It is virtually impossible to have a complete control over a maturing AI. If it becomes intelligent enough, it may always find ways to outsmart its controllers many months before the planned escape from its confined, soft, and hard environment. Therefore, we need a variety of different approaches, which together may be able to control the ‘mind’ of a maturing Superintelligence much better.
If the current exponential progress in the most advanced discoveries in neuroscience and AI continues, it is quite likely that by 2025 we shall have the first Transhumans with increased cognitive capabilities thanks to brain implants. As mentioned earlier, towards the end of this decade, the brain implants will have even more advanced mental capabilities, allowing part of the human’s brain ‘fuse’ wirelessly and continuously with various forms of AI, including the maturing Superintelligence. This will make such people many times more intelligent than an average human, capable of making fast decisions even in a complex situation. This will also have paramount repercussions far beyond the AI field. However, we should look at such an option in the most positive way, specifically in the context of a maturing Superintelligence.
In his interview in May 2020, Elon Musk confirmed indirectly his long-term aim: “Even in a benign [AI] scenario we are being left behind. So how do you go along for the ride? If you can’t beat them, join them.” This can be translated into something like, we shall have ‘proper’ Transhumans within a few years’ time and we desperately need them because they may be our best safeguard against a malicious Superintelligence.
It is quite likely that the first of these Transhumans will indeed be those at the forefront of the most advanced AI developments, such as Elon Musk, Bill Gates or Jeff Bezos and their top AI developers like Demis Hasabi from Deep Mind. Their role will be similar to what happens today, when the top OpenAI or Google’s Deep Mind developers and managers decide, what functionality their applications will have, how those functions will be executed, and how individual people will be able to use them, which may depend on the users’ access rights.
If we consider immense intellectual power and ultra-fast decision-making of such Transhumans, like Elon Musk is planning to become himself, and for whom saving Humanity is the absolute priority, then we may have no better option than to entrust the early Transhumans very soon to act on behalf of the whole Humanity to deliver a friendly Superintelligence. They will have to control it from ‘inside’ by part of their brain being continuously fused to the maturing Superintelligence via a Brain-Computer-Interface (BCI), such as the emerging Neuralink chip, currently embedded in a human skull. Let me then outline one of the possible ways in which we could control the development of Superintelligence.
As progressively larger part of Transhuman’s cognitive functions are replicated in a digital form, it will start an evolutionary process, similar to the evolution of homo sapiens. The most critical difference between humans and almost all animals is the presence of an extra layer in the brain – called Neocortex. By digitizing brain functions and adding a necessary Brain-Computer Interface (BCI), an additional, artificial digital layer of the brain, which I call Transcortex, will be created. In hardware terms, it could be an electronic device semi-permanently attached to the scull and detached only when a hardware upgrade is needed. In the next few years, its software will include wireless communications and other applications necessary for communicating with a particular brain region, or a brain function, and externally with an advanced AI system, a prototype of the future Superintelligence. Such Transhumans will be able to browse the Internet wirelessly by thought alone and perhaps store some of this information in a small memory store in the Transcortex. But by about 2025 we should have the first digitized copy of a particular brain function embedded in a digital circuit and hardwired to an advanced AI system. So, from that point onward, the communication with a maturing Superintelligence will be via an individualized BCI channel, supported by digital circuits at both ends. It is like we would have a physical Poste Restante box at the Post Office but in this case it will be our digital chip fused with a particular digital infrastructure of the prototype Superintelligence. In this way, Transhumans could control Superintelligence from ‘inside’. Here is a potential implementation path of such an approach:
Once we have achieved that, it will be a major breakthrough in controlling the maturing Superintelligence. A biological neural network of millions of neurons supporting a particular brain function, such as speech, decision making, memory etc.) will be replicated by a network of artificial neurons, which we can already create. This will be an exact digitized copy of a fragment of the brain. So, each Transhuman would initially have at least two devices, the Transcortex, which is a Brain Computer Interface attached to his brain/skull, and a digital copy of a brain function – a neural network, which could be rewired wirelessly by an individual Transhuman, with the required access rights.
This is the way selected Transhumans could control 24/7 the development of Superintelligence, only being directly engaged as needed – by an alert. Here is my visualisation of such a process for Transhuman Supervisors or Transhuman Governors (read the diagram from bottom up and from left to right).
However, giving Transhuman Supervisors or Governors such an exceptional power to control the development of Superintelligence is not free of risks. There is no guarantee, that some of those Transhumans will have no urge to dominate us all, using still Immature Superintelligence. That is why I cannot emphasize it enough how potentially dangerous some Transhumans might become even within this decade. Should the wearer of such an advanced implant be, for example, a rich and influential psychopath, or a leader of one of the ‘expansionist’ Superpowers, then this may embolden him to take risks in his attempt to rule the world, which he may have not been able to contemplate as a purely biological human. Although the problem is very difficult to solve, it should not be ignored by those that might become the de facto leaders of the future World Government.
Therefore, as soon as such implants become available (perhaps as soon as in 2023-24), they should only be dispensed by an international licencing authority, such as the Global AI Governance Agency, which I proposed earlier. Unfortunately, the European Artificial Intelligence Board (EAIB), proposed by the EU Commission, will start operation about 2025 (if nothing changes in the scope of the current proposal). Therefore, in the meantime there must be another, temporary option where such a licence will be given by a national agency, such as FDA in the USA (which actually is already evaluating such a proposal from Neuralink, as of 3/2/2021). It would monitor the production and distribution of such implants. It may allow the use of most of them, after registration, to anyone with perhaps some minor restrictions (e.g., that it is unable to read other people’s thoughts). However, the implants which may significantly enhance cognitive abilities of their owners, so that they may become a potential threat to a society or even the world as a whole, would be only distributed by a special licence to a strictly selected and vetted individuals.
In the future, such an Agency as EAIB might licence such brain implants for most countries, including for the leaders of international organisations, such as the EU, UN, International Court of Justice and of course some top scientists. They might be licenced for a specific duration, e.g., for the time of being in office, and digitally disabled once they leave the office. Unfortunately, even if we have an international law banning such unlicensed brain enhancements for achieving superhuman intelligence in this way, it will happen nevertheless because some people, like top AI scientists developing this technology, and those with money, power, and influence, may get such implants via informal routes. And how about the autocratic state leaders? Therefore, the risk of developing a rouge advanced AI, which may threaten our civilisation will still be there and has to be continuously monitored.
But let us focus on the licensed, mainstream use of such BCI implants for the selected Transhumans, who would then have some ability to monitor the behaviour and operation of an advanced AI from ‘inside’. To increase the effectiveness of such control of the AI development, I propose to use a multi-layered system. It would be a similar process to a three-step verification of a banking transaction. In such a process, a decision to proceed must be confirmed by different media and via different communications channels, like for banking transactions it may be a computer, a mobile phone (you need to know the number) and an application on the mobile phone (you must have such an application and know the password for access). To control a maturing Superintelligence in this way, any time it wants to execute a sensitive decision, it will have to be confirmed by an authorized human, who will be hard-wired to it (e.g., a Senior development Vice-President, who will not be linked by a BCI implant). Additionally, it would have to be authorized by a Transhuman, whose some part of the brain will be wirelessly communicating with a maturing Superintelligence all the time.
What I propose here should be in place latest by about 2023-24, at least in some most advanced AI development centres. Otherwise, a maturing Superintelligence may already outsmart its developers in some way and create serious but still localized problems (I do not expect there would be a global Superintelligence development centre by then). To avoid even a local loss of control we may need to rely solely on AI scientists and developers to create a safe development environment by being in charge of the whole process. This interim period of local control may last until a de facto global Superintelligence development centre is created, perhaps by EAIB. In that interim period, probably till about 2030, such control might be executed as in the picture below:
We already have everything you see in this picture, at least in a prototype form, that is needed to create the first Transhuman, apart from the implant for a Transhuman. So, we already have several thousand Starlink’s satellites, with about 20,000 more launched in the next few years into the Lower Earth Orbit (LEO) by that company, also set up by Elon Musk, the British OneWeb and Project Kuiper owned by Amazon’s boss – Jeff Bezos. There are already artificial neurons connecting with biological neurons without causing any immunity problems as well as wireless communications with the brain – Brain Computer Interface (BCI) either via implants in the skull or using special head gear. Finally, the total number of computers connected in a Google network alone is now probably reaching 10 million processors (2.6 million in 2016), plus petabyte size memories.
Finally, we have about 10 top AI companies engaged directly into developing Artificial General Intelligence (AGI), which in this book we call Superintelligence. They are all American, apart from DeepMind, which is British (part of Google’s Alphabet), and which really pushed AI on an exponential growth path. Another important company is Nuralink, also founded by Elon Musk, which has produced the first wireless brain implants.
But perhaps the most important organization is OpenAI, founded in 2015 by Elon Musk and Sam Altman. It is the only organization which, as the name suggests, discloses its code to other developers, and is set up as a non-profit, transparent organization (in 2019 it created a for profit-subsidiary, which however must follow the mission and the ethics of the not-for-profit parent). But perhaps the most significant for the purpose of controlling the development of Superintelligence is its mission: “OpenAI’s mission is to ensure that artificial general intelligence (AGI)—by which we mean highly autonomous systems that outperform humans at most economically valuable work—benefits all of humanity. We will attempt to directly build safe and beneficial AGI, but will also consider our mission fulfilled if our work aids others to achieve this outcome” . In 2016 its CEO, Sam Altman said this about the OpenAI’s intended role “allow wide swaths of the world to elect representatives to a new governance board”. OpenAI’s biggest achievements to date is Natural Language processor called Generative Pre-trained Transformer 3, known as GPT-3, API – a multipurpose application for accessing other products, DALL-E – a Transformer model that creates images from textual descriptions, and CLIP does the opposite: it creates a description for a given image.
We can visualize the monitoring of the whole process of maturing Superintelligence by substituting todays’ Internet with all its resources and browsers, such as Google or Edge as the future Superintelligence with its vastly superior resources and capabilities, which ultimately may lead to the creation of a superintelligent and conscious being. The Internet is controlled by the World Wide Web Consortium (W3C), fuonded in 1994 by Tim Berners-Lee. In 2019 it had over 400 members, which included businesses, non-profit organizations, universities, governmental entities, and individuals. One of the cornerstones of W3C is its independence from government authorities, its self-regulation, and self-control, including the safety of the data traffic. Superintelligence is now at the level of development of the Internet in the early 90’. Although the impact of Superintelligence on our civilization is far greater, we can still apply a similar mechanism for its development and regulation as in the early days of the Internet.
If OpenAI really follows its mission then it might be an ideal candidate initiating the creation of a Superintelligence Consortium, modelled on the Internet monitoring W3C Consortium, as a means of controlling from the ‘inside’ the maturing Superintelligence.
If such a Superintelligence Consortium is set up, then their members may not only produce various software and hardware directly linked to AI but also the brain implants, such as those currently produced by Neuralink or similar devices produced by other companies. Such devices will form the basis for wireless communications between a human and a maturing Superintelligence.
The procedures for implanting digital chips into the brain will be regulated on a similar basis as the development of medical drugs, including the phased trials. This of course could only be done on a national basis. In practice it means such approval will be granted by the USA agency, such as the Federal Drugs Administration (FDA), since almost all most advanced AI companies and organisations engaged into the development of the Artificial General Intelligence (AGI), called in this book – Superintelligence, are there. Neuralink has already applied for the authorization of its brain implants into humans to the FDA. There can of course be other solutions provided by other companies, not necessarily members of the Superintelligence Consortium.
Once the brain implants have been tested for a reliable, harmless wireless communication with an advanced AI, such as GPT-3 successor, and authorized for implementation into humans by FDA in the USA, the Superintelligence Consortium will select the candidates who will become the first Transhumans. The criteria for selection will probably be established by the Superintelligence Consortium but the selection process itself should be done by an independent body, which may include specialists from several domains, e.g., AI scientists, neuropsychologists, biologists, chemists, physicists, etc.
A candidate will have to agree to observe certain legal and ethical rules and agree for some procedures, such as:
- Have an implant inserted and removed, or a digitally disabled when requested by the Superintelligence Consortium
- Undertake psychological/ psychometric tests
- Maintain confidentiality of the innovations discoveries and test results (to prevent unauthorized copying of the whole process by potentially malicious users)
- Be open and trustful and ready to disclose all information requested by the Superintelligence Consortium or any other authority (if requested by a judicial court)
- Agree to share part of his memory related to communications with Superintelligence, and later on, if such need arises, his thoughts with other Transhumans in the team.
Once the first Transhumans have been selected, they will become Transhuman Supervisors, members of the Transhuman Supervisors Board, who may work in the following way:
- The selected candidates will have their implants inserted and start communicating wirelessly with the developed prototype(s) of Superintelligence. Every year the scope of their interaction and the depth of control and monitoring of the maturing Superintelligence will widen, which may be necessary as its capabilities will increase exponentially. The wireless continuous communications with Superintelligence should enable the monitoring in real time.
- The role of the Transhuman Supervisors, fully subordinated to the Superintelligence Consortium, will be to minimize the risk of developing by accident or intent a malicious Superintelligence. Since part of their brain will be wirelessly and continuously connected to Superintelligence via the brain implant, they will be able to control it from the ‘inside’.
- It may be necessary at some stage for Transhuman Supervisors to communicate continuously and wirelessly between themselves (i.e., reading each other’s thoughts via something like an external digital memory area, probably at a subconsciousness level, so it should not disturb their lives too much. Only the Transhumans who agree would be connected.
- Only authorized personnel of the Superintelligence Consortium would have access to that area. This may be used for monitoring the working of the Superintelligence prototype. As it matures, the Transhuman Supervisors will gradually be communicating more and more often with it directly by thought alone.
Transhuman Supervisors will discuss any potential problems (and opportunities) with the Superintelligence Consortium to modify the Superintelligence development process as needed. But they also may increase their own interdisciplinary knowledge exponentially by having access to their own large wirelessly integrated digital memory and processing capabilities. In just 3-5 years from becoming Transhumans, they may be far more intelligent than any biological human in any aspect of human knowledge.
Therefore, it is quite likely that in the second half of this decade some views and suggested decisions of the Transhuman Supervisors Board may not be understood even by the members of the Superintelligence Consortium. And yet, they may become indispensable in solving global problems fast and in the most effective way, such as climate change. However, the danger is that they may be perceived as non-elected dictators, whom people should not trust. This dilemma can only be solved, by extending the scope and the role of the Transhuman Supervisors Board, which is discussed in teh next sub-tab.