Thomas W. Malone

“We'll see that the history of humanity is largely the history of human superminds, of how humans in groups, like hierarchies, communities, markets, and democracies, accomplish things that individual humans could never have done alone.”

“For a long time, the most important contribution of computers won't be artificial intelligence, it will be hyper-connectivity, connecting human minds to each other in new ways and at unprecedented scales.”

“It's about how human computer superminds will be smarter than anything our world has ever seen.”

“In fact, it's not too much of an exaggeration to say that almost all our important problems are solved by groups of people rather than by individuals alone.”

“We'll see that it was the collective intelligence of human groups, not the intelligence of individual humans, that first differentiated our human ancestors from all their animal relatives.”

“And we'll see that over time, information technologies, like writing and the printing press, allowed groups to become dramatically larger and more intelligent.”

“Even though the individual's actions need to be connected, the individuals in a supermind do not need to cooperate with each other or have the same goals. A company called Innocentive, for instance, has online contests where scientists and technologists compete to solve difficult problems, like how to synthesize a particular chemical compound. But even though the problem solvers are competing with each other, not cooperating, their actions are connected because they are all working on the same problem.”

“And when evaluating a group's intelligence, it is often useful to consider overall goals for the group that are important to the observer, even if none of the individuals in the group has those goals.”

“In other words, we found that groups have a form of general intelligence just as individuals do.”

“In fact, a group's collective intelligence score was a much better predictor of how well the group did on these more complex tasks than either the average or the maximum individual intelligence of the group members.”

“we found that the most collectively intelligent groups were those with an intermediate level of cognitive diversity. In other words, groups where the members had very different cognitive styles weren't as smart, perhaps because they couldn't communicate effectively with one another.”

“It would be ideal if instead we could just observe a group doing what it ordinarily does and use those observations to accurately estimate the group's intelligence.”

“People have the most control when machines act only as tools, and machines have successfully more control as their roles expand to assistants, peers, and finally, managers.”

“In other words, general AI has seemed about 20 years away for the last 60 years.”

“My own view is that, barring some major societal disasters, it is very likely that general AI will appear someday but likely not until quite a few decades in the future.”

“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.”

“In other words, we should move from thinking about putting humans in the loop to putting computers in the group.”

“Of course, no real group will ever achieve perfect intelligence all the time.”

“It never forgets anything and it uses everything it knows to make every new decision.”

“It also figures out lots of other things that are knowable from the knowledge it already has.”

“Together, these five cognitive processes are the building blocks of intelligent behavior, whether it occurs in individuals or in groups.”

“Some systems with only specialized intelligence can get by with doing only one or two of these processes, but any system with general intelligence must do them all reasonably well.”

“In fact, you can view a hierarchy as a supermind that mobilizes large numbers of individuals to achieve the goals of whoever controls the hierarchy.”

“But sometimes the politicians in a government, the managers in a company, or the workers in either kind of bureaucracy are able to get a lot of what they want even when it's not what the voters or stockholders really want.”

“For instance, there will usually be at least a few humans at the top of these hierarchies using their general intelligence to direct the organization's overall strategy and manage special case exceptions.”

“On the other hand, as I argued in my 2004 book, The Future of Work, it will probably be even more common for cheap communication to help decentralize decision-making in many parts of our economy.”

“Cheaper communication means it's economically feasible for many more people to have much more information than in the past.”

“They can also receive bonuses of between five and ten times their salaries, determined by the opinions of their peers.”

“One former employee, for instance, says it is a lot like high school, where the popular kids have a lot of informal power.”

“But the key aspect of a democracy is that the ultimate decision-making power comes from all the members of the group, not just its leaders.”

“But if the average voter is more likely than not to vote for a bad outcome, then the more voters there are, the more likely it is that the outcome of the election will be bad.”

“In other words, elections lead to good outcomes only if the average voters have both the knowledge and the motivation to vote wisely.”

“In this way, you can have your interests represented by people with very specialized knowledge and enough time to consider decisions thoughtfully.”

“But it seems unlikely that a system like this would be a lot worse than the representative democracies we have today, and it's quite easy to imagine that it could be a lot better.”

“In addition to aggregating people's values and preferences, democracies can also be very useful as a way of aggregating people's opinions about what is true.”

“And after there was enough data on how good the volunteer's predictions actually were, the best predictor of any one volunteer's future accuracy turned out to be that volunteer's accuracy in the past.”

“For instance, they found that group predictions were more accurate when they put higher weight on more recent predictions, on predictions from people who had been more accurate in the past, and on people who updated their predictions more frequently.”

“told Ignatius that the superforecasters from the Good Judgment Project performed about 30% better than the average for intelligence community analysts who could read intercepts and other secret data.”

“The first process would be designed to predict the outcomes of actions we might take, such as increasing taxes or changing labor laws.”

“By dividing the work of democracy in this way, we let the people who are good at making accurate predictions vote on the results various actions would likely produce, and then we let those who know what we want vote for the results we want most.”

“While they are remarkably good at efficiently allocating resources to fulfill human desires, they are completely blind to things that are not included in the prices that buyers have to pay.”

“A government might impose a tax on pollution, for instance, thus harnessing the vast power of the markets to allocate resources in ways that minimize pollution while still efficiently providing the other goods and services that consumers want.”

“the researchers found that the algorithm made more accurate predictions than 11 of the 14 counselors.”

“About 60% of the studies found that the algorithms were significantly better than the people at making these predictions, and the rest found no significant difference between the two.”

“This combination of diverse approaches resulted in overall predictions that were better than either the people or computers made alone.”

“One of the intriguing possibilities about using a market for prediction is that it provides incentives for people to get involved if and only if they can be useful.”

“For instance, in routine situations, like predicting airline travel on a normal day, computers can often do a good job, and humans have little incentive to participate.”

“Or if they enjoy participating as a kind of entertainment, like gambling, then the money they lose will provide even more incentive for participation by the people and bots who are actually good at predicting.”

“If your bots are better than mine at making accurate predictions, then you will make more money than I will.”

“But we also want our bots to know their own limits and participate only when they have a reasonable chance of being successful.”

“In fact, it can even work with simple prediction algorithms, like statistical regression, that have existed for decades.”

“Wikipedia software also maintains records of what each contributor has done, an important part of building and recognizing reputations in this community”

“In a process similar to scientific peer review, the technicians themselves, not higher-level managers, decided which tips were useful and which weren't.”

“The original online discussion included 13 pages of text, and was filled with digressions and repetitions.”

“But when Mark translated the discussion into an argument map in his system, he found that all the discussion content could be summarized in the following eight lines.”

“I'm surprised that the approach hasn't already been more widely used in business, politics, and elsewhere.”

“Instead of mutual agreements between buyers and sellers in the market, decisions about resources would be based on societal norms that are embodied in the way you're rated by other people and in the algorithms for computing reputations.”

“Why wouldn't a talented person, for example, try to hide his or her abilities and contribute just enough to satisfy society's minimal expectations?”

“Second, simply by making communication faster and easier, new technologies will often increase the speed of evolution in ecosystems composed of superminds.”

“When organizations committed to the ideals of democratic decision-making grow, he argued, at some point it becomes impossible for everyone to participate in decision-making.”

“These elite members then acquire a vested interest in maintaining their positions and in preserving the organization itself, even when that requires doing things that contradict the organization's original goals.”

“Quite often, it is being able to attract more human members.”

“having significantly more soldiers on one side than the other is often a decisive advantage”

“So, in summary, if people choose to participate in the superminds that give them more of what they want, and if the superminds that attract more people survive and reproduce more effectively, then, in general, the ecosystem will choose the superminds that give more people more of what they want.”

“So even though it doesn't always happen, ecosystems often favor the superminds that give more people more of what they think they want.”

“All these smaller decisions are coordinated by a relatively small number of ever-changing prices for the items about which the decisions are being made.”

“A manager in a hierarchy can deal with problems like this by quickly punishing Ron, but markets require some mechanism outside the market itself.”

“This is the best explanation we have for why hierarchically managed companies exist in our economy instead of every worker being an independent contractor who negotiates with others every day to do the work that needs to be done.”

“That means they can make decisions that are good for the group as a whole, even though some of the individuals in the group would never have agreed to them.”

“It also means that the benefits markets do create always get distributed in ways that everyone, at least in some sense, is happy with.”

“Instead, we'll let hierarchical governments supervise markets on behalf of the whole community, but we'll supervise the governments in two ways.”

“Therefore, as information technologies continue to increase the size of groups that can make decisions together, it's very likely that markets will take over even more of these decisions.”

“For instance, we should expect that IT-enabled innovations like liquid democracy will enable more democratic voting in governments and in businesses, and that markets will continue to take over functions in our society that were previously done by other types of superminds.”

“It found all 10 balloons in only 8 hours and 53 minutes.”

“But if that money were the only incentive, anyone who wanted to look for balloons would be motivated not to tell anyone else about the challenge, because each new person recruited would have been more competition for the reward money.”

“So the MIT team also gave a share of the reward money to the people who helped find the people who found balloons.”

“It turned out that the average of all the guesses, 1,197 pounds, was a remarkably accurate estimate of the actual weight, 1,198 pounds.”

“When you average their answers, the errors should cancel each other out, leaving an accurate estimate.”

“20 to 40 percent of the class members often make guesses that are more accurate than the average.”

“a number of other researchers have similarly found that the median is a better way of combining a group's guesses than the average.”

“For instance, in one compelling experiment, when group members saw information about one another's guesses, their guesses moved closer together, and they became more confident that their guesses were right. But their guesses actually became less accurate.”

“So even though cheap communication enables larger groups to exist, we still have to be careful about decision-making methods, since some ways of using that communication capability may actually make the group less smart.”

“found that the further away the problem solver's technical field was from the field in which the problem originated, the more likely he or she was to solve it.”

“Only the problems that stump the experts in the field get posted in the first place.”

“Of the people who submitted proposals, people who had no graduate school training and no previous climate change experience were just as likely to become finalists as those with the opposite characteristics.”

“by reducing the costs of communication and coordination, new technologies are making it possible for groups of people with very diverse kinds of knowledge and problem-solving approaches to work together at a scale and with a degree of collaboration that has never before been possible in all of human history.”

“And hyper-specialized workers can sometimes contribute unknowingly to overall goals they would never support.”

“the system was automatically tracking and managing interdependencies that might otherwise have required time and effort from people to manage.”

“Then the points each global proposal receives are distributed automatically among the authors of the global proposal itself and the authors of all the national and other lower-level proposals it incorporates.”

“Another limitation of the neural net approach is that even if machines could learn some of these patterns, people wouldn't usually be able to understand how the machines reached their conclusions.”

“Part of the reason these two pieces of information were not connected has to do with restrictions on sharing information between agencies like the CIA and FBI.”

“One of the advantages of using a technology like Bayesian networks is that both humans and computers can understand and estimate the same probabilities.”

“In Chapter 5, we talked about how, in a perfectly intelligent supermind, all decisions would take into account everything that is known to every member of the group.”

“Instead, it would be because each person added his or her own information to a system that automatically calculated the combined implications of everyone's information, and because the system then let each person see the implications that were most relevant for his or her own decisions.”

“They increase the size of a supermind's working memory for complex arguments.”

“Perhaps the same is true of groups, and by increasing the size of a group's working memory, we may be able to increase the group's intelligence.”

“The key point here is that Harry's colleagues shouldn't try to computerize the whole crime-solving process once and for all.”

“Eventually, the machines can just automatically take the actions that humans always approve.”

“It usually requires picking a combination of superminds that can work together as a larger supermind.”

“To evaluate these options, Lafley and his colleagues specified conditions for success that would have to be true for each choice to be successful.”

“For instance, when P&G evaluates a strategic option, it first evaluates the conditions for success that it thinks are least likely to be true.”

“However, many options are eliminated after evaluating only one or two questions, thus saving large amounts of unnecessary evaluation effort.”

“A more interesting possibility would be to create a conditional prediction market, in which many people make predictions about what the eventual costs would be if the product is produced.”

“If the product is not produced, everyone just gets his or her money or points back.”

“Either way, it might be useful to let people enter detailed arguments online for and against the different points of view, which could inform the people who are participating in the prediction market.”

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

“By systematically tracking over time how accurately and how early people predict technological advances and other kinds of breakthroughs, we could identify the people who can do this better than others. Then we could ask these people to take a second look at some of the crazy ideas that other people reject.”

“But if purchasing specialists separately estimate whether materials costs would be acceptable with a recession and without one, and if marketers do the same for sales volume, then a Bayesian network could automatically combine all these estimates with separate projections by economists about the probability of a recession.”

“The results would thus be an integrated prediction that draws upon expertise from economists, purchasing specialists, and marketers, all automatically combined by computers.”

“The end result would be a handful of the most promising options from which the human managers of the company would make their final choices.”

“People who cause the emission of greenhouse gases create costs for everyone on the planet, but they have to pay only a tiny fraction of those costs themselves. And since these costs are not included in today's markets, markets essentially ignore them when they make their decisions.”

“We could use contest webs, augmented by a network of people and computers with many different kinds of expertise, to evaluate the overall results of the various combinations of proposals. That way, the social pressure and consensus decision-making of the world community could be much more precisely targeted towards specific combinations of actions that would be most useful.”

“But I think the burden of proof is extremely high for anyone who argues that the outcome this time will be different from all the other times technologies have eliminated jobs in the past 200 years.”

“The closer I get to creating an SAI, the more you would be willing to pay me to stop, so I would have incentives to get as close as possible to creating a dangerous SAI.”

“An obvious solution to that problem is to do what most modern societies do, let a democracy oversee the hierarchy that monitors and regulates AI development.”

“If we assume, as I do, that we are probably at least decades away from anything approaching human-level AI, it's quite possible that our methods for global governance may have developed sufficiently before we actually need them.”

“But if you own a dangerous wild animal like a tiger, you are liable for whatever the tiger does even if you haven't been negligent at all.”

“When we are tempted to develop automated systems that can override human judgment, like safety measures and semi-autonomous vehicles, we should carefully consider whether and how humans can override the overrides.”

“But we'll see in this chapter that we can build a definition of consciousness by drawing on studies of consciousness in the fields of philosophy, psychology, and neuroscience. and we'll see that there are many important senses in which groups can, indeed, be conscious.”

“The only place all the different functional points of view come together is at the very top, which leads to an unusually centralized form of information integration.”

“both your brain and Apple depend on a kind of centralized information integration to function, and thus in this sense, both qualify as conscious.”

“So to analyze our groups, we simply measured, at each point in time, whether or not each individual in the group was active, sending a message or talking. Then we used this information to calculate a value of phi for each group.”

“The measured value of phi for the internet grew significantly over the six-year period we studied from 2008 to 2014.”

“Each time these transitions occurred, new technologies enabled larger groups of people to communicate with each other more easily.”

“Democratic governments are sometimes overthrown by dictators, and in some cases, such as with Hitler in Germany, democracies elect leaders who later become dictators. but in the long run as robert wright says there is a direction an arrow in human history in which technologies arise that enable new richer forms of non-zero-sum interaction that is interactions in which all participants are better off from having interacted i am persuaded by wright's argument, and I think it is very likely that we will continue to see movement toward what you might think of as the end point of this process, global scale integration of all the people, computers, and other forms of intelligence on our planet.”

“The global mind is an ecosystem, so it wants whatever the most powerful actors in the ecosystem want.”

“As we've seen, the most powerful actors on our planet today are mostly groups of humans organized in superminds such as governments, businesses, markets, democracies, and communities.”

“To do that, as we saw in chapter 10, they usually need to provide their members with whatever they want.”

“So perhaps the most important way we as individuals can help the global mind make wise choices is to make wise choices ourselves.”

“And in large markets, democracies, or ecosystems where the key decisions result from interactions among vast numbers of individuals it can be harder still to understand why these choices are made.”

“Is developing such a global mind our destiny? I think it is.”

“Next, we use this framework to classify descriptions of 13,275 AI software applications and a worldwide tally of 20.8 million robotic systems.”

“Most of the market value is used in information-based activities (72%), especially creating information (36%), and only 12% is used in physical activities.”

“Using the methods for automated classification and human editing described in Sections 6.1 to 6.3, our first result was a comprehensive ontology of work activities including approximately 40,000 total activities.”

“This means that if we discover that AI is applicable to a parent activity (e.g., “Write”), we have principled grounds for hypothesizing that it may be applicable to the child activities (e.g., “Write report,” “Write email”) even before verifying this empirically.”

“These activities may be less amenable to automation or may offer comparatively limited value when delegated to AI systems, given prevailing human preferences, accountability concerns, and organizational norms.”

“Over this two-year period, the number of AI applications increased by a factor of 6.0 times, while coverage expanded by only a factor of 1.2 times.”

“Development remains concentrated within a relatively narrow subset of task categories, indicating that advances in AI capabilities have thus far translated into depth in a few activities rather than breadth across the space of human activities.”

“Thus, we use the word “atomic” here to mean a single activity that captures one specific functional operation and that is usually represented as a verb-object pair.”

“Unlike clustering-based hierarchies that group activities by statistical co-occurrence, our ontology groups activities by functional inheritance.”

“However, the AI-AI IAA is significantly higher than any of these human-referenced agreements, indicating that the models are more self-consistent than they are with human annotators, and than human annotators are amongst themselves.”

“For our purposes, the most important implication of these results is that the AI classification pipelines agree as much with human raters as other humans do, and this indicates that the automatic classifications are acceptable substitutes for human classifications.”

“To preserve structural coherence, the platform requires each node to be linked to at least one generalization, blocks edits that would orphan nodes, flags near-duplicate nodes (based on lexical + embedding similarity) to discourage redundant node creation, and maintains node-level change histories with attribution.”

“Across these variants, retrieval was the dominant bottleneck: if the correct node was missing or drowned in noisy candidates, downstream selection could not recover.”

“The full-context SPFO variant provided the best overall tradeoff between agreement with human judgments, specificity, hallucination rate, and cost, which is why it is used in the final pipeline.”

“| Human-Human | 0.525 [0.465, 0.586] | 0.584 [0.525, 0.649] | 0.557 [0.514, 0.609] | 0.258 [0.161, 0.349] | 0.348 [0.260, 0.433] | 0.307 [0.234, 0.372] | | AI-AI | 0.812 [0.726, 0.895] | | | 0.703 [0.566, 0.822] | | |”

“This imbalance suggests that, despite significant research progress in reasoning, planning, and analytical capabilities, these functions have not yet translated into widespread, monetized AI applications at scale.”

“This imbalance highlights substantial untapped opportunities for AI deployment, particularly in activities that involve deeper reasoning, decision support, and complex task orchestration, areas that may represent the next frontier for economically impactful AI systems.”

“The paper also describes a field test of this method in an online community of over 50,000 people who are developing proposals for what to do about global climate change.”

“A key insight of this paper is that just as markets in a supply web can combine the results of other markets, so, too, contests in a contest web can combine the results of other contests.”

“This means that integrated proposal authors can use any combination of other proposals they want to without having to get permission from the authors of the other proposals.”

“In other words, even though proposal authors in a single contest compete with each other, the CoLab Points system provides incentives for them to collaborate with authors in higher- and lower-level contests.”

“However, as Table 3 shows, of the 348 uses of proposals at higher levels, 28% were uses of their own work, and 72% were uses of work by non-overlapping author teams.”

“As Table 4 shows, the averages of these ratings for the finalist integrated proposals are not significantly different from basic proposals (except for the presentation dimension where they are somewhat worse).”

“We showed that it was possible for a diverse crowd of people from all over the world to simultaneously create multiple, sensible integrated proposals using a combination of their own and other’s work.”

“As shown in Fig. [2,](#page-3-0) when the human alone outperformed the AI alone, the combined human–AI system outperformed both alone with an average pooled effect size for human–AI synergy of g = 0.46 (t104 = 5.06; two-tailed P = 0.000; 95% CI, 0.28 to 0.66), a medium-sized effect[21](#page-8-10). But when the AI alone outperformed the human alone, performance losses occurred in the combined system relative to the AI alone, with a negative effect size for human–AI synergy of g = −0.54 (t104 = −6.20; two-tailed P = 0.000; 95% CI, −0.71 to −0.37), a medium-sized effect[21.”

“With most of the decision tasks studied in our sample, however, both the human and the AI system make a complete decision, with the humans usually making the final choice. Our results suggest that with these decision tasks, better results might have been obtained if the experimenters had designed processes in which the AI systems did only the parts of the task for which they were clearly better than humans. Only 3 of the 100+ experiments in our analysis explore such processes with a predetermined delegation of separate subtasks to humans and AI.”

“Only 3 of the 100+ experiments in our analysis explore such processes with a predetermined delegation of separate subtasks to humans and AI. With the four effect sizes from these 3 experiments, we found that, on average, human–AI synergy (g = 0.22, t104 = 0.69; two-tailed P = 0.494; 95% CI, −0.42 to 0.87) occurred, but the result was not statistically significant (see Supplementary Information section 2.6 for a more detailed discussion of these experiments).”

“Most (>95%) of the human–AI systems in our dataset involved humans making the final decisions after receiving input from AI algorithms. In these cases, one potential explanation of our result is that, when the humans are better than the algorithms overall, they are also better at deciding in which cases to trust their own opinions and in which to rely more on the algorithm's opinions.”

“In this case, we hypothesize that, since the people were less accurate, in general, than the AI algorithms, they were also not good at deciding when to trust the algorithms and when to trust their own judgement, so their participation resulted in lower overall performance than for the AI algorithm alone.”

“In our broad sample of recent experiments, the vast majority (about 85%) of the effect sizes were for decision-making tasks in which participants chose among a predefined set of options. But in these cases we found that the average effect size for human–AI synergy was significantly negative.”

“In order to apply the phi metric, we characterized the state of the group in terms of which group member was communicating at which point in time (see Methods).”