In this blog, Dr. Sean Fitzgibbon, PhD, Data Scientist at Fivecast introduces the concept of augmented intelligence; exploring the value of leveraging artificial intelligence to enhance rather than replace the decision-making processes of humans and how this can significantly enhance OSINT best practices.
What is Augmented Intelligence?
Augmented intelligence (also known as intelligence amplification) refers to the process of using technology to assist and enhance the decision-making performance of humans. Augmented intelligence proposes a partnership between technology and humans so that the combined decision-making performance is greater than the sum of its parts. This symbiotic relationship between humans and computers was first described by computing pioneers as early as the 1950s. However, the modern derivation of augmented intelligence typically defines artificial intelligence (AI) and machine learning as the specific tool used to augment human decision-making. This is driven partly by the modern explosion of AI performance, but mostly due to the ability of AI to learn from the human, which makes it a natural fit to complement the human and create a truly symbiotic process for augmenting intelligence. As such, augmented intelligence is often considered a subset of AI.
The intersection of Artificial and Human Intelligence
Traditionally, commentators have pitched AI as our competitor, with the ultimate goal of AI research being to outstrip human performance. This approach has led to some genuinely impressive results; however, this is usually in very narrow tasks. It is still debated as to whether AI will ever be able to reach or exceed human performance in the wide scope of intellectual skills that humans naturally excel. Nonetheless, current state-of-the-art AI can perform some specific tasks much better than a human.
Augmented intelligence does not pit AI against humans but seeks to leverage the complementary strengths of both the AI and the person, to deliver a powerful framework for decision-making. A typical augmented intelligence workflow, such as open-source intelligence (OSINT) gathering, might involve using AI to process large volumes of data, very quickly without fatigue or attention issues, and identify and extract patterns in the data that might be valuable to the human analyst. In turn, the analyst reviews these distilled patterns and applies their insight to interpret, extract knowledge, and make improved decisions. Ideally, to complete the loop, the analyst would then provide feedback to the AI, allowing it to learn to process the data better. Depending on the context, augmented intelligence may rely more on AI or on the human, but ultimately the final decision depends on the human which can be especially important for various regulatory and ethical reasons.
Augmented intelligence in the wild
A major strength of augmented intelligence is that it can be immediately implemented by organizations. They do not need to wait for the perfect AI to be developed because the human partner can understand and compensate for any specific weakness in the AI. As such, many organizations have already adopted augmented intelligence strategies to add value to their business. Here we present just a couple of examples.
Netflix and chill
Content providers, such as Netflix, use AI-enabled recommendation systems to suggest content for a viewer by comparing their interaction with the service to that of other users with similar tastes and preferences. These tools surface the content with the highest probability of user interest to enhance the user’s ability to make a good viewing decision. However, the ultimate viewing decision is still deferred to the user. Furthermore, the ongoing interaction of the user with the service (e.g., viewing history, title ratings, etc.) provides ongoing feedback to the AI, allowing it to improve its suggestions over time.
Enhancing medical diagnosis
Medical diagnosis is an obvious candidate for augmented intelligence given regulatory constraints that require oversight by medical professionals and because, as a society, we do not yet have the trust in technology to hand over such a critical task to AI alone. In a recent grand challenge to evaluate tools for automated detection of breast cancer, the winning team presented a deep-learning AI tool that could identify cancer with an impressive accuracy of 92.5% (AUC) which is almost as high as a human pathologist at 96.6% (AUC). However, even more impressively, when the AI was used to support the pathologist’s decision-making in an augmented intelligence style approach, the performance increased beyond both the pathologist and the AI to 99.5% (AUC). Here, the augmented intelligence is greater than the sum of its parts, and furthermore, regulatory and trust concerns can be mitigated by having the human in the loop.
Staying safe on the roads
AI for autonomous driving of vehicles has progressed rapidly in recent years, and advocates proclaim that the technology is ready for deployment. Irrespective, society does not yet have sufficient trust in the AI to permit fully autonomous driving and therefore regulatory constraints requiring human drivers exist in most jurisdictions. However, this has not stopped manufacturers from integrating AI technology into modern vehicles resulting in an augmented intelligence driving scenario. It is common in modern vehicles to have technology such as automatic braking to stop the vehicle if a collision is detected, lane support systems to advise the driver if they are veering out of their lane and potentially to correct the steering, and driver monitoring systems to notify the driver if they are not sufficiently alert. All these systems enhance the decision-making and performance of the driver but leave the ultimate responsibility of operating the vehicle to the driver. If proponents of autonomous driving are correct, then we should continue to see integration of new augmenting technologies into vehicles, along with a gradual prioritization of the AI over the driver.
Augmented intelligence for OSINT
OSINT is the process of collecting, analyzing, and extracting meaningful insights from publicly available data sources including social media, news feeds, blog sites, discussion forums, and more. OSINT is particularly challenging for the human analyst because of the overwhelming volume of multi-media data, the velocity at which topics of interest arise and then disappear, and the variety of platforms and structures in which the data is represented. However, whilst challenging for humans, this is where AI has its core strengths.
Fivecast ONYX is an augmented intelligence platform for OSINT that uses advanced data collection and integrates AI-enabled analytics and machine learning technologies to enhance the decision making of skilled intelligence analysts to detect threats in vast amounts of unstructured multimedia data. Numerous AI tools are incorporated into the Fivecast ONYX workflow including multi-lingual text detection, AI-expanded term detection, quote and text similarity detection, object and concept detection, user trainable image and logo detectors, and sentiment and emotion scoring. This framework helps the analyst to focus on the high-risk data to more efficiently identify threats and respond accordingly.
The Way Forward for Leveraging AI
Augmented intelligence has been referred to as the “real AI” and it has been proposed that it is the “right” way forward for integrating AI in our lives. By retaining the human in the loop, augmented intelligence can mitigate many of the barriers to AI adoption including regulatory constraints, ethical concerns, and public trust, all of which are essential considerations when deploying OSINT. Furthermore, the performance of augmented intelligence can surpass both AI and human performance alone by leveraging the complementary strengths of the person and the AI. We witness this first-hand in our customer base when intelligence analyst teams can leverage automated and AI-enabled risk analytics technology that frees them up to apply their considerable experience and human insights into solving complex intelligence challenges faster. Finally, augmented intelligence is a framework that can be implemented immediately by organizations to improve existing decision making and add value to their business.
Augmented intelligence solutions are not static. As AI develops, and trust in the AI deepens, then the relative importance placed on the AI in the human-AI partnership will naturally increase, and in some contexts the human may no longer be needed in the loop. However, in a great many areas it seems unlikely that people will ever be fully replaced, rather the role the person plays will adapt over time to complement the AI.
