Learning from the Birds and the Bees: Start Looking at Swarm AI to Harness the Power of Human Insights and Opinions for Your Enterprise

For several decades, computer scientists and artificial intelligence (AI) specialists have been studying the behavior of social animals and insects to try and export the benefits of such systems beyond the natural world. Since the 1980s, AI specialists have been developing algorithms modeled on natural swarm behaviors to build AI systems to mimic natural swarm systems’ efficiency and accuracy. The application of such “swarm intelligence” to AI has become known as “swarm AI.” More recently, advances in AI and the rise of digital and online interactions have moved swarm AI algorithms out of the field of academia and into real-world scenarios, where they are already delivering tangible benefits for enterprise users across functions and industries.

In particular, swarm AI is proving itself adept at helping organizations—and individuals—make complex decisions that require tapping into the collective opinion of large groups of people, whose opinion reflects public sentiment or the amalgamated knowledge of specialists. Just one example includes amazon founder Jeff Bezos turning to his Twitter followers to ask how to best invest his money in a worthwhile cause. The overwhelming number of responses was subsequently processed by up and coming player Unanimous AI to cull the group’s collective wisdom. Swarm AI is rapidly being adopted by the healthcare, financial, and market research firms in particular. Organizations within these industries especially should start evaluating swarm AI now or risk falling behind their competitors in a cut-throat marketplace.

Real-world applications of swarm AI technology within enterprises

The use of swarm AI to solve internal or consumer-facing business intelligence issues by making more efficient use of enterprises’ existing resources is growing rapidly. The growing connectivity enabled by social media and other online platforms is further accelerating adoption and use cases. Across multiple industries, swarm AI is proving adept at optimization tasks, routing conundrums, and staff scheduling challenges, to name but a few. Industries pioneering the use of swarm AI include:

• Doctors, including radiologists and oncologists, are leveraging platforms like Unanimous AI’s to make more accurate diagnoses by opening a case up to the scrutiny of their professional peers to leverage their collective expertise and converge on the best answer. A study by Stanford University showed that eight radiologists using Unanimous AI’s platform could diagnose pneumonia more successfully based on chest x-rays than the individual doctors or even a machine learning system could.
• Banking and financial services. Investment banks, hedge funds, and other institutions that invest on clients’ behalf are expressing strong interest in swarm AI tools. By drawing on the intuition and expertise of a broader population and by gauging market behavior on a large scale, professionals at such financial organizations managing portfolios or trading stocks can better predict the probability of a positive or adverse event that could impact the performance of the assets in their remit. They can draw on the expertise of all the experts within their own organization or external professionals.
• Polling and market research. Traditional field surveys and exit polls are increasingly failing to predict the outcome of political elections, including the last US presidential election and the result of the UK’s Brexit referendum. Instead, polling firms are turning to swarm AI systems, which factor in regional and local variances instead of removing them. As such, they yield a more accurate representation of a population’s sentiment and intent. Market research firms are also following suit, leveraging the same techniques to get a better idea of, for instance, which film is most popular with critics and audiences despite their box office revenues.
• Major telecommunications firms including British Telecom, France Telecom, and MCI WorldCom are already using swarm AI tools to ease congestion in their internet traffic. In a telecommunications network, some routes are heavily trafficked at certain points in the day, while others are left clear. Hewlett Packard engineers built a suite of “digital ants” that can be deployed to scan routes and alert engineers to uncongested ones where surplus traffic can be redirected, thus disrupting traffic more equally across the network. This means better customer service and hence improved client loyalty for such firms.
• Manufacturing and logistics. Manufacturers have begun turning to swarm AI to optimize their production lines and factory floors, mimicking how bees draft in non-worker bees during peak times to get tasks accomplished more efficiently. For instance, Unilever used swarm AI algorithms from the Bios Group to respond in real-time to machinery breakdowns or repair delays by automatically updating production schedules around the unexpected event to ensure they did not cause production delays, deploying people to parts of the floor that they are not normally assigned to in order to maintain the pace.

Swarm AI leverages the best of the natural world to enable enterprises to make informed decisions efficiently

“Swarm intelligence” (SI), the concept underpinning both natural and artificial swarms, refers to the “collective behavior of decentralized, self-organized systems, natural or artificial,” and it was first used in a computer science context in 1989. Populations that display such behavior are known as SI systems. They consist of multiple independent agents whose local and largely random interactions generate broader “intelligent” behavior patterns that the population as a whole engages in to solve complex problems. Artificial SI systems, in turn, are sets of algorithms programmed to follow SI patterns observed in nature. More commonly, these algorithm sets are known as “swarm AI.”

Despite being small, birds, fish, ants, bees, and bacteria engage in group behaviors that allow them to solve complex challenges such as defending themselves effectively against predators, finding the shortest route between two points, and scouting for and communicating the location of a food source. Individually, these organisms would be unable to solve such complex challenges and achieve optimal outcomes. But, by pooling their resources, they can accomplish what organisms their size otherwise couldn’t. For instance, when bees are scouting for locations to establish a new nest, the hive has to converge on an option together. Hundreds of worker bees are dispatched to scout the territory, and upon return, they perform a dance that communicates their findings to the group. Individual worker bees have to try to influence the rest of the group to follow them to their preferred direction and keep vying against each other until one scout persuades the rest to follow.

Swarm AI is finding the most traction in the form of “middleman” software, for example, a digital platform that organizes scattered networks of individuals into “swarms” through real-time, closed-loop systems. These systems allow individuals to act as a “hive mind” to form predictions and answer queries and polls. In testing, human swarms have repeatedly proven capable of amplifying human knowledge and generating predictions far more accurately than individuals could in sports match outcomes, elections, and entertainment award programs. Enterprises are looking closely at swarm AI because it allows them to mine maximum value from readily available resources—the knowledge, skills, and opinions of employees, customers, and members of the public—quickly and cheaply. There are no leaders or permissions in swarms, meaning decisions don’t have to go through a central authority and that information is easily and rapidly shared. This makes swarm AI decision-making very rapid and transparent. As we shall see, swarm AI offers numerous advantages over other forms of AI and analytics for enterprises interested in more informed decision making at a lower cost.

In particular, swarm AI boasts several characteristics that ought to make it interesting to enterprises wanting to reduce expenditure and improve their quality of service:

• Efficient use of resources by organizing knowledge that already exists in the company. Drawing on employees, customers, or a public demographic, depending on the task at hand, swarm AI allows enterprises to leverage the collective intelligence of a population that would otherwise have remained incoherent, dispersed, and unusable.
• Amplification of knowledge by feeding human input into a powerful AI model. Swarm AI systems allow enterprises to combine the awesome analytical and organizational capacities of an AI system with human strengths such as instinct and nuanced judgment to get a fully rounded perspective on a given situation.
• Accuracy and granularity by getting all participants in a swarm to submit an answer simultaneously rather than linearly. Swarm AI can ensure that earlier respondents do not influence or bias subsequent ones, as has been known to happen in elections and prediction markets. Rather than merely finding the average sentiment of a given group, swarm AI factors in all outliers and produces the answer that the entire collective has converged upon.
• Explainability and auditability by using human knowledge as input data. This inherently makes a swarm AI system more understandable and transparent. Although the input data may, of course, contain biases held by the human participants, it is at least clear where those biases came from, avoiding “black box” AI. At a time when the need for explainable AI is intensifying, swarm AI could ease the regulatory burden and risk of reputational damage for enterprises using AI.

The diversity of swarm AI offerings keeps growing—start scouting the market to find one for your company

Thanks at least in part to the years that swarm AI has spent in an academic context and the value it displayed there, the technology has found a comfortable home with enterprise buyers, and demand has kept growing. The market is a large one for specialist companies that can find a niche in it, and several candidates are developing swarm AI solutions tailored specifically to the needs of large enterprises:

• Unanimous AI. Launched in 2014, US-based Unanimous AI is arguably the best-known swarm AI vendor in the market. The company has developed a cloud-based platform that allows users across the globe to log in and become participants in a real-time, closed-loop intelligence system or swarm moderated by AI and run on deep neural networks. In 2016, Unanimous AI correctly predicted the outcome of the Kentucky Derby Superfecta, netting betters who followed the hivemind’s advice hundreds of dollars on $1 bets. The platform amalgamates users’ combined knowledge, biases, and opinions into a single answer to a given question. Mimicking a bee swarm, users answering a question have to move an icon across the screen, moving with or against the collective, and try to persuade others to move their icons towards their preferred outcome. By getting everyone to answer at the same time, Unanimous AI ensures that earlier responses don’t bias subsequent ones; by factoring in all responses, it achieves an answer far more accurately than an average derived in a poll or survey could. This approach is yielding powerful results. The company says that such public-facing predictions are simply good testing grounds for the platform and that it’s focused on selling to enterprises to amplify their business intelligence. It already has large clients from medicine, the public sector, and banking and financial services.

Exhibit 1: The Unanimous AI platform in action: processing input data to generate reliable answers

Source: Unanimous AI (screenshot)

• Convergent AI. A US-based company, Convergent AI (previously Axon AI) launched in 2014 and, uses swarm intelligence and poly-agent principles to build predictive analytics solutions that help large enterprises solve business challenges more efficiently and with more of their own company data. Its learning algorithms prioritize variances in input data, attuning insights to dynamic local data. The algorithms have also been optimized to ingest as many data types as possible, thus enabling companies to leverage the vast amounts of business intelligence they normally leave untouched. In particular, Convergent AI focuses on helping multiple industries deal with challenges including cybersecurity threat assessment and response; IP and asset protection; maximizing marketing campaign effectiveness; and streamlining logistics, for instance ensuring flights are taking off on time.

Exhibit 2: Convergent AI PatentID swarm AI solution

Source: Convergent AI (screenshot)

• US-based Lexalytics, founded in 2003, is taking a highly modular approach to swarm intelligence. The company has been developing AI modules which, despite functioning as standalone analytics solutions with specialized functionalities that can be mix-and-matched, deliver their full value by cooperating as a single swarm super-solution. This interaction behavior has been modeled on bird-flocking patterns, which enables the modules to adjust their actions based on real-time feedback loops from its fellow modules. As with Convergent AI, Lexalytics’ algorithms learn in real time from input data variances and can self-adjust to imbibe new forms of information as they appear. Together, the modules cooperate to realize whatever objective the client has set. The modules are particularly well-suited to executing functionalities including data analytics and data visualization for unstructured data. Lexalytics is targeting its solutions at data scientists and business intelligence professionals within organizations that have to process over 100,000 documents a month and have strong data security and privacy requirements, for example, regulators, healthcare companies, and banks.

Exhibit 3: Lexalytics technology features

Source: Lexalytics (screenshot)

Risks of and best practices for implementing swarm AI in your enterprise

Despite the benefits such enterprises are already reaping from swarm AI, awareness of the technology’s potential shortcomings and effective techniques for avoiding or at least mitigating them are essential for successful swarm AI projects. Below, we look at some of the major risks and emerging best practices for enterprises interested in or already using swarm AI should have on their radar.

Inevitably, due to the complexity of any form of AI technology, and the even more complex nature of the human beings acting as the input data in swarm AI systems, swarm AI comes with risks and challenges enterprises should be aware of. These include:

• “Death spiral” behavior, where ant colonies blindly follow the swarm to their doom. SI systems need to watch for rogue players and incorrect opinions, particularly in use cases such as medical diagnoses and stock trading.
• AI talent is thin on the ground, making it hard for enterprises to attract and retain it. With swarm AI, there is a double ask for talent: to maintain these systems, some knowledge not only of AI but also of natural systems is required, narrowing the talent pool even further and leaving users dependent on third-party expertise.

However, the good news is that since the risk factors posed by swarm AI are largely to do with human nature, there are some easy steps enterprises can take to mitigate them. Best practices for implementing swarm AI safely include:

• Implement robust change management. Enterprises looking to leverage swarm AI have to prioritize education campaigns to get participants on board and understand the technology’s value proposition, given how dependent swarm AI is on human cooperation.
• Start in low-stakes domains and expand. Given the risk of “death spirals” and the potential lack of swarm AI experts within an end client’s organization, a sound best practice for enterprises would be to apply swarm AI to relatively low-stakes domains and functions as a test-bed, and scale up from there.

Bottom line: Human insight can be turned into an invaluable resource. Start looking at swarm AI to do the job, because your competitors will be.

Human intuition and opinions are becoming an increasingly invaluable data source as the Digital OneOffice model, which prioritizes consumer satisfaction, becomes the standard for all industries. Swarm AI holds immense potential because it can do the heavy lifting in filtering this data. Swarm AI is a perfect example of humans and machines working together and doing what they do best to deliver superior outcomes for enterprise users.

Swarm AI makes quantifying people’s intuition, opinions, beliefs, and convictions a reality, organizing an otherwise subjective and ephemeral form of information into a usable format, enabling enterprises to gauge what consumers, the public, and their employees are really thinking. Make no mistake—your competitors will be looking closely at swarm AI. Don’t get stung—be a busy bee and start doing your swarm AI research.