The Imperfectionists: Strategic Mindsets for Uncertain Times
Robert McLean & Charles Conn
192 pages, Wiley, 2023
Nonprofits and companies have a tendency to assume away uncertainty. Many are satisfied with “strategies” that are essentially just three-year budgets, without reference to the probability of the desired inputs or outcomes occurring. In a world where uncertainty is front and center and accelerating—from pandemic to technology change and economic disruption—there is no normal to return to, so these weak planning efforts posturing as strategy look more and more like fantasy. Many organizations end up paralyzed, afraid to step into uncertainty and stuck in a cycle of doing more of the same.
There is another way, which we term being an imperfectionist. Imperfectionists learn to tolerate uncertainty and ambiguity by employing a suite of six strategic problem-solving mindsets. Imperfectionists are curious, they look at problems from several perspectives, and gather new data and approaches, including from outside their current industry. They deliberately step into risk, proceeding through trial and error, utilizing nimble low consequence and reversible moves to deepen their understanding of the unfolding game being played, and to build capabilities. Imperfectionists succeed with dynamic, real-time strategic problem-solving, confidently moving forward while others wait for certainty or make impetuous bets.
We have been researching strategy as problem-solving for three decades. Our first book, Bulletproof Problem Solving: The One Skill That Changes Everything (Wiley 2019), described the skillsets required for great problem-solving. The Imperfectionists adds strategic mindsets for developing organizational direction in uncertain times—like the ones we are in right now.
The book is based on 50 case studies, half of which come from nonprofits and the impact investment space. We hope you’ll enjoy the excerpt below, which features Collective Intelligence, a strategic mindset for uncertain times.—Charles Conn and Rob McLean
* * *
Joy’s Law
Bill Joy is well known as the cofounder of Sun Microsystems, now part of Oracle. He coined what has become known as Joy’s Law: “No matter who you are, most of the smart people work for someone else.” The logical corollary of Joy’s law is that you have to find a way of accessing that intelligence. As Bill Joy put it, “It’s better to create an ecology that gets all the world’s smartest people toiling in your garden for your own goals. If you rely on your own employees, you’ll never solve all your customer’s needs.”
For some it comes as a shock to have to see the world this way. We spent a lot of our careers trying to get the smartest people in the room, and linking ourselves to other experts outside the room. The implications for how organizations solve problems are huge. How does our recruiting need to change if we are drawing more on others in the ecosystem? Do we know how to put together really diverse teams? What role should so-called experts play when innovation demands new competence? How do we organize to put Joy’s Law, and more broadly collective intelligence, to work?
Framing Collective Intelligence
When collective intelligence is mentioned in conversation there is often much nodding that ‘yes, we know what that is—the wisdom of crowds is guessing how many beans there are in a jar.’ We think there is a lot more to it. We see collective intelligence playing a key role in how organizations compete. That suggests we should be respectful of experts, but sceptical of historically-derived perspectives in settings where the rules are rapidly changing. You should embrace diversity of perspectives in your own teams, and you should look outside your own four walls for rich sources of self-disruption.
Fishface and Sustainable Tuna Catch
Fishface is the name of The Nature Conservancy (TNC) project to use machine learning to identify the different species of tuna caught by fishing boats. It uses a variety of computer vision technologies to automate real-time at-sea collection of data on the species and numbers of fish caught, thereby enabling fisheries managers to make evidence-based resource management decisions. The project addresses a perennial problem in fisheries management, the lack of reliable knowledge about the size of catches of target and non-target species relative to sustainable harvest quotas. Globally 34 percent of fisheries are now overfished and another 60% cannot sustain any additional fishing pressure.1 Further, TNC estimates that 90% of all fisheries have no effective management in place.2 In the absence of precautionary management frameworks and easy ways to verify compliance with regulations on sustainable catch, conservationists fear that fisheries will collapse and this will have disastrous implications for the three billion or so people who count seafood as a significant source of animal protein. TNC’s solution is to turn real-time data on fish catch by species into a risk management tool that will allow large operators to validate that their catches comply with sustainability commitments and regulations. This would bring fisheries into line with supply chain verification of sustainability in land-based agriculture.
The catalyst for FishFace was TNC winning the popular vote of the Google Impact Challenge for Australia in 2016. Some of the prize money was used to identify a machine learning algorithm through a Kaggle competition that offered prize money of $150,000. Participants had to predict fish numbers of different species using video data collected from fixed cameras on board fishing vessels. The competition was open for five months and 2293 teams took part, making this one of Kaggle’s most popular competitions.
Participants received a training dataset of 3792 images and a testing dataset of 1000 images from the fishing boat cameras. The teams had to classify the fish into eight species, including yellowfin tuna, albacore tuna, dolphin and shark. A public leaderboard score was computed for results on the 1000 test images. Felix Yu, who came third in the public leaderboard stage, highlighted the challenges of crafting an algorithm from the video data, which included only small samples of some species, unclear images of fish fins (an important identifier), and the impact of wave action on the images.
Fast forward to 2022 and the good news is that the FishFace algorithms have been used onboard a fishing vessel in Indonesia with an accuracy of 90-95 percent, an acceptable level for reporting compliance. The next step is to develop a minimum viable product that will work on the 100,000 large fishing boats that account for 50 percent of the global seafood catch.3 The TNC team led by Mark Zimring is partnering with Amazon’s AWS, leaders in cloud computing, to develop an at-sea data uploading solution.
TNC’s FishFace is problem-solving an issue of global significance, harnessing real-time data capture with a pattern recognition learning engine to crack an age-old problem in fisheries management. It is a great example of the power of competitive crowd-sourcing to solve complex problems.
Ancestral Wisdom: ‘Right Way Fire’
Imagine sitting in a helicopter in Northern Australia next to an indigenous ranger feeding glycol incendiaries—fire bombs—each about the size of a golf ball, into a machine. The machine drops these mini incendiaries onto the tropical savanna grasslands below, where they ignite into small fires. In that seat, you would be learning firsthand about a partnership between indigenous rangers and conservation groups, including The Nature Conservancy Australia, where Rob is a Trustee. This has led to the reintroduction of millennia old fire management techniques into Australia, now adopted as a model for the whole country.
In contrast to modern western practices of fire management that focus on fire suppression, Australian indigenous communities have been relying on early dry season burning for tens of thousands of years. Called ‘right way fire’ by these communities, it’s an approach which benefits land management and prevents destructive, larger-scale fires. It’s a kind of collective intelligence that we think of as ancestral wisdom—historical problem-solving solutions that had been ignored or forgotten.
Western science complements this set of practices, passed down the generations, with modern experiments that gauge the amount of greenhouse gases emitted from the savanna, and satellite mapping that shows the extent and intensity of the fires. Carbon credits are generated based on the net abatement of greenhouse gases, compared to a baseline of emissions that occur in the absence of early dry season burns. The Australian Government plays its part by registering the carbon credits which are produced, which can then be either sold to government or on the voluntary carbon market.
Conservationists, meanwhile, celebrate the positive impact on flora and fauna of the carefully managed mosaics of unburnt country. These allow birds and animals safe corridors between burns, and promotes plant and seed growth. The rich blend of roles is an extraordinary example of collective intelligence.
Tropical savannas, which represent 16 percent of the earth’s land surface and are present not only in Australia but Africa, South America and parts of Asia, are the most fire-prone vegetation on earth.
The dramatic progress brought about by the reintroduction of early-season burning over the last decade is visible in the two satellite image-derived maps of Northern Australia. The areas coloured red in 2009 are where late and more severe burns occurred, while the green segments are early dry season, or managed, burns. The top section of the chart is a high rainfall zone where ranger teams were in place and carbon projects registered. The middle section is largely Arnhem Land where Otto Campion’s people pioneered the contemporary approach to managed fire regimes. The top right hand segment is Cape York where managed fire regimes were yet to come into place. In Cape York the low rainfall shows a lot of damaging late-season burns, not surprising in the absence of carbon methodology and ranger groups.
Fast forward to 2021: Wildfires have largely disappeared from the high rainfall zone where traditional burning now takes place. It’s an extraordinary change in a landscape, probably one of the most significant we are aware of on our planet. The other change is that there has been a sharp reduction in late burns in the low rainfall zone compared to a decade ago. You can also see some carbon projects extending into the low rainfall zone.
All this has come about because the Australian Government accepted that the indigenous people who managed this land for millennia should return to their historical fire and land management practices, now in partnership with modern science. The combination is a powerful example of collective intelligence benefiting climate, nature and people.
AI Swarms
The intersection of individuals and teams who are utilizing tandem human-machine capability is growing rapidly. Take AI ‘swarm platforms’ for prediction. In a contest to forecast the outcome of 50 English Premier League soccer games, swarm participants correctly predicted 72 percent of winners, while those doing it in ordinary crowds or alone in isolation, were correct only 55 percent of the time. This amounted to a 31 percent increase in accuracy where participants were connected in AI swarms.4 The difference is that participants in swarms ‘think together’ in real time and converge on solutions through interactions governed by algorithms. The swarm process is based on bees, schools of fish or flocks of birds. A corporate leader in this space is Unanimous AI, which describes a swarm as a ‘brain of brains,’ able to achieve super-intelligent results and outperform all the individual members.
How does an AI swarm compare with findings from deep learning? A study at Stanford medical school found that groups of doctors using Swarm AI algorithms were 22 percent more accurate in making diagnoses than the most advanced deep learning algorithm that only used historical data. Clearly, having humans connected with a swarm is producing encouraging findings. DeepMind has made significant progress in creating Alpha Code to write computer code at a competitive level, where it now ranks in the top 54 percent of participants in programming competitions.5 In the spirit of collective intelligence, DeepMind is putting the dataset of problems and solutions on GitHub to spark innovation in problem-solving and code generation.
The cases of AI enabled collective intelligence illustrate a broad suite of problem-solving applications. The traditional model of internal teams focused on problem-solving has an important role to play, but even the most experienced experts will have their views placed in perspective by readily available swarm platforms.
That said, while AI and machine learning will become familiar tools for collective intelligence, they will not always hit the mark as we saw with attempts to use them to design treatment protocols for Covid 19 early in the pandemic. We know that when uncertainty is high and data is thin, AI’s pattern recognition capabilities are at their weakest.
Collective intelligence from crowds also has its limitations. Todd Rose reminds us of collective illusions that occur when individuals subsume their private view to conform with what they think the group wants.6 An example is the tulip mania of 1835 when tulip prices exceeded their weight in gold. He also points to climate change in the US, and the gulf between the private view reflected in its ranking as the #3 most important issue for individuals, and its #33 ranking when people are asked what the public thinks. The way to avoid collective illusions is to keep asking why, as the child in the Curiosity mindset would have us do, and ‘is it really true?’ to avoid the fate of the Emperor whose new clothes revealed all.