Collaborative AI     


Collaborative AI - overview

Many problems in private or public enterprises require collaboration (cooperation, competition, or coordination) from multiple agents (human or machines) to take a decision that maximizes an overall public goal or utility while balancing each agent's individual private goals. Collaborative AI refers to viewpoint that looks beyond an individual’s cognition to include goal-driven tactical, operational, and strategic interactions of individuals with others (including other non-human cognitive agents) in order to develop far superior collective intelligence through computational modeling/evaluations of such interactions/engagements among the collaborating agents.  While Core AI and Secure AI (please see the figure depicting AI dimensions, below) have been two orthogonal flag-bearers of AI for last couple of decades in the enterprise world, the emergence of the third essential pillar, Collaborative AI, completes the Enterprise AI technology space.


Core-AI is about basic machine learning, with recent focus on NLU, Bias-free and Fair AI, AutoML etc., Secure AI focuses on creating tamperproof AI applications, federated ML etc. Collaborative-AI, on the other hand, is about the maturing and evolution of an AI ecosystem towards a self-governed (via Cogniculture) community of Autonomous AI Agents collaborating to Learn/Reason/Decide towards the sustenance, survival and growth of the ecosystem itself. These autonomous agents have certain Core-AI functionalities that they can upgrade over time and that need to be secure (hack proof). The agents may participate in multi-agent Social Machines, necessitating secure AI capabilities. Game Theoretic constructs provide a framework to explore emergent behavior and social learning in multi-agent systems, and thus find natural coherence with Collaborative AI.                                  

IBM's Collaborative AI framework is built on 4 pillars: (1) Intelligent and Autonomous Collab-AI Agents are organized in a Social Machine built on an Agent-oriented Architecture in which the behavior of individual agents is defined based on a Socio-cognitive Architecture; (2) Collab-AI Agents realize high-end social skills through techniques, such as Multi-Agent Reinforcement Learning (MARL) and Theory of Mind (ToM), to enable high-value collaborations with sparse knowledge exchange; (3) Collab-AI Agents and all their declarative and procedural knowledge represented and organized in a Cognitive Knowledge Marketplace (CKM) that orchestrates agents and their skills with a goal to achieve strategic sustainability via pro-social evolution; (4) Game Theory (Behavioral, Epistemic, and Evolutionary PGG paradigms specifically) provides the ability to design mechanisms for effective engagement among AI agents and human business personas, in both single and multi-enterprise settings.

Our current in-market experimentation of this innovative technology is in the domain of Intelligent Workflows for Finance and Accounting processes (e.g. Lead-to-Cash) in the Enterprise B2B space. This agenda has resulted in more than 8 papers in top-tier conferences/journals and more than 10 disclosures filed/to-be-filed in the USPTO.

Please contact Gyana Parija ( for more information on this agenda.


    IBM Alumni
  • Sudhanshu Singh
  • Shrihari Vasudevan
  • Shweta Garg
  • Max Narayanan
  • Rishi Saket
  • Sarthak Ahuja
  • Manish Kataria
  • Monalisa Mohanty
  • Harit Vishwakarma
  • Ramasuri Narayanam
  • Karthik Visweswariah
  • Santosh Srivastava
  • Salil Gupte
  • Nidhish Pathak
  • Vishaal Munusamy Kabilan
  • Joydeep Mondal