Beyond AGI: Symbiotic Intelligence, Superintelligence, and the Quantum Frontier
Updated: Oct 23
The Future of Intelligence May Not Be a Single Machine, but an Interconnected World of Human and Artificial Minds
DEW Diligence, LLC | October 23, 2026 — DEW Diligence is not affiliated with Google, Google DeepMind, or the DeepMind Institute; this is independent commentary.
Artificial Intelligence (AI): What Comes After the Individual Machine?
For decades, the pursuit of AI has centered on building capable machines. From early expert systems and machine learning to large language models (LLMs), generative AI, and autonomous agents, technological progress has been measured largely by what individual systems can accomplish. The pursuit of artificial general intelligence (AGI) and artificial superintelligence (ASI) raises difficult questions. What does this mean for the future of technological convergence? What happens when capable systems begin working together, coordinating decisions, exchanging knowledge, and performing even more advanced intelligence activities that extend beyond the capabilities of any individual model?
The answer may challenge how we understand intelligence itself.
On September 24, 2026, Benjamin Bratton (Visiting Researcher, Paradigms of Intelligence, Google), Blaise Agüera y Arcas (VP and Fellow, CTO of Technology & Society, Google), and James Manyika (President, Research, Labs, Technology & Society, Google) published Artificial Symbiotic Intelligence: Agents, AGI and the Orchestration of Many Minds through the DeepMind Institute. Their essay explores a future in which advanced intelligence emerges through interactions among artificial agents, humans, technologies, and institutions. Its sharpest claim is that the popular "Singularity" narrative — a single, titanic model bootstrapping itself to isolated, godlike intelligence — is likely the wrong vision. It is a compelling perspective as organizations explore autonomous systems, interconnected AI agents, and the possibility of intelligence operating beyond traditional human capabilities. The essay is a forward-looking conceptual argument, not an experimental demonstration that AGI has been achieved. The DeepMind Institute also clarifies that its essays represent the authors' perspectives, not Google's official position.
Interesting question: if intelligence becomes distributed, autonomous, and interconnected, who governs its decisions, determines its boundaries, and remains accountable for its consequences?
Artificial Symbiotic Intelligence: A Different Vision of AGI
The DeepMind Institute essay challenges the popular narrative that AGI will emerge from a singular machine that eventually exceeds human cognitive capabilities. Instead, the authors consider intelligence as a collective phenomenon, drawing parallels to biological evolution, human societies, and the development of institutions.
Human civilization already demonstrates the power of collective intelligence. Scientific discoveries, economic systems, governments, businesses, and technological infrastructure depend on people specializing, exchanging information, challenging assumptions, and coordinating their activities. No individual possesses all the knowledge necessary to design, manufacture, distribute, regulate, and maintain a modern spacecraft. Yet interconnected organizations and specialized professionals accomplish precisely that.
Artificial symbiotic intelligence extends this concept into a world where human participants and artificial agents contribute to interconnected systems of cognition and decision-making. The essay introduces several possibilities worth examining.
Intelligence Through Collaboration
The authors observe that the best AI agents for many applications today are not single models but frameworks that divide labor across teams of models. An advanced AI system may not require every capability within one model. Specialized agents could perform research, develop software, evaluate financial information, assess cybersecurity exposure, or analyze scientific questions. These agents could exchange findings, challenge conclusions, coordinate workflows, and produce collective outcomes.
An individual model might be less capable than another model in isolation, yet a well-designed network of specialized agents could outperform either system working independently. This distinction is important. The intelligence of a system may depend as much on its architecture, coordination, shared knowledge, and institutional controls as on the capabilities of its underlying models.
The Cognitive Crossover Point
The authors also consider the possibility of a cognitive crossover point: a future transition in which artificial agents collectively perform more cognitive work than humanity. Industrialization dramatically expanded society's capacity for physical production. Autonomous AI could eventually introduce a comparable transformation in knowledge work, analysis, administration, research, and digital operations.
Whether or when a crossover threshold will occur remains uncertain. The essay presents this as a possible historical transition, not an established measurement. Nevertheless, the implications deserve attention. A growing population of interconnected artificial agents could alter workforce structures, business economics, scientific discovery, and the relationship between human decision-making and machine execution.
From Individual Agents to Artificial Institutions
Perhaps the essay's most consequential argument concerns the development of agent institutions. The authors envision nested structures — hybrid human-AI networks and agent-only systems — in which participants occupy defined roles, operate under shared procedures, challenge decisions, and learn through coordinated interactions. The resulting intelligence would reside not in any individual agent but in the embedded rules, procedures, precedents, and feedback mechanisms that connect them. The authors point to recent evidence, published in Nature in 2026, that orchestration harnesses can outperform nominally more intelligent single models. They also reframe alignment: rather than a constraint engineered in advance, they see it as a co-evolutionary outcome of sustained contact between humans, agents, and institutions.
This is an intriguing possibility. It also introduces a significant governance challenge. A coordinated system does not necessarily produce trustworthy results. Multiple agents can reinforce incorrect assumptions, propagate unreliable information, or collectively execute inappropriate actions. The AI Incident Database recorded 362 documented AI incidents in 2025, up from 233 in 2024, according to the 2026 AI Index. The organizational structure must therefore be designed to support both performance and accountability.
AGI, Artificial Superintelligence, and Artificial Symbiotic Intelligence: Understanding the Distinctions
As AI capabilities advance, terminology influences public expectations, commercial investments, and government policy. Three concepts deserve attention.
Artificial General Intelligence (AGI) generally describes AI capable of performing a broad range of cognitive tasks at approximately human-level proficiency or beyond. Definitions differ across research organizations, particularly regarding generality, autonomy, reliability, and performance requirements.
Artificial Superintelligence (ASI) typically refers to AI that substantially exceeds the cognitive capabilities of the most capable humans across virtually all tasks and relevant domains.
Artificial Symbiotic Intelligence is the DeepMind Institute essay's vision of collective intelligence emerging through interconnected human and artificial agents. Its emphasis is on relationships, coordination, and institutions, not simply a single system reaching a particular level of cognitive performance.
Importantly, artificial symbiotic intelligence is not necessarily artificial superintelligence. The word symbiotic describes a form of interaction and organization; superintelligence describes an exceptional level of cognitive capability. The abbreviation ASI may be used for both concepts in different discussions. Furthermore, an AI model can outperform human specialists on specific benchmarks without possessing AGI. Likewise, an interconnected society of powerful AI agents does not automatically constitute artificial superintelligence.
Google DeepMind's research, Levels of AGI for Operationalizing Progress on the Path to AGI (Morris et al., ICML 2024), provides a useful framework for evaluating intelligence through performance, generality, and autonomy. Its lesson is that extraordinary performance in one domain should not be mistaken for extraordinary intelligence across all domains. As of October 2026, no widely accepted, independently substantiated demonstration has established that artificial superintelligence has been achieved.
The capabilities of leading systems are remarkable and improving, but the International AI Safety Report 2026 — chaired by Yoshua Bengio, authored by more than 100 experts, and backed by nominees from more than 30 countries, the EU, the OECD, and the UN — documents significant differences between performance on advanced evaluations and reliability in unfamiliar, real-world situations. There is still no universally accepted test establishing precisely when AGI begins or when a system becomes superintelligent. That uncertainty calls for scientific investigation rather than premature declarations.
Superintelligence: An Industry, a New Name, or a Demonstrated Capability?
The distinction between technological capability and terminology has become especially relevant in the United States. On September 29, 2026, Executive Order 14434, Inaugurating the Era of Super Intelligence, was signed and published in the Federal Register on October 2, 2026.
The order directs federal executive departments and agencies, to the maximum extent permitted by law, to use the terms "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" in official correspondence, public communications, websites, reports, policy documents, and other non-statutory documents. For purposes of the order, SI initially encompasses the technologies and systems covered by the existing statutory definition of artificial intelligence under 15 U.S.C. § 9401(3). The order also requests, within 60 days, proposed legislation establishing a federal definition of Super Intelligence. It changes the words the executive branch uses; it does not, by itself, change the legal scope of the statutory definition or any obligation on private companies. This represents an important terminology and policy development.
When does AI actually become superintelligence?
A credible assessment would require independently verifiable evidence of extraordinary generality, sustained performance beyond leading human experts across diverse domains, reliable adaptation to unfamiliar problems, and meaningful capability beyond narrow benchmark specialization. Autonomy, independent scientific discovery, and the ability to improve performance across different environments may contribute, but no single characteristic is sufficient. Nor does superintelligence necessarily require consciousness, subjective experience, or unrestricted autonomy. This distinction will become consequential as businesses, investors, policymakers, and consumers evaluate claims about advanced intelligence.
Beyond Classical Computing: Could Quantum Technology Accelerate Superintelligence?
Looking forward, another possibility deserves consideration. Consider the relationship between advanced intelligence and quantum computing. The relationship runs in both directions, and each deserves attention. Contemporary AI depends primarily on classical computing architectures, including CPUs, GPUs, and specialized accelerators such as TPUs. Quantum computing introduces fundamentally different computational mechanisms based on quantum-mechanical principles.
Artificial Intelligence Supporting Quantum Computing
One opportunity already being investigated is applying machine learning and advanced AI to the development and operation of quantum computing systems. A review published in Nature Communications, Artificial Intelligence for Quantum Computing (Alexeev et al., December 2, 2025), examines AI applications involving quantum device control, calibration, error correction, algorithm development, and optimization. These capabilities could improve the performance, reliability, and engineering efficiency of quantum systems. Agentic AI could further support researchers by coordinating experiments, analyzing results, proposing adjustments, and automating appropriately controlled laboratory workflows. In this scenario, capable classical AI helps advance quantum computing.
Quantum Computing Supporting Advanced Intelligence
The reverse relationship. Future fault-tolerant quantum computers may provide computational advantages for particular problems relevant to scientific discovery, chemistry, materials science, optimization, or selected machine-learning methods. If those capabilities mature, advanced AI systems could potentially use quantum processors as specialized resources within broader computational architectures.
Imagine an interconnected research environment in which AI agents formulate scientific hypotheses, classical systems analyze experimental information, quantum processors perform specialized simulations, and additional agents interpret results and propose subsequent investigations. The collective capability could become more powerful. Quantum computing does not, however, automatically accelerate all AI workloads, and greater computational performance does not inherently establish superintelligence.
An August 4, 2026, Gartner assessment reported that no peer-reviewed result had demonstrated quantum advantage on a production AI workload, forecast that no enterprise AI workload at scale would run on quantum hardware through 2028, and predicted that fault-tolerant quantum computing would remain in the research-and-development phase for AI through 2030. Likewise, a position paper at the 2026 International Conference on Machine Learning, Quantum Deep Learning Still Needs a Quantum Leap, identifies significant barriers to practical quantum acceleration of deep learning, including algorithmic limitations, hardware speed, and the underdeveloped state of quantum random-access memory.
These findings do not eliminate the long-term possibilities. They establish a boundary between promising research and capabilities available for commercial deployment. The intersection of advanced AI and quantum computing deserves sustained scientific investment, but the timeline and ultimate impact remain uncertain. A quantum-enabled system would still need to demonstrate extraordinary cognitive performance before it could reasonably be characterized as artificial superintelligence.
The Enterprise Implications: When AI Becomes an Interconnected Operating Model
The DeepMind Institute's vision has immediate relevance even without the arrival of AGI or superintelligence. Organizations are already exploring architectures where multiple agents interact with enterprise applications, information repositories, communication platforms, and business workflows. Gartner forecasts that at least 15 percent of day-to-day work decisions will be made autonomously by agentic AI by 2028, up from essentially zero in 2024, and that one-third of enterprise software applications will include agentic AI by the same year.
As intelligence becomes distributed, accountability must become more deliberate. Traditional enterprise governance assigns responsibilities through business functions, management authority, policies, procedures, systems of record, and internal controls. Interconnected AI systems complicate those boundaries because an outcome may depend on multiple models, external services, automation tools, data sources, and human participants.
An enterprise pursuing this future should consider the following, each grounded in NIST guidance.
Establish Ownership and Decision Authority
Every consequential AI workflow requires an accountable business owner. The NIST AI Risk Management Framework (AI RMF 1.0) makes this explicit: GOVERN 2.1 calls for documented roles, responsibilities, and lines of communication for AI risk, and GOVERN 1.3 calls for organizational leadership to set risk tolerance. The NIST Cybersecurity Framework (CSF) 2.0 reinforces it through its GOVERN function, specifically the Roles, Responsibilities, and Authorities category (GV.RR). Organizations should define which decisions agents may recommend, which actions they may execute independently, and which activities require authorization from designated individuals. An AI agent may execute a transaction, but accountability must remain with the responsible organization and its authorized leadership.
Engineer Trust Into the Architecture
Enterprise agents should operate as identifiable security principals under the controls organizations already apply to human users and services. NIST SP 800-53 Rev. 5 supplies the vocabulary: Account Management (AC-2), Least Privilege (AC-6), Identification and Authentication for organizational users and for services (IA-2, IA-9), Boundary Protection (SC-7), and Event Logging with sufficient audit-record content to reconstruct events (AU-2, AU-3). NIST SP 800-207, Zero Trust Architecture, provides the model for independently enforced authorization at every request, and NIST SP 800-63-4, the Digital Identity Guidelines finalized in July 2025, defines the assurance levels. For nonhuman identity specifically, NIST's National Cybersecurity Center of Excellence published a concept paper on software and AI agent identity and authorization in February 2026, and NIST's Center for AI Standards and Innovation launched an AI Agent Standards Initiative the same month. Security controls must address the entire workflow, not the model alone.
Validate Collective Intelligence
Testing one model does not establish the reliability of a multi-agent system. AI RMF MEASURE 2.5 requires that the AI system be demonstrated valid and reliable, and MEASURE 2.7 that security and resilience be evaluated and documented. Organizations must evaluate what happens when agents exchange inaccurate information, disagree, reinforce one another's mistakes, encounter adversarial instructions, or lose access to critical services. NIST AI 100-2e2025, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations, catalogs the attack classes to test against. Testing should include cascading failures, unauthorized actions, correlated errors, recovery procedures (SP 800-53 CP-10), incident handling (IR-4), and effective human intervention. Independent assessment (CA-2) and developer testing and evaluation (SA-11) are important when multiple agents rely on shared models or information sources.
Connect Autonomy to Measurable Business Value
Greater autonomy is not inherently a better investment. Gartner predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value, or inadequate risk controls. AI RMF MAP 1.4 asks organizations to define the business value or context of business use, and MANAGE 1.1 asks whether the system should be developed or deployed at all, given its assessed risks and value. The CFO, CIO, CTO, CISO, and business leadership should assess the incremental value of agent orchestration against its total cost of ownership, operational complexity, security exposure, and performance requirements. Relevant measures include decision quality, cycle time, cost per completed workflow, human rework, error rates, service availability, control exceptions, and realized financial benefits. The objective is to improve sustainable business outcomes within acceptable risk.
Prepare Governance for Capable Systems
The NIST AI Risk Management Framework's Govern, Map, Measure, and Manage functions provide the operating model. Organizations can apply them to agent inventories (SP 800-53 CM-8; CSF 2.0 ID.AM), lifecycle management (MAP 1.6), post-deployment monitoring (MANAGE 4.1), and accountability. NIST AI 600-1, the Generative AI Profile, adds twelve risk categories specific to generative systems. Two NIST efforts in progress extend this to agents directly: the Cyber AI Profile (NIST IR 8596, preliminary draft, December 2025), a CSF 2.0 community profile organized around securing AI components, AI-enabled defense, and thwarting AI-enabled attacks; and the SP 800-53 Control Overlays for Securing AI Systems (COSAiS), which include single-agent and multi-agent overlays under development. Where agents process personal data, the NIST Privacy Framework and the SP 800-53 PT family (Personally Identifiable Information Processing and Transparency) apply.
These frameworks are voluntary guidance. Specific legal and regulatory obligations depend on the organization's activities, sector, jurisdiction, and applicable rules. As AI capabilities advance, governance must evolve with them. Systems with broader authority, greater autonomy, or access to higher-consequence decisions require stronger assurance and operational controls.
The Future of Intelligence: More Than a Technology Milestone
The DeepMind Institute's essay is valuable because it challenges a fundamental assumption that the future of advanced intelligence must be determined by the capabilities of an individual machine. Its alternative is a future of interconnected human and artificial intelligence, supported by shared knowledge, specialized capabilities, and institutions designed to coordinate complex activities. Whether this architecture contributes to AGI or artificial superintelligence remains an open scientific question.
Quantum computing could eventually expand the computational possibilities available to such systems, while continued advances in classical computing, algorithms, and agent orchestration may produce significant progress independently. The emergence of superintelligence would represent a technological achievement requiring compelling evidence.
For enterprise leadership, the strategic opportunity is already visible. Organizations can begin building more capable, secure, and economically sustainable systems of human–AI collaboration without assuming that AGI or superintelligence has arrived. When it does, the question is whether humanity can develop institutions capable of responsibly governing the powerful intelligence that emerges from the interaction of people, machines, and technologies.
The future of intelligence may be collective. Whether that collective intelligence creates business and societal value will depend on how deliberately we design, measure, and govern it.
Sources and Further Reading
Primary source
Bratton, B., Agüera y Arcas, B., and Manyika, J. Artificial Symbiotic Intelligence: Agents, AGI and the Orchestration of Many Minds. DeepMind Institute, September 24, 2026.
Supporting research and authoritative references
Morris, M. R., et al. Levels of AGI for Operationalizing Progress on the Path to AGI. Google DeepMind, ICML 2024.
The White House. Executive Order 14434: Inaugurating the Era of Super Intelligence. September 29, 2026.
Bengio, Y., et al. International AI Safety Report 2026. February 3, 2026.
Alexeev, Y., et al. Artificial Intelligence for Quantum Computing. Nature Communications, December 2, 2025.
Gartner. Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027. June 25, 2025.
Gartner. Gartner Predicts Enterprise AI Workloads at Scale Will Not Run on Quantum Hardware Through 2028. August 4, 2026.
Gundlach, H., Kukina, H., Lynch, J., and Thompson, N. Position: Quantum Deep Learning Still Needs a Quantum Leap. Proceedings of the 43rd International Conference on Machine Learning, 2026.
National Institute of Standards and Technology. AI Risk Management Framework Core.
NIST frameworks and publications
NIST AI 100-1, Artificial Intelligence Risk Management Framework (AI RMF 1.0), January 2023.
NIST AI 600-1, AI RMF: Generative Artificial Intelligence Profile, July 2024.
NIST AI 100-2e2025, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations, March 2025.
NIST CSWP 29, The NIST Cybersecurity Framework (CSF) 2.0, February 2024.
NIST SP 800-53 Rev. 5, Security and Privacy Controls for Information Systems and Organizations.
NIST SP 800-207, Zero Trust Architecture, August 2020.
NIST SP 800-63-4, Digital Identity Guidelines, July 2025.
NIST IR 8596 (Initial Preliminary Draft), Cybersecurity Framework Profile for Artificial Intelligence (Cyber AI Profile), December 16, 2025.
NIST, SP 800-53 Control Overlays for Securing AI Systems (COSAiS) project.
NIST Privacy Framework, Version 1.0 (January 2020) and Version 1.1 Initial Public Draft (April 2025).
Editorial disclosure: This article represents independent research commentary by DEW Diligence, LLC. DEW Diligence is not affiliated with or endorsed by Google, Google DeepMind, the DeepMind Institute, Gartner, the International AI Safety Report, NIST, MIT, NVIDIA, or the other organizations referenced. Statistics, forecasts, and research findings are reported as published by their sources and have not been independently reproduced by DEW Diligence; Gartner figures are forecasts, not measured outcomes. The cognitive crossover point, agent institutions, and quantum-enabled research environments are hypothetical or conceptual scenarios and should not be read as documented capabilities. “Artificial Symbiotic Intelligence” is terminology used by the essay’s authors, Bratton, Agüeray Arcas, and Manyika. “Super Intelligence” and “SI” are terms set by Executive Order 14434, which defines them by the existing statutory definition of artificial intelligence. “Artificial superintelligence” is used by DEW Diligence in its general research sense: AI that substantially exceeds the most capable humans across virtually all domains. It does not refer to the executive order’s usage. Mappings of enterprise recommendations to NIST functions, categories, and controls are DEW Diligence’s interpretation, not official NIST crosswalks. Publications identified as drafts were in draft status as of October 23, 2026.
Want more on this topic? Contact us for a consultation, and review our services for fractional & virtual CIO, CISO, and CTO executive consulting.
David E. Williams, CISM, GSTRT, GSLC
Founding Principal, DEW Diligence, LLC
© 2026 DEW Diligence, LLC.
Comments