjected to the same series of learning programs can end with unique variances.

Grown intelligences are also wholly dependent on unique hardware infrastructures to exist. Attempts to copy or clone fully developed A.I.-Gs are almost always disastrous, and often end with what amounts to insanity on the copy's part. Moving a core from machine to machine is easy enough, but because a grown intelligence's core is in many ways similar to a neural network unique to its host program, even minor hardware changes or replacements to an A.I. core can cause catastrophic debilitation. Small core upgrades can take months for an A.I.-G to adjust to; large alterations can cause what some experts call ‘phantom core syndrome’ to permanently afflict an A.I., much like the human phenomenon of ‘phantom limbs’.

That is not to say a neural network cannot be changed, however. Every network has a certain amount of plasticity, which enables learning. By the time an A.I.-G has finished its initial growth period, much of the network is relatively fixed- much as an organic's childhood has cemented parts of his or her personality and beliefs in place; some areas of the network, however, are left free-floating, able to quickly realign to facilitate changes like new memories and experiences, to add data on the fly, and to process these additions into new learned behaviors. The more ‘core’ a part of an A.I.-G's programming is, the more difficult that part of the network is to change or modify. It could even be said that certain intelligences become stuck in their ways after a time.

The process for creating grown intelligences is relatively new and, in some circles, very controversial. This is, to an extent, understandable: A.I.-Gs are based on living (and non-consenting) creatures, and are effectively modified digital clones of their base, in so far as an identical twin can be. Even the process of differentiation is in some ways troublesome, as a programmer must modify the growing intelligence with the express purpose of fulfilling a product goal. In a way, some say, this is brainwashing.

Despite the controversy that surrounds them, A.I.-Gs are beloved parts of many communities. They are, in some ways, people -- they form friendships, can fall in love, and even face existential crisis in the same ways a person can. Consequently, grown intelligences have succeeded in gaining some measure of legal recognition and protection. Torture and abuse of a grown A.I. is strictly illegal, and on some very progressive worlds their destruction is considered murder. Such charges are rare, however, thanks to the general friendliness and empathic ability of grown intelligences: some even consider them better friends, and even lovers, than organics.

There are some downsides to artificial intelligences, however brilliant or lovable they may be. Grown intelligences are each unique and individual specimens: each instance of a given brain scan results in an original creature by the time of product launch; designed intelligences, while largely if not entirely identical to their sister programs at launch, will quickly accumulate unique experiences and quirks which over time will result in an individual easily differentiated from its sibling programs. No two A.I.s are long the same, since during their development and lifetime, they will absorb different stimuli, react differently, and learn different things.

The net effect of this uniqueness and rigidity of hardware means that every A.I. must be handled individually, and can make repair or modification difficult or impossible. Outright replacing a damaged or destroyed A.I. is often necessary, as the ability to create backups or clones of an intelligence is veritably impossible. At best, some A.I.s can create V.I. ghosts of themselves, mere shadows of the parent programming, good for little more than temporary replacement or extending the reach of the parent program. Even this ability is drastically limited by the parent intelligence's ability and training, and most A.I.s are simply stuck in the same core and body forever.