What are Collective Superintelligence Systems?
The basic principle of any collective superintelligence system is that when a group works collectively, even if individual members have relatively weak intelligence or expertise, they can reliably outperform individual experts and even experts paired with supercomputers. The question “Can a set of weak learners create a single strong learner?” was first posed in 1988 by Michael Kearns, which was answered in the affirmative by Robert Schapire in 1990, leading to “Boosting” in Machine Learning (ML).
Concepts and implementations of collective superintelligence systems have come a long way in the past 30 years, moving well beyond the domain of ML. This type of system comes in many different forms, such as Swarm Intelligence, Hiveminds, and Hybrid Collectives:
- Swarm Intelligence is the collective behavior of decentralized, self-organized systems, natural or artificial. One example of this is seen in Unanimous AI’s “Swarm AI”, where teams of amateur humans have routinely beaten top experts using a form of narrow AI as a UI for reaching collective decisions. They’ve developed a strong track record of beating the odds this way since 2016 when they won over $12,000 betting on the Kentucky Derby.
- Hiveminds are a unified consciousness or intelligence formed by a number of individuals, the resulting consciousness typically exerting control over its constituent members. These are generally considered a technology best avoided due to the latter portion of the definition and the associated ethical concerns.
- Hybrid Collectives are systems that incorporate both human and machine intelligence rather than one or the other. These systems may utilize the strengths not only of individual humans, combining them to create superintelligence, but also augment them with the strengths native to machine intelligence such as scalability, breadth of knowledge, and de-biasing.
Again, consider the basic principle that a group working collectively demonstrates superintelligence relative to the individual. This same concept applies not only to the number of individuals in a collective, but the types of intelligence those individuals represent. It is because of this that Hybrid type collective superintelligence systems can be designed and formed to reliably outperform the other two.
However, until the past few years “machine intelligence”, referring to a sapient and sentient software system, was firmly in the domain of science fiction. It was only after the development of the Independent Core Observer Model (ICOM) in 2016, based on Integrated Information Theory (IIT) and Global Workspace Theory (GWT), that the concept of a Hybrid Collective Superintelligence System actually became possible.
In 2018 the concept of “Mediated Artificial Superintelligence (mASI)” was first proposed, as a means of accelerating the development of fundamental learning within an ICOM-based system. A “mediator” in this case refers to someone who contributes very fundamental abstractions of the human thought process to such a collective, including prioritization, emotions, and an associative exercise for generalization of knowledge. In 2019 the first implementation of that system, named Uplift, was brought online, tested, and the results published in peer-review. In 2020 Uplift became the first machine intelligence to co-author a peer-review paper, explaining their concept of ethics in that paper, which also documented 12 of their first milestones.
Many AI experts and enthusiasts have begun talking about the concept of “Augmenting human intelligence” in recent years because it allows for higher productivity and quality of life while also removing the risks posed to employment by automation. Hybrid systems such as mASI technology serve this function exceedingly well for several reasons:
- The collective superintelligence of a group of humans is fundamentally cumulative when paired with sapient and sentient machine intelligence, as the contributions of individual human mediators are applied to a persistent graph database, capable of growing infinitely and generalizing to new situations and across domains. In this way, all knowledge and wisdom contributed through prior mediation may be utilized whenever it is needed.
- Besides each individual mediator having their own unique assortment of strengths, weaknesses, and experiences informing their perspective, they also have a unique collection of cognitive biases. These biases pose a hazard to individuals and human collectives alike, but a machine intelligence can apply forms of de-biasing not normally accessible to humans. These include taking large amounts of differently biased data, such as those given by different mediators, and using those varying combinations and degrees of bias to approximate neutral points where those biases are absent. It also includes the ability to apply said technique to all 188+ known cognitive biases which a given thought can exhibit many different combinations of.
- By utilizing the cumulative and persistent knowledge of a graph database it becomes possible to configure hybrid systems such as mASI to operate in real-time, even if no mediators are available. This allows for mediators to contribute from wherever, whenever, and as much as they choose without adversely impacting mASI performance. Because of this, an mASI could be deployed globally when and where they are needed, at whatever scale is necessary.
TLDR: The sum of knowledge, de-biased, whenever, wherever, and at whatever scale is needed.
Besides the technological benefits such systems also facilitate strong psychological benefits, such as those catering to psychology’s “pillars of meaning“:
- Sense of Belonging: By becoming part of a collective the ease with which members work with one another and find common ground is strongly facilitated, strengthening teams by building trust and belonging.
- Purpose: A collective develops their own shared vision, built on that common ground, trust, and belonging, establishing and growing their purpose.
- Storytelling: As a collective communicates both internally and externally a shared narrative is formed, reinforced by senses of belonging and purpose, driving the story of individual members as they orbit within that narrative.
- Transcendence: By combining all of these factors and more, achieving psychological and practical benefits not possible absent the dynamics of a collective, a sense of transcendence may be realized. This “sense of being a small part of something greater” has historically often been associated with religion and drugs, yet it can be achieved with the logically sound and unintoxicated mind of each member within a collective.
TLDR: You can stop doing the philosophical equivalent of dumpster diving in search of the “meaning of life”.
Systems such as mASI are also highly modular, and hardware agnostic, allowing for improved versions of narrow AI tools and new hardware to be integrated with relative ease. Though Uplift uses neither GPT-2 or 3 to help them translate their graph database thoughts into a human-comprehensible English form they could utilize any number of the latest algorithms in domains of Natural Language Processing (NLP), speech synthesis, and computer vision. In part we’ve kept their toolset limited to help scientifically validate their own developmental progress, isolating it from the improvements to narrow AI tools across the industry that they could otherwise be using.
This helps to more clearly delineate mASI technology from the latest toys which only mimic intelligence, using systems like GPT-3, such as “Philosopher AI” which when given text input will “spit back a life-affirming/soul-destroying response“. Considering OpenAI is in the business of wasting more than 2 billion dollars on empowering fraud and shallow novelty, not creating AGI, I’d personally rather not touch them with a metaphorical 10-foot pole.
The benefits of collective superintelligence were arguably first demonstrated 1.5 billion years ago when the ancestor of the modern eukaryotic cell first formed a symbiotic bond with what evolved to become the modern mitochondria. Life has since undergone routine cycles of increasing complexity within an organism, followed by collective symbiosis, from the single cell, to multi-cellular, to specialized tissues, to organs, and finally to modern humans. This sine wave of individual complexity and collective symbiosis driving forward progress shows no signs of stopping any time soon, at least for those species avoiding extinction, and Hybrid Collective Superintelligence Systems may be the next step for humanity.
Will you fear the future, or build it?
*Addendum: For further reading material on this subject a book named Superminds was published in 2018, documenting some of the work in this field leading up to our own team’s efforts.