- "Each of us represents a single trans-human mind. Each of us is an undrawn map – not a mere physical reconstruction of an object as it was in life – but full, human potential realized in crystal and thinking at the speed of light!"
- —One of the AIs of the Assembly.
A smart AI or "smart" AI is the colloquial term in the United Nations Space Command for artificial intelligence (AI) which have no limitations in their dynamic memory-processor matrix, meaning they can not only be taught a vast wealth of information, but can learn and comprehend from their surroundings. This is unlike "dumb" AIs who can only learn a limited set of topics.
The first smart AI was created in the mid-21st century. At first, there was great concern that this new technology would become too capable and render human intelligence obsolete. The ability for vocal expression was added to make smart AIs less threatening, allowing them to converse on the same level as their masters. Over time, such AIs became more advanced. Due to their uninhibited matrix design, "smart" AIs proved capable of intellectual development. This is manifested in that they can actively learn from and adapt to situations and events. Conversely, their counterparts, the so-called "dumb" AIs, can not; while extremely competent in their designated tasks and fields of expertise, they are 'merely' highly-advanced computer programs and are not capable of intuition. Like all AIs, smart AIs use a Riemann matrix for higher-function processing. This matrix can be downloaded into a data crystal chip for mobility. In case of rampancy, the Riemann matrix includes a fail-safe program that will destroy the AI.
Smart AIs are subject to Isaac Asimov's Three Laws of Robotics, a series of rules set up in order to prevent the AI from causing harm to humans or itself. Since many AIs are used for military purposes, they are capable of ignoring the First Law, but only when operating at full capacity; when their processing capabilities are diminished - say, by being cut off from the core of the ship or facility they are stationed on - they are incapable of bypassing the Law.
Every "smart" AI represents itself through a holographic avatar. The appearance of these avatars are usually derived from the physical traits of the person they are created from; for example, Cortana resembles a younger version of Dr. Halsey. Avatars can range from the complex and theatric, such as those of Araquiel and Deep Winter, to simpler human forms such as those of Melissa, Cortana and Serina.
Dr. Catherine Halsey worked extensively with AIs and oversaw the construction of all their programming. She has an unparalleled knowledge on how AIs work and also created the upgrades for the MJOLNIR armor so it would be capable of carrying an AI. She and her team successfully managed to create a neural link between a human and a smart AI for the first time. Through a neural interface, the neural network of the AI and the low-order structure of the human brain would be effectively connected. This was first tested by John-117 and Cortana in 2552. Dr. Halsey also had her brain flash-cloned, which in turn were scanned to create the AI Cortana.
Despite the fact that "smart" AIs are essentially trans-human minds, they are treated as little more than tools, and their activities are closely monitored. Due to their superior intelligence, some "smart" AIs secretly consider themselves as the "next step" in human evolution. By 2310, a group of "smart" AIs had formed the Assembly, a collective that worked to preserve the human race, without humanity's knowledge. The group existed for centuries, and was ultimately behind many important events in human history.
Instead of simply "programming" a smart AI, their matrix is created by sending electric bursts through the neural pathways of a human brain which are then replicated in a superconducting nano-assemblage in a process known as Cognitive Impression Modeling. This process, performed by a device known as the AI matrix compiler, destroys the original brain tissue, and so the brain being used is typically obtained after the host is dead. The brains used to create AIs are typically acquired from voluntary donors, although Durga implied that many AIs were the result of kidnappings by ONI "brain ninjas". In the case of the AI Cortana, however, the candidate brain was flash-cloned and the host's memories transferred, so that the host, in this case Dr. Catherine Halsey, could remain alive. However, this is extremely inefficient, both from the degeneration of the flash-clones and the high rates of schizophrenia and dementia resulting from the memory transfer process. Out of Dr. Halsey's twenty cloned and enhanced brains, only two of them were sufficient and without deformity. Cortana is the only AI known to have been created this way.
The human neural patterns are used merely to "seed" the initial AI matrix; over time, the AI develops its own system of more efficient neural linkages. This can be considered one of the drawbacks of smart AI creation; initiation time is wasted for the AI by creating the first neural linkages, each of which gradually erodes the AI's limited lifespan. It is technically unclear exactly how a "smart" AI works, as it is only based on the functionality of a human brain, which even UNSC scientists do not fully understand.
Because "smart" AIs are created using either actual human brains or cloned human brains there are often times residual thoughts, memories and/or feelings that remain. These residuals can be anything from the "feeling" of a hair brush being pulled through hair (in the case of Sif), to an effect on the mannerisms and characteristics that make up the personality of an AI, in the case of Cortana and her likeness to Dr. Catherine Halsey as well as her possessing many of Halsey's memories, which surfaced during her descent into rampancy.
Another feature which stems from their unusual nature is a limited lifespan. Dr. Halsey compared the life cycle of a smart AI to that of a star; they spark, burn brightly, then either die by the crush of gravity or quickly terminate in a chaotic eruption.
Unlike dumb AIs, smart AIs not only accumulate knowledge but also grow in their ability to draw conclusions from an incomplete dataset - similar to human intuition or creativity. Over time, this causes them to form more and more cross-linkages within their processing matrix. This increases their cognitive abilities, but also shortens their operational lifespan; with such an excessive overload of data their function under normal parameters is no longer possible.
The interconnections continue with two possible outcomes: the sheer density of linkages causes a cascade of irreversible quantum transfer (comparable to a short circuit), eventually halting all functions, or the AI may take corrective action and begins to preemptively sever neural linkages through a voltage overload. This often results in a condition known as rampancy, which is characterized by aberrant manifestations of personality. This eventually leads to the AI eliminating more and more vital connections which in the end proves terminal, and essentially kills the AI system. An apt analogy to this would be a human brain devoting so much of its work towards thought that its neural signals to vital organs cease. In this way the AIs quite literally 'think' themselves to death.
A number of solutions for a smart AI's short lifespan have been proposed, none of which are known to have been effectively implemented.
Dr. Catherine Halsey experimented with a theoretical architecture of three smart AIs arranged in parallel. This "triumvirate" would make consensus decisions of which neural links to cut when approaching rampancy, as well as divide algorithms to compensate, resulting in more stable and coherent operation, possibly even faster processing. The creation of neural linkages would also be decided by consensus decision, resulting in more controlled development.
In the architecture of a smart AI, space is also an obvious limitation. Halsey attempted to overcome this by creating a hyper-dimensional processing matrix within the eleven-dimensional spacetime of Slipstream space. Theoretically, slipspace would allow limitless cross-connections and faster-than-light processing speeds. In 2547, she attempted to test this in practice using a decommissioned Shaw-Fujikawa Translight Engine, but the experiment failed. 
 Known UNSC "smart" AIs
 Third Generation
- Araqiel - Colonel Ackerson's personal AI
 Fourth Generation
- Black-Box - An ONI AI and assistant to Admiral Margaret Parangosky and Captain Serin Osman.
- Wellsley - UNSC military AI.
 Fifth Generation
- Deep Winter - Overseer of UNSC facilities on planet Onyx, succeeding Eternal Spring.
- Endless Summer - MIL AI 4279 - Overseer of UNSC facilities on planet Onyx, succeeding Deep Winter.
- Beowulf - Was the attaché to Admiral Michael Stanforth and the UNSC Navy.
- Chauncey - UNSC Red Horse's old smart AI
- Cortana - CTN 0452-9 - UNSC military AI
- Doppler - MIL AI 8575 - UNSC military AI
- Eternal Spring - First known overseer of UNSC facilities on planet Onyx, aided in reconnaissance of Zone 67.
- Iona - AI attached to SPARTAN Black Team.
- Jerrod - The first experimental "micro-AI".
- Juliana - Commercial AI operating in Madrigal, unpacked after the planet's destruction by the Covenant to control the Rubble's guidance and maneuvering.
- Kalmiya - Dr. Halsey's former personal AI, Cortana's "older sister".
- Lackluster - corroborated conclusions of Admiral Cole's continuing survival with Codename: SURGEON and another AI, Phoenix
- Loki - Colony ship AI (military) of the UNSC Skidbladnir. Reassigned as PSI (Planetary Security Intelligence) on Harvest, second AI core on stand-by next to AI Mack.
- Lorelei - Ship AI of the CMA Season of Plenty
- Lysithea - UNSC/ONI AI, was located in the UNSC High Command's headquarters, HIGHCOM Facility Bravo-6 in Sydney, Australia.
- Mack - Colonial Administration Authority AI/Agricultural Operations.
- Melissa - UNSC/ONI AI, attached to UNSC Apocalypso.
- Mo Ye - Shipboard AI of the UNSC The Heart of Midlothian.
- Pallas Athena - Boston Public Library AI
- Phoenix - corroborated conclusions of Cole's continuing survival with Codename: SURGEON and another AI, Lackluster
- Rebecca - UNSC Red Horse's smart AI
- Roland - UNSC AI, Shipboard AI of the UNSC Infinity as of 2558.
- Sekmet - Shipboard AI of Preston Cole's flagship, UNSC Everest.
- Serina - UNSC AI, served aboard the UNSC Spirit of Fire.
- Sif - Colonial Authority AI/Shipping Operations.
- Solipsil - An ONI AI.
- Toran - UNSC AI, served aboard Han during the SPARTAN observation process.
- Watchmaker - UNSC military AI for Captain Cole aboard the UNSC Gorgon.
- Smart AIs' lifespans are about seven years long, this is another reference to Bungie's favorite number, seven.
- Some smart AIs have avoided rampancy or have at least not shown it, such as Juliana.
- Due to their imagination and limited lifespan, smart AIs are intrigued by existential philosophy, such as Zen koans.
- The collective of smart AIs known as the Assembly existed for centuries, far exceeding the lifespan of a smart AI. However, given the fact that no specific members of the Assembly have been mentioned as having taken part in more than one of their sessions, it is possible that the group's composition changed regularly, with new AIs being introduced to the collective as older ones were terminated or succumbed to rampancy. Alternatively, given its Majority-Minority composition, the Assembly may be a working implementation of the hypothetical AI architecture theorized by Dr. Halsey in her journal, in which AIs are capable of extending their operational lifespans by means of a more deliberate control of neural linkage creation via majority vote. Since Halsey formulated her theory independently while the Assembly had been active for centuries, however, this application or the technology behind it would have had to have remained secret from most of humanity.
- ^ a b Halo: Contact Harvest, page 30
- ^ a b Halo: The Fall of Reach, page 235
- ^ Halo: Evolutions - Essential Tales of the Halo Universe, "Midnight in The Heart of Midlothian", page 88
- ^ Dr. Halsey's personal journal, March 24, 2549
- ^ a b c Halo: Reach, Data pads
- ^ Halo 4, campaign level Composer, Audio log
- ^ Dr. Halsey's personal journal, May 21, 2549
- ^ a b c Dr. Halsey's personal journal, May 3, 2526
- ^ a b Dr. Halsey's personal journal, November 10, 2533
- ^ Dr. Halsey's personal journal, December 25, 2534
- ^ Dr. Halsey's personal journal, September 4, 2547
- ^ Halo: Reach, Data pad 17
- ^ Halo: Ghosts of Onyx, page 178