ibm research

IBM Research – India

Established in 1998, IBM Research – India is a hub for science and technology innovation. With locations in Gurgaon and Bengaluru, the lab is focused on a wide array of projects across hybrid cloud, AI, climate & sustainability and quantum computing applications.

Meet the team

Celebrating 25 years of innovation at IBM Research India

Research areas, trustworthy ai.

As enterprises aim to build better, smarter and automated ML models, teams must also consider ways to validate and manage data before building and deploying applications in the real world. Doing so ensures that the model is based on data that is trusted and built using principles that can be explained. At IBM Research – India, we are working to systematically understand the challenges enterprises face in data quality assessment to help build trust into the model before it is fed to a ML pipeline.

Learn more about Trustworthy AI  

Testing for AI

• Data and AI Security

Data Quality in AI

Metadata Management

• AI for Code

We’re building AI models, designing program analysis techniques, optimizing software engineering, and contributing to the field of AI for Code. We’re investigating fundamental tasks such as software understanding, source code modelling, multi-code view representation learning, as well as industrial use cases like application modernization, code translation, and question and answering.

Learn more about AI for Code  

Functional Analysis of Code

Code Representation Learning

Code Translation

• Conversational AI

We’re working on ways to model user intents using machine learning methods and rule-based dialog flow to enhance the field of conversational AI. We’re building on frameworks such as Watson Assistant, DialogFlow, and lex to help ensure deep learning models are applicable for use in enterprise settings. At IBM Research – India, we are working on various approaches that could help bring the deep learning based models closer to their use in enterprise settings.

Learn more about Conversational AI  

Continuous Improvement of Chatbots

Foundation Models for Conversation

• Foundation Models

We are witnessing the emergence of highly scaled decentralized applications, like central bank digital currencies (CBDC), decentralized finance (DeFi), and non-fungible tokens (NFT) that have stringent privacy requirements and require seamless interoperability across applications. We're working on building the next generation of tools that answer the needs of these applications with privacy-preserving and interoperable digital assets.

Learn more about Security  

Foundations and Practice of Distributed Ledger Interoperability

• Quantum Safe
• Trusted Decentralized Systems

Hybrid Cloud

A hybrid cloud environment must be reliable for organizations to adopt it at scale for mission critical needs. Critical functions span across performance, security, compliance, cost of ownership, resilience and elasticity. As businesses become increasingly agile to meet dynamic consumer and market needs, these specifications are evolving at a rapid pace. The reliability of a hybrid cloud environment must support continuous and agile modernization at all levels of the IT stack.

Learn more about Hybrid Cloud  

Application Aware Networking

• Hybrid Cloud Platform

Multi-cloud Compliance as Code Automation

Sustainable Computing

Platform modernization with move2kub.

• Application Modernization

Climate and Sustainability

To respond to the global climate emergency, we're using AI and hybrid cloud to accelerate discovery of climate mitigation solutions. We're improving carbon emission performance through optimization and capture, and preparing enterprises for the impact of climate change. Enterprises are under significant pressure from investors, consumers, and policymakers to act on climate change mitigation by disclosing their GHG emissions and committing to the reduction of emissions from their industrial activities including operations, manufacturing, logistics, and supply chains.

Learn more about Climate and Sustainability  

Carbon Performance Engine

Greenhouse Gas Monitoring

Renewables forecasting.

• AI for Business Automation

At IBM Research – India, we are developing new AI powered solutions for resource planning and decision optimization in the context of manufacturing and supply chain with the goal of improving process efficiencies and reducing cost. We are developing advanced analytics solutions that can prescribe set points to control operations of various assets using real-time machine data. By analyzing the business and IT operations data, process executions can be optimized by identifying automation opportunities and exploring different automation strategies that include Robotic Process Automation, and intelligent agents for automation.

Learn more about AI for Business Automation  

AI for Business Process

Resource and Operations Management for Industry 4.0

Location details.

IBM India Pvt. Ltd IBM Research-India Manyata Embassy Business Park 8th Floor, G2 Outer Ring Road Rachenahalli & Nagawara Villages Bengaluru, KA 560045

Join our team

We’re always looking for people excited to make a difference. See our open positions and help us invent what’s next.

• See open positions in India

Imagine a world where machines aren’t confined to pre-programmed tasks but operate with human-like autonomy and competence. A world where computer minds pilot self-driving cars, delve into complex scientific research, provide personalized customer service and even explore the unknown.

This is the potential of artificial general intelligence (AGI), a hypothetical technology that may be poised to revolutionize nearly every aspect of human life and work. While AGI remains theoretical, organizations can take proactive steps to prepare for its arrival by building a robust data infrastructure and fostering a collaborative environment where humans and AI work together seamlessly.

AGI, sometimes referred to as strong AI , is the science-fiction version of artificial intelligence (AI), where artificial machine intelligence achieves human-level learning, perception and cognitive flexibility. But, unlike humans, AGIs don’t experience fatigue or have biological needs and can constantly learn and process information at unimaginable speeds. The prospect of developing synthetic minds that can learn and solve complex problems promises to revolutionize and disrupt many industries as machine intelligence continues to assume tasks once thought the exclusive purview of human intelligence and cognitive abilities.

Imagine a self-driving car piloted by an AGI. It cannot only pick up a passenger from the airport and navigate unfamiliar roads but also adapt its conversation in real time. It might answer questions about local culture and geography, even personalizing them based on the passenger’s interests. It might suggest a restaurant based on preferences and current popularity. If a passenger has ridden with it before, the AGI can use past conversations to personalize the experience further, even recommending things they enjoyed on a previous trip.

AI systems like LaMDA and GPT-3 excel at generating human-quality text, accomplishing specific tasks, translating languages as needed, and creating different kinds of creative content. While these large language model (LLM) technologies might seem like it sometimes, it’s important to understand that they are not the thinking machines promised by science fiction. 

Achieving these feats is accomplished through a combination of sophisticated algorithms, natural language processing (NLP) and computer science principles. LLMs like ChatGPT are trained on massive amounts of text data, allowing them to recognize patterns and statistical relationships within language. NLP techniques help them parse the nuances of human language, including grammar, syntax and context. By using complex AI algorithms and computer science methods, these AI systems can then generate human-like text, translate languages with impressive accuracy, and produce creative content that mimics different styles.

Today’s AI, including generative AI (gen AI), is often called narrow AI and it excels at sifting through massive data sets to identify patterns, apply automation to workflows and generate human-quality text. However, these systems lack genuine understanding and can’t adapt to situations outside their training. This gap highlights the vast difference between current AI and the potential of AGI.

While the progress is exciting, the leap from weak AI to true AGI is a significant challenge. Researchers are actively exploring artificial consciousness, general problem-solving and common-sense reasoning within machines. While the timeline for developing a true AGI remains uncertain, an organization can prepare its technological infrastructure to handle future advancement by building a solid data-first infrastructure today. 

How can organizations prepare for AGI?

The theoretical nature of AGI makes it challenging to pinpoint the exact tech stack organizations need. However, if AGI development uses similar building blocks as narrow AI, some existing tools and technologies will likely be crucial for adoption.

The exact nature of general intelligence in AGI remains a topic of debate among AI researchers. Some, like Goertzel and Pennachin , suggest that AGI would possess self-understanding and self-control. Microsoft and OpenAI have claimed that GPT-4’s capabilities are strikingly close to human-level performance. Most experts categorize it as a powerful, but narrow AI model.

Current AI advancements demonstrate impressive capabilities in specific areas. Self-driving cars excel at navigating roads and supercomputers like  IBM Watson ® can analyze vast amounts of data. Regardless, these are examples of narrow AI. These systems excel within their specific domains but lack the general problem-solving skills envisioned for AGI.

Regardless, given the wide range of predictions for AGI’s arrival, anywhere from 2030 to 2050 and beyond, it’s crucial to manage expectations and begin by using the value of current AI applications. While leaders have some reservations about the benefits of current AI, organizations are actively investing in gen AI deployment, significantly increasing budgets, expanding use cases, and transitioning projects from experimentation to production.

According to  Andreessen Horowitz (link resides outside IBM.com ) , in 2023, the average spend on foundation model application programming interfaces (APIs), self-hosting and fine-tuning models across surveyed companies reached USD 7 million. Nearly all respondents reported promising early results from gen AI experiments and planned to increase their spending in 2024 to support production workloads. Interestingly, 2024 is seeing a shift in funding through software line items, with fewer leaders allocating budgets from innovation funds, hinting that gen AI is fast becoming an essential technology. 

On a smaller scale, some organizations are reallocating gen AI budgets towards headcount savings, particularly in customer service. One organization reported saving approximately USD 6 per call served by its LLM-powered customer service system, translating to a 90% cost reduction, a significant justification for increased gen AI investment.

Beyond cost savings, organizations seek tangible ways to measure gen AI’s return on investment (ROI), focusing on factors like revenue generation, cost savings, efficiency gains and accuracy improvements, depending on the use case. A key trend is the adoption of multiple models in production. This multi-model approach uses multiple AI models together to combine their strengths and improve the overall output. This approach also serves to tailor solutions to specific use cases, avoid vendor lock-in and capitalize on rapid advancement in the field.

46% of survey respondents in 2024 showed a preference for open source models. While cost wasn’t the primary driver, it reflects a growing belief that the value generated by gen AI outweighs the price tag. It illustrates that the executive mindset increasingly recognizes that getting an accurate answer is worth the money. 

Enterprises remain interested in customizing models, but with the rise of high-quality open source models, most opt not to train LLMs from scratch. Instead, they’re using retrieval augmented generation or fine-tuning open source models for their specific needs.

The majority (72%) of enterprises that use APIs for model access use models hosted on their cloud service providers. Also, applications that don’t just rely on an LLM for text generation but integrate it with other technologies to create a complete solution and significantly rethink enterprise workflows and proprietary data use are seeing strong performance in the market.

Deloitte  (link resides outside IBM.com) explored the value of output being created by gen AI among more than 2,800 business leaders. Here are some areas where organizations are seeing a ROI:

• Text (83%) : Gen AI assists with automating tasks like report writing, document summarization and marketing copy generation.
• Code (62%) : Gen AI helps developers write code more efficiently and with fewer errors.
• Audio (56%) : Gen AI call centers with realistic audio assist customers and employees.
• Image (55%):  Gen AI can simulate how a product might look in a customer’s home or reconstruct an accident scene to assess insurance claims and liability.
• Other potential areas : Video generation (36%) and 3D model generation (26%) can create marketing materials, virtual renderings and product mockups.

The skills gap in gen AI development is a significant hurdle. Startups offering tools that simplify in-house gen AI development will likely see faster adoption due to the difficulty of acquiring the right talent within enterprises.

While AGI promises machine autonomy far beyond gen AI, even the most advanced systems still require human expertise to function effectively. Building an in-house team with AI,  deep learning ,  machine learning (ML) and data science skills is a strategic move. Most importantly, no matter the strength of AI (weak or strong), data scientists, AI engineers, computer scientists and ML specialists are essential for developing and deploying these systems.

These use areas are sure to evolve as AI technology progresses. However, by focusing on these core areas, organizations can position themselves to use the power of AI advancements as they arrive.

Improving AI to reach AGI

While AI has made significant strides in recent years, achieving true AGI, machines with human-level intelligence, still require overcoming significant hurdles. Here are 7 critical skills that current AI struggles with and AGI would need to master:

• Visual perception:  While computer vision has overcome significant hurdles in facial recognition and object detection, it falls far short of human capabilities. Current AI systems struggle with context, color and understanding how to react to partially hidden objects. 
• Audio perception:  AI has made progress in speech recognition but cannot reliably understand accents, sarcasm and other emotional speech tones. It also has difficulty filtering out unimportant background noise and is challenged to understand non-verbal expressions, like sighs, laughs or changes in volume. 
• Fine motor skills:  It’s conceivable for AGI software to pair with robotics hardware. In that instance, the AGI would require the ability to handle fragile objects, manipulate tools in real-world settings and be able to adapt to new physical tasks quickly. 
• Problem-solving:  Weak AI excels at solving specific, well-defined problems, but AGI would need to solve problems the way a human would, with reasoning and critical thinking. The AGI would need to handle uncertainty and make decisions with incomplete information. 
• Navigation : Self-driving cars showcase impressive abilities, but human-like navigation requires immediate adaptation to complex environments. Humans can easily navigate crowded streets, uneven terrain and changing environments. 
• Creativity : While AI can generate creative text formats to some degree, true creativity involves originality and novelty. Creating new ideas, concepts or solutions is a hallmark of human creativity.
• Social and emotional engagement:  Human intelligence is deeply intertwined with our social and emotional abilities. AGI would need to recognize and understand emotions, including interpreting facial expressions, body language and tone of voice. To respond appropriately to emotions, AGI needs to adjust its communication and behavior based on the emotional state of others.

AGI examples

However, once theoretical AGI achieves the above to become actual AGI, its potential applications are vast. Here are some examples of how AGI technology might revolutionize various industries:

Customer service

Imagine an AGI-powered customer service system. It would access vast customer data and combine it with real-time analytics for efficient and personalized service. By creating a comprehensive customer profile (demographics, past experiences, needs and buying habits), AGI might anticipate problems, tailor responses, suggest solutions and even predict follow-up questions.

Example : Imagine the best customer service experience that you’ve ever had. AGI can offer this through a perception system that anticipates potential issues, uses tone analysis to better understand the customer’s mood, and possesses a keen memory that can recall the most specific case-resolving minutiae. By understanding the subtleties of human language, AGI can have meaningful conversations, tackle complex issues and navigate troubleshooting steps. Also, its emotional intelligence allows it to adapt communication to be empathetic and supportive, creating a more positive interaction for the customer.

Coding intelligence

Beyond code analysis, AGI grasps the logic and purpose of existing codebases, suggesting improvements and generating new code based on human specifications. AGI can boost productivity by providing a hardcoded understanding of architecture, dependencies and change history.

Example:  While   building an e-commerce feature, a programmer tells AGI, “I need a function to calculate shipping costs based on location, weight and method.” AGI analyzes relevant code, generates a draft function with comments explaining its logic and allows the programmer to review, optimize and integrate it.

Navigation, exploration and autonomous systems

Current self-driving cars and autonomous systems rely heavily on pre-programmed maps and sensors. AGI wouldn’t just perceive its surroundings; it would understand them. It might analyze real-time data from cameras, LiDAR and other sensors to identify objects, assess risks and anticipate environmental changes like sudden weather events or unexpected obstacles. Unlike current systems with limited response options, AGI might make complex decisions in real time.

It might consider multiple factors like traffic flow, weather conditions and even potential hazards beyond the immediate sensor range. AGI-powered systems wouldn’t be limited to pre-programmed routes. They might learn from experience, adapt to new situations, and even explore uncharted territories. Imagine autonomous exploration vehicles navigating complex cave systems or drones assisting in search and rescue missions in constantly changing environments.

Example:  An AGI-powered self-driving car encounters an unexpected traffic jam on its usual route. Instead of rigidly following pre-programmed instructions, the AGI analyzes real-time traffic data from other connected vehicles. It then identifies alternative routes, considering factors like distance, estimated travel time and potential hazards like construction zones. Finally, it chooses the most efficient and safest route in real time, keeping passengers informed and comfortable throughout the journey.

The vast amount of medical data generated today remains largely untapped. AGI might analyze medical images, patient records, and genetic data to identify subtle patterns that might escape human attention. By analyzing historical data and medical trends, AGI might predict a patient’s specific potential risk of developing certain diseases. AGI might also analyze a patient’s genetic makeup and medical history to tailor treatment plans. This personalized approach might lead to more effective therapies with fewer side effects.

Example : A patient visits a doctor with concerning symptoms. The doctor uploads the patient’s medical history and recent test results to an AGI-powered medical analysis system. The AGI analyzes the data and identifies a rare genetic mutation linked to a specific disease. This information is crucial for the doctor, as it allows for a more targeted diagnosis and personalized treatment plan, potentially improving patient outcomes.

Imagine an AGI tutor who doesn’t present information but personalizes the learning journey. AGI might analyze a student’s performance, learning style and knowledge gaps to create a customized learning path. It wouldn’t treat all students the same. AGI might adjust the pace and difficulty of the material in real time based on the student’s understanding. Struggling with a concept? AGI provides other explanations and examples. Mastering a topic? It can introduce more challenging material. AGI might go beyond lectures and textbooks. It might create interactive simulations, personalized exercises and even gamified learning experiences to keep students engaged and motivated.

Example:  A student is struggling with a complex math concept. The AGI tutor identifies the difficulty and adapts its approach. Instead of a dry lecture, it presents the concept visually with interactive simulations and breaks it down into smaller, more manageable steps. The student practices with personalized exercises that cater to their specific knowledge gaps and the AGI provides feedback and encouragement throughout the process.

Manufacturing and supply chain management

AGI might revolutionize manufacturing by optimizing every step of the process. By analyzing vast amounts of data from sensors throughout the production line to identify bottlenecks, AGI might recommend adjustments to machine settings and optimize production schedules in real time for maximum efficiency. Analyzing historical data and sensor readings might help AGI predict equipment failures before they happen. This proactive approach would prevent costly downtime and help ensure smooth operation. With AGI managing complex logistics networks in real time, it can optimize delivery routes, predict potential delays and adjust inventory levels to help ensure just-in-time delivery, minimizing waste and storage costs.

Example:  Imagine an AGI system monitors a factory assembly line. It detects a slight vibration in a critical machine, indicating potential wear and tear. AGI analyzes historical data and predicts a possible failure within the next 24 hours. It alerts maintenance personnel, who can proactively address the issue before it disrupts production. This allows for a smooth and efficient operation, avoiding costly downtime.

Financial services

AGI might revolutionize financial analysis by going beyond traditional methods. AGI could analyze vast data sets encompassing financial news, social media sentiment and even satellite imagery to identify complex market trends and potential disruptions that might go unnoticed by human analysts. There are startups and financial institutions already working on and using limited versions of such technologies.

By being able to process vast amounts of historical data, AGI might create even more accurate financial models to assess risk and make more informed investment decisions. AGI might develop and run complex trading algorithms that factor in market data, real-time news and social media sentiment. However, human oversight would remain crucial for final decision-making and ethical considerations.

Example:  A hedge fund uses an AGI system to analyze financial markets. AGI detects a subtle shift in social media sentiment toward a specific industry and identifies a potential downturn. It analyzes historical data and news articles, confirming a possible market correction. Armed with this information, the fund manager can make informed decisions to adjust their portfolio and mitigate risk.

Research and development

AGI might analyze vast data sets and scientific literature, formulate new hypotheses and design experiments at an unprecedented scale, accelerating scientific breakthroughs across various fields. Imagine a scientific partner that can examine data and generate groundbreaking ideas by analyzing vast scientific data sets and literature to identify subtle patterns and connections that might escape human researchers. This might lead to the formulation of entirely new hypotheses and research avenues.

By simulating complex systems and analyzing vast amounts of data, AGI could design sophisticated experiments at an unprecedented scale. This would allow scientists to test hypotheses more efficiently and explore previously unimaginable research frontiers. AGI might work tirelessly, helping researchers sift through data, manage complex simulations and suggest new research directions. This collaboration would significantly accelerate the pace of scientific breakthroughs.

Example:  A team of astrophysicists is researching the formation of galaxies in the early universe. AGI analyzes vast data sets from telescopes and simulations. It identifies a previously overlooked correlation between the distribution of dark matter and the formation of star clusters. Based on this, AGI proposes a new hypothesis about galaxy formation and suggests a series of innovative simulations to test its validity. This newfound knowledge paves the way for a deeper understanding of the universe’s origins.

What are the types of AGI?

AGI would be an impactful technology that would forever transform how industries like healthcare or manufacturing conduct business. Large tech companies and research labs are pouring resources into its development, with various schools of thought tackling the challenge of achieving true human-level intelligence in machines. Here are a few primary areas of exploration:

• Symbolic AI:  This approach focuses on building systems that manipulate symbols and logic to represent knowledge and reasoning. It aims to create a system that can understand and solve problems by following rules, similar to how humans use logic.
• Connectionist AI (artificial neural networks):  This approach is inspired by the structure and function of the human brain. It involves building artificial neural networks with interconnected nodes to learn and process information based on vast data.
• Artificial consciousness:  This field delves into imbuing machines with subjective experience and self-awareness. It’s a highly theoretical concept but might be a key component of true intelligence.
• Whole brain emulation:  This ambitious approach aims to create a detailed computer simulation of a biological brain. The theory is that consciousness and intelligence might emerge within the simulation by copying the human brain’s structure and function.
• Embodied AI and embodied cognition:  This approach focuses on the role of an agent’s physical body and its interaction with the environment in shaping intelligence. The idea is that true intelligence requires an agent to experience and learn from the world through a physical body.

The AGI research field is constantly evolving. These are just some of the approaches that have been explored. Likely, a combination of these techniques or entirely new approaches will ultimately lead to the realization of AGI.

Operationalizing AI is the future of business

AGI might be science fiction for now, but organizations can get ready for the future by building an AI strategy for the business on one collaborative AI and data platform, IBM watsonx™. Train, validate, tune and deploy AI models to help you scale and accelerate the impact of AI with trusted data across your business.

More from Artificial intelligence

Breaking boundaries: postgresql 16 is now available on ibm cloud.

2 min read - PostgreSQL Version 16 is now available on IBM Cloud®. The latest version of IBM Cloud® Databases for PostgreSQL includes critical features that offer unmatched reliability and scalability for clients' data storage needs. Advanced features like enhanced parallel query performance and accelerated indexing provide significant performance gains to your applications. With IBM Cloud's robust ecosystem of extensions and plugins, PostgreSQL v16 empowers customers to build and manage sophisticated, high-performance applications easily while continuing to improve data management at scale. In addition, customers can continue to access powerful IBM…

IBM and TechD partner to securely share data and power insights with gen AI

3 min read - As technology expands, at TechD, we know that the quality of generative AI (gen AI) depends on accurate data sourcing. A reliable and trustworthy data source is essential for sharing information across departments. Through the implementation of generative AI we are able to expand our knowledge to many individuals easily, quickly and efficiently becoming a resource. In today's rapidly evolving digital world, immediate responses are crucial for delivering outstanding user experiences. Our partnership with IBM facilitates the delivery of scalable…

4 ways generative AI addresses manufacturing challenges

4 min read - The manufacturing industry is in an unenviable position. Facing a constant onslaught of cost pressures, supply chain volatility and disruptive technologies like 3D printing and IoT. The industry must continually optimize process, improve efficiency, and improve overall equipment effectiveness. At the same time, there is this huge sustainability and energy transition wave. Manufacturers are being called to reduce their carbon footprint, adopt circular economy practices and become more eco-friendly in general. And manufacturers face pressure to constantly innovate while ensuring…

IBM Newsletters