Superminds

Individuals collaborating intelligently to achieve goals comprise a “supermind.”

Founding director of the MIT Center for Collective Intelligence Thomas W. Malone describes a supermind as “a group of individuals acting together in ways that seem intelligent.” To “seem intelligent,” a supermind’s activities must have identifiable goals.

“The global mind isn’t something that either exists or doesn’t. Instead, like collective intelligence and superminds, it is a perspective – a way of looking at the world.”

Two kinds of intelligence exist. Specialized intelligence achieves specific goals effectively in a given environment. General intelligence can achieve a range of goals in various environments and has the capacity to learn and adapt. Humans possess general intelligence, and computers perform only specialized tasks. Computers and humans together can achieve remarkable things.

“Long before we have general AI, we can create more and more collectively intelligent systems by building societies of mind that include both human and machine agents.”

A standardized test to evaluate collective intelligence found the groups that scored the highest possessed elevated social perception, equal participation and a larger proportion of women. Another factor was the group’s “cognitive diversity” – when it comprises “verbalizers, object visualizers, and spatial visualizers.” The more collectively intelligent the group, the faster it learns. Having a group of intelligent people isn’t sufficient to create high collective intelligence. The group must work together to achieve goals.

AI will augment human general intelligence with specialized intelligence.

Artificial intelligence won’t look like robots from science fiction. It will be a tool or an assistant. As tools, machines won’t increase individual intelligence. They will augment collective intelligence by facilitating faster and better human communication.

“Sometimes the best and most innovative ideas are so unusual that most people don’t recognize them in their early stages.”

As assistants, machines can collate large amounts of data and make suggestions. They can act as managers by directing traffic, for example, or by delegating calls in a call center. The machines won’t look like people; the tasks they perform will drive their design. The exceptions would be robots performing human-like tasks, such as serving a meal, making a bed or engaging in sex.

“How can people and computers be connected so that – collectively – they act more intelligently than any person, group or computer has ever done before?”

AI remains specialized, because general intelligence is currently beyond programmers’ reach. For general intelligence, machines require millions of pieces of data to understand even commonsense activities, such as going to the drugstore for medication. Databases such as the Cyc project compile such data, but much more needs to occur before a computer can process information that a five-year-old child could process automatically.

“In order to do effective sensing, any supermind needs to gather and interpret information about the world.”

Programmers can “teach” machines by providing them with big data, or by writing programs so the machine can “learn” from experience as humans would. For example, Google and Stanford University gave a computer 10 million images to sort into categories, asking it only to look for patterns using a “deep learning” process that simulates the human brain. It sorted the millions of photos into 20,000 discrete categories.

“Climate CoLab is not just about deciding what proposed actions are best. It’s primarily about creating good ideas for what to do in the first place.”

General AI will probably be an AI that has access to many smaller, specialized AI “agents,” forming a “society of mind” in which the sum of the agents proves more intelligent than any single agent. Computers should be brought into the human “group” to contribute to improving collective intelligence.

Technology will transform the five superminds that make up human society.

All human achievement comes from group intelligence that falls into the following types of superminds:

• Hierarchies
• Democracies
• Markets
• Communities
• Ecosystems

Almost all businesses are hierarchies. They can – by using technology – automate processes that lower-level workers perform. As technology lowers the price of communication and information retrieval, workers gain autonomy, which increases their creativity and motivation.

“We should move from thinking about putting humans in the loop to putting computers in the group.”

Democracies rely on knowledgeable voters for positive outcomes. Democracies remain the most efficient way to aggregate voter opinions, but they do not always exhibit strong collective intelligence. In markets, group decisions “emerge” from millions of small actions by each participant. Accurate predictions of these actions have great value. Algorithms prove cheap and effective at market predictions.

“In every past case where technology destroyed jobs, markets eventually created even more new jobs.”

Communities are superminds from which all other superminds evolved. Social norms dictate a member’s value, so a positive community reputation matters. Technology could enable a “cyber-socialism” which allocates resources according to algorithms that assign social value to people’s actions. This system would address individual needs, rather than reward people for consumption or production.

The most successful superminds attract people by offering more of what they want.

Ecosystems are the environments in which superminds interact. They determine their success based on the survival of the fittest. They reflect the “desires of the most powerful” and reward what works, not what is good. If technology makes superminds smarter and faster, the ecosystem itself will evolve quickly. The ecosystem doesn’t care about the welfare of its members. The most successful superminds attract people by offering more of what the people want. Often the larger the group, the more successful it will be.

“A key aspect of collective memory, as opposed to individual memory, is that collective memory usually requires communication between individuals.”

Which type of supermind would the ecosystem choose for making decisions that secure the most “net benefits” to society? Each supermind has different strengths and weaknesses.

“When machines are doing the routine work that used to be done by people, people will often do new things that were never done before.”

Markets usually need to be overseen by governmental hierarchies to deal with problems like people not fulfilling contracts. Democracies can create very large benefits from group decision-making, but making decisions in a democracy usually requires a great deal of time and effort from many people. Communities have evolved to delegate most decision making to other superminds.

How will IT affect the balance of power among superminds? It can increase the size of the groups making decisions while reducing the costs thus making markets and hierarchies more attractive.

Bigger groups are often smarter groups that can tackle the world’s most complex problems.

Since 2009, Climate CoLab has crowdsourced ideas from 100,000 experts and citizens all over the world to address one of the most important problems facing humanity: climate change. A larger group of diverse people brings more and sometimes unconventional ideas. Technology provides a cheap and efficient way to organize them. Climate CoLab funds winning ideas, which often come from people lacking a specialized background. Large groups of people with unusual approaches can innovate solutions to complex problems. Technology brings these people together on a scale never seen before.

“Superminds whose primary members are humans often exhibit a general long-run tendency to do what’s good for the people in them.”

The next level of coordination would involve “contest webs” in which smaller proposals for itemized problems – such as “crowdfunding solar energy” – integrate into subproposals for national solar projects. These, in turn, incorporate into a global proposal that awards “solar dollars” for emissions reductions. Contestants would achieve pay not only by competing, but by collaborating. These contest webs can fuel education, health and corporate planning.

A “cyber-human sensing” system could be part of a “perfectly intelligent” supermind.

Effective leaders possess sound “sensemaking” skills — a capacity to tackle ambiguous situations — and with the help of big data, this can harness and find patterns to analyze situations and predict outcomes. Connecting physical objects such as cars and houses to the internet sometimes enables gathering data to anticipate and fix a problem before it happens. One can even imagine a fully integrated cyber-human sensing system as part of a perfectly intelligent supermind that makes decisions that take into account all the relevant information anywhere in the world.

“Perhaps a wise global mind shouldn’t just serve the desires of its own members; perhaps it should serve some larger purpose.”

As a simple example, the Human Diagnostics Project augments the medical community all over the world to support diagnoses.

In “cyber-human learning loops,” humans would input data into IT that would store and identify common patterns, make predictions, and offer suggestions and solutions. Two “robot scientists,” Adam and Eve, effectively worked in laboratory settings – generating hypotheses, conducting experiments and making discoveries. Their tasks remain routine, but they free time for scientists to pursue more difficult work that requires creativity. 

Machines will not replace humans, but will allow them to pursue higher goals that often require social intelligence.

Many corporations generate, sort and prioritize ideas to remain innovative and competitive in their markets. This is strategic planning. With IT in the loop, companies could generate more ideas than ever – through “contest webs,” for example, with participation from consumers and experts alike. From this a “cyber-human strategy machine” might emerge, which would evaluate millions of strategies for one company. As they learn, computers could take over more and more of the process.

Some concern exists that machines will take away human jobs. But, as in the past, new jobs will appear in the wake of automation – ones that only humans can do. Those who succeed will often possess better social intelligence. In demand would be physicians, caregivers, social workers, athletes and entertainers. Education would be more accessible online; investors and worker associations could pay for it. Governments could give tax incentives to create jobs or redistribute income directly to those who lost jobs to automation.

Does a risk exist that a superintelligent AI (SAI) could take over the world? As AI assumes more control over large systems, such as power grids, humans could weaponize AI for nefarious purposes. But AI itself is unlikely to possess such motives. To mitigate SAIs, make their creators accountable for their actions and have human intelligence guide their goals and activities.

Most superminds demonstrate consciousness in varying degrees.

Philosophers, psychologists and scientists have sought for generations to define consciousness, and assigning consciousness to a group remains problematic. Helpful criteria exist. An entity is conscious if:

• It is aware of and reacts to external stimuli.
• It is self-aware and can tell others about its internal changes.
• It is directed towards an intentional goal.
• It can integrate information.
• It has a sense of what it is “like” from experience.

According to these criteria, most types of superminds can be said to be conscious. All five, for instance, react to changes in their respective environments, particularly markets and democracies. Businesses and governments are intentionally goal-directed; markets exhibit intent differently. Almost all superminds can integrate information. Humans have difficulty grasping what it might be “like” to be a supermind.

Assess consciousness in a system by whether the information it generates is more than the sum of its parts – like a brain made up of neurons – and whether those parts are interdependent. Measuring “integrated information” (which “phi” represents) in groups – such as people communicating over the internet – reveals a consciousness emerging over time.

The “global mind” is not an entity but a “perspective.”

People often fare better in large groups with “societal evolution” tending towards the greatest good for the most people. Evidence exists that the next supermind will be global in scope, encompassing all other superminds and comprised of machines and people coordinating together to solve the world’s problems. This will be not so much an entity as a perspective. 

To order to survive and propagate, the global mind must give its members what they want. A “wise” global mind has to want the right things. It would have to serve a greater purpose beyond ego and desire. As the global mind expands, it is difficult to predict what problems and choices it will face. It is reasonable to believe that greater comprehension is possible as machines assist humans in understanding the superminds in which they participate. This will help make a better world for everyone.