Thomas W. Malone: Superminds; The Surprising Power of People and Computers Thinking Together

Like individual plants and animals, superminds can be categorized into species:

• Hierarchies – where people with authority make decisions others are required to follow.
• Democracies – where decisions are made by voting.
• Markets – where decisions are made by mutual agreement among trading partner.
• Communities – where decisions are made by informal consensus or shared forms.

And the final type that encompass all others:

• Ecosystem – where decisions are made based on who has the most power and the greatest ability to survive.

IT usage is creating development. But new trends are not only important for development of AI, one of the important tasks it is doing is creating groups of people and computers that, together, are far more collectively intelligent that was ever possible before.

Superminds are a group of individuals acting together in ways that seem intelligent. Collective intelligence is the result of groups of individuals acting together in ways that seem intelligent. When we talk about intelligence, definition from a group of 52 psychologists is: “Intelligence is a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather, it reflects a broader and deeper capability for comprehending our surroundings – “catching on”, “making sense” of things, or “figuring out” what to do.”[1]

There are two kinds of intelligence: specialized and general. General collective intelligence means “group versatility” or “group adaptability”. Question that author is working on is, can we measure group intelligence. First, they defined tasks to be measured. They used Joseph McGrath framework for classifying group tasks and they choose tasks from each category in his framework: generating, choosing, negotiating and executing. They found that groups have a form of general intelligence, just as individuals do. The average and maximum intelligence of the group members was correlated with the group’s collective intelligence, but this correlation was only moderately strong. Three factors were significant: the average social perceptivenes of the group members (person social intelligence), the degree to which group members participated about equally in conversation, proportion of women in the group. Social intelligence is important for collective intelligence, but so are diverse cognitive styles. We have three types: verbalizers, object visualizers and spatial visualizers. Verbalizers are good at reasoning with words; object visualizers are good at dealing with the overall properties of images; spatial visualizers are good at analyzing images part by part. Groups that are more collective intelligent also learn faster.

One of the important question author is asking himself is how computers will work with humans. So far computers have mainly help humans. The main area was communication. But now they are evolving into:

• assistants – like WatsonPaths software that IBM is developing together with Cleveland Clinic to help doctors with diagnoses,
• peers – like Crowd-Forge system that is using online workers to write documents like encyclopedia articles,
• managers.

When we think about AI, it is important to notice, that it will be in majority invisible. But general AI is still difficult to achieve. If not for other reasons, then because coding is complicated. Google estimates that their services are based on 2 billion lines of code. Maybe one of the best way to create general AI, is to create a collective intelligence, that combines, inside a signle system, many different kinds of AI. But before achieving that, we can create more and more collectively intelligent systems by building societies of mind that include both human and machine agents. Humans can supply the general intelligence and other specialized skills that machiens don’t have. The machines can supply the knowledge and other specialized capabilities that people don’t have. And the groups of people and computers together can act more intelligently than any other person, group, or computer has ever done before.

So what is perfect intelligence. It does the best job possible given the information and other resources available. To act intelligently, there are five basic things you need to do:

• you have to decide what actions to take,
• you need to create possibilities for one or more course of action and then to identify and choose good actions, you need informaiton about the world you are acting in by:
  • sensing
  • remembering,
• at heart of intelligence is ability to learn from experience, to observe patterns in the environment and to improve your own action over time.

So how can superminds make decisions?

• Hierarchies – technology is enabling hierarchies to automate some decision making and can provide some decentralization (especially in highly automated organizations. Decentralization is possible since cost of communication is going down due to new technology capabilities.
• Democracy – the most visible examples today are governments. Elections lead to good outcomes only if the average voters have both the knowledge and motivation to vote wisely. Democracies usually can’t commit themselves to future actions in the same way as hierarchies can. Without voters who are well-enough informed to vote sensibly, democracies don’t make much sense. New technologies now make it possible to create forms of democracy that combine the best of both direct and representative democracies. These new forms have names like delegative democracy, proxy democracy and smartocracy. Author like the expression liquid democracy. When you train, select, and combine the best forecasters from a more or less random online crowd of part-time workers, you get result that are substantially better than those from multibillion-dollar apparatus of the US intelligence community. One system that is doing similar work is Watson. It includes different computational agents, each with a different kind of expertise, producing evidence for or against different possible answers. Over time, machine-learning algorithms build into the system refine the weights they give to these “opinions” from the different agents.
• Markets – unlike voters, who all have to live with the same president, and unlike members of hierarchy, who have to follow orders from their superiors, participants in a market are not bound by any decisions they didn’t agree to. A market is the first type of supermind out of all three where no individual sees the whole problem for which the group is trying to make a decision. New technologies are reducing cost of communication and provide automation tools. A lot of automation is based on predictions. Both people and computer can do them. But computers are better at doing this. But in a test, researchers find out that markets in which both people and software bots traded together worked better than those with only people or only bots trading. In financial markets, investment managers increasingly relying on quantitative, often AI-based, trading algorithms.
• Communities – they are everywhere. The information-processing capabilities of new technologies can dramatically after how communities reach consensus and establish reputations. One example is Wikipedia. Information technology can play an important role in both bringing communities together and splitting them apart. Idea is that IT makes it possible to create publiclyy visible reputations for everyone in a group, derived from detailed tracking of their actions. Imagine that very sophisticated algorithms analyze all this data to calculate a multidimensional reputation for you based on what you have contributed and used relative to your abilities and needs. Idea of cyber-socialist economy is, unlike a purely market-based economy, that it takes into account needs and abilities, not just what they consumer and produce. China is already experimenting with something it calls a social credit system.
• Ecosystems – require some overall framework for cooperation: hierarchical authority, democratic choices, market agreements or community norms. Ecosystems are a different type of supermind. They are environment in which all the other interact. But they also make decisions. New technologies will essentially increase the level of competitiveness for all the superminds in an ecosystem. Communication improvement will increase speed of everything. Superminds characteristics are based on ideas (memes) that can be transmitted by many forms of communication and then imitiated in a vast number of different ways. The goals of supermind and their members are not always the same. But when ecosystems are composed of superminds whose members are people, the ecosystems, in the long rung, generally try to provide the greatest good for the greatest number of people. In this light having smarter superminds and faster spread of ideas should increase ability of an ecosystem to achieve positive goals. And so from that point of view, IT should help ecosystems to become smarter.

When comparing superminds, one dimension is to look which one is the best for certain decisions. We should look for net benefit of their decisions, best benefits for the smallest costs. We want superminds that have strong benefits, small cost and even distribution of their decision making.

• Costs of decision making:
  • Hierarchy – medium
  • Democracy – high
  • Market – medium +/-
  • Community – medium
  • Ecosystem – low
• Benefits of decision making:
  • Hierarchy – high
  • Democracy – high
  • Market – medium –
  • Community – medium
  • Ecosystem – low
• Distribution of benefits of decision making
  • Hierarchy – low
  • Democracy – medium
  • Market – high
  • Community – medium
  • Ecosystem – low

In very large groups with lots of decisions to make, markets often have a lower cost of decision making than the other three supermind types. But sometimes it is more expensive than in hierarchies. Governments on the other hand can redistribute income, provide funding for research that benefits everyone but that it would not be fund by anybody else. Voting about everything would be too costly, so we let market decide for us, but even market needs some way to resolve disputes between buyers and sellers. So, we let hierarchical governments to supervise markets on behalf of the whole community. IT is likely to increase the size of groups and decrease costs of group decision making. And because of that superminds that suffer the most because of high decision-making costs will benefit the most from IT development. All of superminds will benefit from this development, but market and democracy will probably benefit the most.

The most important ways new technologies will help superminds get smarter are:

• Involving more individuals
• Organizing the work in new ways

Approaches like crowdsourcing for problem solving or just use of brute-force effect of bigger groups doing some tasks are representative improvement in supermind actions. Large group also have potential to use Wisdom of Crowd, plus big groups are more likely to include people with unusual knowledge. One example of potential to solve hard problems with crowd knowledge is project Foldit – online games for solving problem of protein molecules folding. But there is also some limitation of size. Optimal size of group is between 5 and 10.

New ways of organizing work include one or more of the following three elements:

• Dividing the work in new ways,
• assigning tasks in new ways and
• coordinating interdependencies among tasks in new ways.

Some of the examples are companies like Topcoder, now part of Wipro or project from Amazon Mechanical Turk. Because new information technologies make it possible to communicate across the planet instantly and nearly for free, hyperspecialized workers will be able to take advantage of global economies of scale for the specialized tasks they do.

It is important to tackle interdependencies among the activities that different group members do. Interdependencies can be classified in: flow dependencies (one individual creating something that will be used by another), sharing dependencies (multiple individuals sharing same resources), fit dependencies (different individuals making pieces that must fit together.

Another area of potential improvement in superminds working is smarter sensing. Sensemaking, ability to make sense of what is happening in ambiguous situations, is one of the core capabilities of effective leaders. Sensemaking is also a capability that groups can possess collectively. By far the most visible way technology is improving collective sensing today is by using big data and data analytics. The technologies that make it possible to analyze big data and do new kind of collective sensing also raise important issues about who owns the data and who has the right to use it. More and more things will be electronically sensed and recorded in the future. People want be able to analyze all those data, they will need computers, but computers will need people to understand subtle nuances of things. In a perfectly intelligent super-mind, all decisions would take into account everything that is known to every member of the group.

Smarter remembering is another are of improvement. A key aspect of collective memory, as opposed to individual memory, is that collective memory usually requires communication between individuals. Writing was important milestone in this development of collective memory, other improvements were made with lowering costs of storing large amounts of memories. We can divide memory in working memory and long-term memory. IT can increase the size and reliability of supermind’s long-term memory.

Smarter learning is next in line. There are two important approaches to learning – exploitation and exploration. One of the most important ways IT can help supermind learn is by helping group members to share the lesson that individual learned separately. One way is creating cyber-human learning loops. We can use elements like remembering previous cases, recognizing common patterns, automating common patterns.

When we talk about usage of superminds in order to better solve real-life problems, one area where we can check this is corporate strategic planning. Michael Porter is talking about three generic strategies: cost leadership, differentiation and focus. Technology can help introducing more ideas from wider circle of employees into strategic planning. Also, machines could help strategy teams remember good ideas that others have had in similar situation. Machines can suggest actions like integrating forward or backward, outsource more of the things you do internally or move into related market segments. Amazon used data-science to develop detailed models of many parts of its business: how customers respond to prices, ads and recommendations; how supply-chain costs vary with inventory policies, delivery methods and warehouse locations; and how load balancing and server purchases affect software and hardware costs. We are still a long way from having anything like a complete computer models of even a single company, much less the whole economy. Even though automated simulations can be incredibly helpful for making and combining predictions, they are not enough. People (who are not perfect at predicting these things either) will probably still need to use their best judgment to make final decisions after computer simulations have done what they can. Such an ideal strategy machine would use combination of people and computers. It will be kind of ecosystem supermind that combines markets (the contests), communities (proposing ideas), democracies and hierarchies (evaluating ideas).

In the world of biological organisms, the phenomenon we call intelligence and the one we call consciousness are frequently linked. One reason why consciousness is so hard to define is that it is inherently subjective. Five most common ways of defining consciousness are:

• Awareness: an entity is conscious if it reacts to stimuli in the world.
• Self-awareness: an entity is conscious if it reacts to (and can tell others about) changes in itself.
• Goal-directed behavior: an entity is conscious if it takes intentional action to achieve goals.
• Integrated information: an entity is conscious if it integrates many kinds of information.
• Experience: an entity is conscious if there is something “it is like” to be that entity.

If you take materialist perspective, the answer to the question of whether groups can be conscious is almost certainly yes. There is not yet any broad scientific consensus about how to systematically measure consciousness, but there have been some very intriguing attempts. Giulio Tononi calls its theory of consciousness integrated information theory. The core claim of integrated information theory is that consciousness is associated with something called integrated information (represented by phi, Greek letter). It is the amount of information a whole system generates that is more than just a sum of what its parts generates. Your conscious perception of anything generates integrated information. When the brains of conscious people were stimulated, they responded with complex patterns of neutral activity that were both widespread across the brain (integrated) and highly variable (information complex). Since new technologies will allow us to build more intelligent combination of people and computers, it will make sense to think about potential of consciousness of superminds.

Robert Wright’s book Nonzero is writing about tendency, not absolute law, but general tendency, for humans to form larger and larger groups that lead to a net improvement in human welfare. A good name for vast global supermind is global mind. Peter Russell and Howard Bloom, have called it global brain. The hyperconnectivity is making the global mind harder and harder to ignore.

To be intelligent you need to be good at getting whatever you want, to be wise, you also need to want the right things. People often have a kind of hunger for meaning in their life. A single neuron in a brain can’t begin to understand how the whole brain works. Single ant can’t understand how colony works, but people have enough intelligence to understand quite a bit about how superminds to which they belong works. But when superminds are getting more and more complex, this becomes harder, we will probably need to make ourself comfortable with partial understanding of all complex superminds around us.

---

[1] In the book on the page 24