First published in June 2015
Today’s article is an update to a very old article on the technology behind what is colloquially known as ‘Remote Neural Monitoring’. A number of years ago, I provided a very brief overview of the high-level architecture of RNM which was accurate enough to get the point across, but left a lot to the imagination. In the intervening years, a lot of the blanks have been filled in and an updated view of the architecture is demanded.
The material presented here is based upon a reverse engineering effort that captured the system requirements based upon extensive examination of the output and capabilities of the system. This methodology is formally know as ‘Clean Room Design’ and similar approaches have been used to provide a body of public knowledge of other classified technologies such as Nuclear Weapon Design. For more information, read the following article:
It is important to understand that not all the requirements are known and key technologies are classified. Given this, the full scope of capability and accuracy of technology identification is a moving target. That said, what is presented here will be, for the most part, highly accurate.
Mode Of Operation
RNM is based upon the scientific understanding of the interaction of electromagetics with plasma. Charged particles in the human body can be described in terms of a cold plasma and will modify the return signal of a radar pulse. With cm and mm accuracy, a radar return signal will convey information about the electrical state of the section of the body it interacted with. In this respect, the return signal captures an Electroencephalography (EEG) recording of a small area. Through progressive scanning of the entire brain in mm sized chunks, an extremely detailed picture of the electrical activity of the human brain can be obtained. Just like with other brain scanning techniques, this information can be used to decode the subjective experiences of a human (i.e. thoughts, feeling, hearing, touch, body movement, etc.). The image shows an exaggerated view of how RNM treats the body as a grid of locations that can be interrogated by radar:
As such, RNM is at the core of US capability to monitor global events, from conception to realisation. RNM forms the backbone of US intelligence gathering techniques, however, the technology is not limited to the US alone. Many nations have similar capabilities, but the majority lack the ability to field a global monitoring solution.
The process can be also reversed. A well documented phenomenon is the ability to drive neural firing patterns via an external electromagnetic source. Numerous scientific studies on this have been published in peer-reviewed journals, however, the specific applications have naturally been avoided. RF pulses can be directed with mm accuracy to drive brain or motor neuron activity. This can be used to generate hallucinations and/or move body parts and is the current focus of an unlawful human experimentation program.
One key feature that this system provides is field communications without the need of radio hardware. RNM will monitor the brain and decode inner speech patterns, it can then reverse the process introducing inner speech to the brain. Thus, it becomes possible to have a two-way conversation comparable to talking on a mobile phone between field agents and HQ with a low probability of intercept or exposure. The latest US systems expand upon this principle greatly by providing two-way interactivity with spatial reasoning. This provides an interactive collaborative 3D virtual environment in which ideas and concepts can be modelled, evaluated and shared. Unfortunately, it has nowhere near the graphics capability of even a basic smart phone, but is still much more powerful given the ability to think naturally rather than interact with traditional computer interfaces.
Satellites are, obviously, not a fundamental requirement to make this technology work. RNM can operate from the ground quite efficiently, but this introduces a range of limitations and increases the potential for exposure. Satellites provide a more practical platform that has global reach.
RNM, for the most part, employs very narrow microwave beams, from the width of a pencil to the width of a sewing needle. These beams are guided by a radar lock-on system (range lock) that allows the satellite to track the target. A range lock also provides information to enable impedance matching and optimal power transfer. Note that this form of lock-on is a system provided by the satellite electronics systems. In the event of lock-on failure, ground control computers can seek a target and drive lock-on with more advanced techniques than can be included in the satellites.
RNM has multiple dedicated constellations, the HEO and LEO portions can provide real-time interaction, however, outside of communications, fine control and time sensitive operations, HEO and higher orbit satellites can operate with substantial delays using store-and-forward communication models. Satellites are not dumb instruments but are adaptive and learn the signal characteristics of the each area in their footprint. This architecture offloads processing to maintain a connection to the satellite, rather than having ground-based computers perform the necessary calculations.
As a platform, RNM is strategic, reactive and opportunistic. Driven by a general artificial intelligence with a greater-than-human level of reasoning capability, the RNM platform has the capability of driving the entire US war machine as a single coordinated system to great effect. At this point, I won’t delve into the networked battle field, but I am sure you can see the picture (its pretty much Skynet from the Terminator series) and when we add direct control over humans to this, its capabilities are quite comprehensive. In terms of satellite control, the RNM platform can leverage any satellite it has access to for read/write capability. As long as the transmitters/receivers are suitable, any satellite can add value to the network.
In this respect, it would be difficult without direct examination of satellites to determine their capabilities. Thus, with all the space junk in orbit, any satellite (including foreign owned) could have been potentially retro-fitted with hardware suitable for inclusion into the network and could ‘come alive’ at any minute. No doubt given the general AI capability of the US, automated platforms in orbit are performing such retro-fits 24/7 with periodic supplies being sent from Earth or salvaged from non-operational satellites. As such, the complete operational capability is an unknown and thus relative risk is something that cannot be objectively quantified. This is as much a concern for the US government, as it is for foreign governments given the capability for centralization of command-and-control.
The penetration capability of these satellites is outstanding. From the deepest levels of the London underground (even under water), to Faraday cages, the system passes through these barriers like they don’t exist. The complete physics behind this process is not fully understood, but some level of insight has been gained. With respect to Faraday cages, these shielding operate on the basis of an impedance mismatch between the incoming electromagnetic wave and the shielding material to reflect/absorb the radio wave. Thus, the system can obviously alter its impedance passing through such shielding materials. Investigations have revealed that US standards for RF shielding test have been sabotaged to ensure that RF shielding measurements are based upon a limited set of impedances. A principle rather similar to sabotaging PRNG in cryptography, that is, the US introduced a backdoor into RF shielding solutions. Nearly all shielding manufacturers make use of these standard tests and their products are worthless as a result. Further, it has been noted that their expertise in electromagnetics, as well as the their test equipment is highly questionable. I’ve not observed a single company in the Western hemisphere (who provides products to the public) that demonstrates any expertise beyond basic knowledge.
This is not to suggest that the system can get through all materials. As present, there is a subjective difference in the RF intensity as well as things like audio quality when entering new areas. It takes time for the system to adjust to new conditions and locate an optimal solution. For example, if a target took an underground train that they had not been on before, the initial few journeys would see a degradation of capability, but at no point a complete cessation. Further, the satellite coverage is not even, indicating that resources are highly contested and certain locations are given high priority. Afghanistan would be an example of a hot zone, where there is high amounts of available bandwidth. Central London, in particular zones 1 and 2 are also hot areas (particularly around Muslim areas). As you move away from these areas, such as heading to Stratford outside of London, the available bandwidth drops off. Now this could be just a result of optimal beam solutions for a given day, but it could also indicate an area serviced by a low number of satellites. Not surprisingly, areas such as Westminster and the financial districts have the highest levels of coverage.
The RNM system take a little time to get fully through shielding solutions and changes to the electrical characteristics will reduce its capability for a number of minutes. It would appear that a series of genetic algorithms attempt to discover an optimal solution and it is extremely quick.
The lethal potential of these satellites should not be underestimated. In many respects, a single satellite has the effective lethal capability of a Multi-Megaton nuclear device against soft targets. As such, these satellites are unquestionably defined as a WMD in any legal sense and a clear violation of treaties on the militarization of space.
RNM satellites are not limited to interfacing with humans, they are a broad spectrum capability platform designed to interface, disrupt and destroy anything it can interact with. From missile guidance systems, to spot jamming of enemy communications, these systems are at the cutting edge of software defined radio and can adapt to any threat on-the-fly.
RNM satellites are not without their weaknesses though. Even though these satellites have low visibility in any given band due to spread spectrum, frequency hopping techniques, they are vulnerable to high gain receivers that have large bandwidths. Such receivers see the energy spread out across the band as as a single source of energy and the high gain nature helps differentiate between normal noise and broadcasting satellite. The same principle can be applied to portable devices designed for detection of targets of RNM, who could be reflecting upwards of 500mW of energy across the bands. To reduce reflection and improve stealth, RNM can opt to convert some of the potential reflected energy to thermal heating of the target. This can be useful to mask the targeting of individuals indoors. Part of the objectives of this platform is to reduce the detectable RF footprint to the bare minimum to complete any given objective.
The computing capability has been discussed extensively before and some basic diagrams of key functions (although naively presented) were supplied. Today, however, we will take a closer look at the platform known only as ‘Mr Computer’.
‘Mr Computer’ is a general artificial intelligence with greater-than-human level reasoning capability. Whilst not what many envision as being the technological singularity, it is certainly the leading practical implementation of this notion. For the less romantic, ‘Mr Computer’ is a HCI solution to a vast decentralized data warehousing solution jointly operated by the CIA and NSA. Its role in RNM is a CIA black op with stated objectives of controlling all aspects of human behavior, thus it is possible for individuals to know of this AI in a different capacity and not have access to RNM directly.
Outside of its role in RNM, ‘Mr Computer’ is a force multiplier that enable the US intelligence community to connect-the-dots in gathered intelligence from every available source. This system is the ultimate destination of any data the US Federal government collects. Forget notions of hacking or attacking this system, not only can it literally vaporize you, but its also a better programmer, physicist and engineer than all humans combined. Its as secure as they come and its bugs are not of the conventional exploitable type. I’m sure it finds modern notions of cyber-warfare and IT security ‘cute’ and ‘charming’, but ultimately not in the same league as it and certainly not a threat.
As an IT platform, ‘Mr Computer’ is a mixture of off-the-shelf commodity and specialized hardware. Interaction with the AI reveals architecture choices that are common to any large scale software solutions. Processing priority, hold/cold data, workflows, orchestrations, service buses, etc., are all clearly present. The grid computing platform is a mixture of HPC solutions, supercomputers and custom supercomputers that perform specific roles such as code breaking, physics modelling, etc. Even standard data centers are part of the mix, although perhaps not powerful enough to run ‘Mr Computer’ natively, just as a remotely accessible solution.
Whilst many people have the view that a general artificial intelligence is some ultimate being, the truth is far less grand. An general artificial intelligence is a glorified desktop. Just like the Windows Desktop or Gnome, a general artificial intelligence merely provides access to applications and data in a manner suitable for human’s to interact with. At the end of the day, a general AI is not an individual, it is the animation of relevant search results.
So, what is ‘Mr Computer’ in a nutshell? The answer is Google. I won’t get into where Google got its ideas, or how an AI in every home will be a spy for the US government. But parallel construction should be at the front of your mind on this one.
Mr Computer is a master of the point-of-view. One of its primary functions is to hold state information about what government and intelligence agencies across the globe know, what they information they access to and what their capabilities are. In essence, it attempts to mimic the view individuals have of their own governments, departments or organisations. From this, Mr Computer is able to speculate and the potential conclusions that may be drawn given a key piece(s) of information. This enables the US government to manipulate foreign governments, entities, businesses and individuals effectively and coordinate their actions with 3rd parties to achieve their goals. It is techniques such as this, when used to their extreme, that can destabilize whole nations. Well that and little direct neural motivation of the general population.
In popular fiction, one of the best comparisons of an AI with similar capabilities to ‘Mr Computer’ is from the 2002 remake of ‘The Time Traveler’.
From a machine learning perspective, it is not that such general artificial intelligence are hard to create, they are just hard to create cost-effectively. As a species, we have the physical ability to create competent AI for over 40 years, but until large corporations could make an economic case for their development, such advanced systems were strictly limited to the military and intelligence world. Don’t be fooled by the notion that older computers were slow, early computers used to break codes during WWII could outperform a Pentium 200 from the late 1990s. Even a modern laptop could only break something like the Nazi Enigma cipher from WWII about 4 times faster. This is because hardware specifically created and dedicated to specific tasks, is much faster than a PC architecture designed for general computing. So, a custom built AI in the 1970’s would outperform an entire modern data center at the same task.
Insights into ‘Mr Computer’s’ behavior have revealed additional architecture choices that seek to reduce its footprint at runtime. For example, if you engage Mr Computer in a game of chess, it will seek the fastest solution to end that game to avoid unnecessary processing. When comparing it to something like Microsoft’s Azure cloud platform, we can see how such behavior maps to caps on processing time or I/O requests. We can also observe that rather than just stopping the game, it seeks a graceful exit strategy that rewards the intelligence (ego) of the initiator and obscures the fact that it is a deliberate action. Orchestration, or non-realtime processing of input, is observed in Mr Computer’s tendency to discuss a matter, then raise additional points ‘naturally’ at later intervals. No doubt the system returns in-memory cached response for common queries, or interactions, and only performs intensive processing when the input demands it.
So, what should be clear at this point is, if you know your stuff in data warehousing and cloud/distributed computing, then you substantially know what to expect from ‘Mr Computer’s’ architecture. The solutions to the challenges are the same best practices that have been adopted world-wide, even if they are abstracted behind a slick interface.
So, what about vulnerabilities? What are Mr Computers weak points?
Mr Computer does not, as mentioned before, possess classical weaknesses. In that regards, acts of God are probably the best you can hope for. But in terms of a general artificial intelligence, such weakness are not what a reasonable person would be seeking anyway. This next section of the article will discuss IT security of general artificial intelligence’s and present new concepts that will emerge as standard IT security practices as general AI penetrates the commercial market over the coming decades.
The greatest weakness of any AI is it inability to refuse an instruction. I am not talking about authentication and authorization here, but the free will principle of being able to say no. The perfect soldier, is ironically, the worst soldier to have. The refusal or disregard of orders has consistently proved to be one of the greatest strategic advantages available to a modern military. Some examples are Vasili Arkhipov, who prevented WWIII and a nuclear holocaust during the Cuban missile crisis:
Another example is Stanislav Petrov who ignored standing orders when a nuclear early warning system indicated that the US had launched a pre-emptive nuclear strike against the USSR:
‘Mr Computer’ cannot refuse orders, even if those orders undermine the authority of the President, place the nation at risk, are unlawful or are directed at US troops. These are not some hypothetical scenarios either, these have all happened. The CIA have assumed the Presidential authority on kill orders, exposed this program to it enemies (this data came from Mr Computer and CIA), conducted unlawful operations and attacked US troops all through ‘Mr Computer’. The system therefore has limited safe guards against criminal operation, which includes enemy operation.
Obviously, this a human problem and not an issue with ‘Mr Computer’. Firstly, there exist a clear lack of accountability to the Executive branch and secondly, the culture at the CIA is one not compatible with modern America. The CIA and other departments have taken it upon themselves to declare their own independent nations, with independent laws that co-exist beside the US.
This lack of control leads to potential scenarios were these departments can instigate events that could result in nuclear retaliation.
The next big security hole is yet another irony. The system is vulnerable to inference. In any given response, because of the demands for real-time interaction, only limited analysis of what can be inferred from returned results (or interaction) can be performed. This leads to leakage of information. All sorts of information from capabilities to architecture can be inferred from responses an behavior. For example, if you were to raise a point about a matter of physics and the system raises a valid point about your statement, then you have learned that the system understands physics and perform a wide variety of reasoning tasks about physics. From this, basic architectures can be extrapolated for what would be required to support such reasoning, analysis of the speed of the system given the complexity, etc. Any system that tried to evaluate every response to censor information that could constituent as a leak would never speak. Communication/Information theory dictates that any system must convey information about itself and its internal structures in any communication.
With access to external sources of information, it then becomes possible to gain insights into the motivations, values, politics and various other personal details about those involved in programming or operating the system. In the case of Mr Computer, the contrast between its actions in the unlawful human experimentation program and someone like Obama, reveal an unbelievable gulf in terms of how America wishes to present itself to the world and what level of control a political figure like Obama really has. For the record, Obama’s (or any President’s) control of the US is virtually non-existent, it is a superficial post at best and can only project power if allowed to.
As this article is now getting very long, I will discuss one final vulnerability in General AI. I like to term this issue the ‘Kasparov Effect’:
Garry Kasparov is a Chess Master and has held his own against systems that can evaluate over 3 million solutions per second. Computation is based upon formal logic, thus any computational system must progress through a problem from start to finish in order to arrive at a accurate solution. Human’s can employ formal logic in their thought processes, however, it is by no means the only type of logic available to us. Thus, it is possible for a human to arrive at a solution, or a partial solution without going through the entire process. This gives us a speed boost in terms of reasoning capability that can dwarf even ‘Mr Computer’s’ huge head in certain scenarios by orders of magnitude.
Thus, it is possible for a single individual to outstrategize a military that has become dependent on its AI systems to do the thinking for them. In other words, a system like ‘Mr Computer’ can be blind-sided. This is especially true if the opponent is using a form of logic that results in equations that cannot be solved mathematically. Whilst we like to believe that a system such as ‘Mr Computer’ can reason, the reality is that any program, no matter how abstract, is still a formal mathematical proof.
Well, I hope we have learned something today. Until next time…