SustainChain is AI for Good
At the US Coalition on Sustainability, we believe in the power of advanced digital technology to quickly guide leaders toward the products, partnerships, and solutions required to achieve the UN Sustainable Development Goals. At the core of the SustainChain™ platform is a data engine that continually learns from its users and their efforts - and we named it DaVinci™ after the eminent inventor’s intelligence and vision. Learn more about SustainChain from Jarno Kartela, lead architect of our DaVinci technology.
How will AI help you solve the problem that SustainChain proposes to address?
Given the scale and complexity of SustainChain’s mission; to drive meaningful action recommendations across the entire Sustainability ecosystem), it would be very inefficient to match the right member to the right action initiative, to the right organization, to the right activity, to the right content or to another member in the right time in any other way than with data and applied artificial intelligence. Should we attempt to do that, the resulting logic would infinitely scale to different if-when statements until impossible to manage.
Therefore, SustainChain uses a reinforcement learning based, self-optimizing and self-exploring artificial intelligence system called DaVinci as its backbone, resulting in a fully real-time data-enabled optimization and recommendation system that can ultimately guide any recommendation decision between different actors in the platform and do that in a way that constantly helps the platform to learn, adapt and evolve, improving both overall impact and the experience for its members.
What datasets do you have?
The core data asset - is a fully real-time stream of signals which include a triplet of member, object and weight that is used to online train DaVinci whenever new data is created in the platform via interaction or via the creation of new objects (like action initiatives, organizations, content, or new members).
An analytical data asset – rich analytical warehouse and object storage-backed data asset consisting of every piece of data that is created in the platform, that is enriched from outside sources or that is directly created by DaVinci during the constantly ongoing reinforcement learning process. This can be used for offline policy optimization, feature engineering, analytical reasoning, monitoring, reporting or model retraining purposes.
The core idea of DaVinci is to fully utilize all data assets available while also creating completely new, mission-critical insights from exploring what past data only does not tell us.
Do you currently have access to this data? If not, how do you plan to collect or access them?
Access to data explained above is available in real time, but there will be a need for enriching the data asset with external sources to further improve capability in serving SustainChain’s community and mission. That said, the most interesting data asset is created by the constant exploration of the reinforcement learning system, as it has the capability of exploring the impact of any new object in SustainChain immediately after it is created. The resulting data asset is critical since it allows the platform and its users to truly learn something new, instead of solely looking at or mirroring past performance.
Tell us about how you would use the data in an AI model. What data would your model consume, and what information or decisions would it produce?
All data created in the platform, enriched from outside sources or created by the AI system itself is fed back to the intelligence engine for learning purposes as fast as possible. This is done in real time through online learning and the intelligence engine itself does optimized exploration of all new assets as they are created.
The model consumes a real-time stream of events which is created partly as a byproduct of the reinforcement learning system and its recommendation mechanisms and more importantly as a result of more and more users interacting with the platform to further their initiatives. The model also consumes features about the objects being recommended or acted on which can range from typical object metadata assets to enriched, industry-specific features.
Importantly, there is virtually no separation between learning and providing decisions – they are two sides of the same coin in SustainChain’s setup. The intelligence engine can optimize for any decision in the platform that has a set of choices, like recommending action initiatives to members or recommending members to other members and will learn from its own actions in real time. Since it does exploration - which is effectively just very intelligent randomization, it is much less prone to bias from historical decisions and less prone to create unwanted ethical concerns as all possible futures are explored.
How does SustainChain currently use AI?
SustainChain’s backbone is its intelligence engine DaVinci, which runs recommendations in the platform and creates new insights through its online learning and online exploration mechanisms.
Importantly, the recommendation system is a reinforcement learning based system which allows it to recommend anything that has a set of choices. It is by no means a standard product/content recommendation system designed for a single recommendation purpose, but a multifunctional decisioning engine that is used to optimize all relevant decisioning problems occurring on the platform.
Given the scale, complexity and impact of SustainChain, there are limited possibilities of running the platform with anything else than a AI- and data-backed intelligence engine nor there are better ways to achieve the amount of new insights and understanding.