Technology giant IBM said artificial intelligence (AI) will enter its most promising innovation era starting in 2023 as a result of three key subareas of this technology.
In recent decades, the field of AI has experienced tremendous scientific advancements — from vast improvements in processing power and computational efficiency to the emergence of more flexible and reusable AI systems that succeed in a broader range of application and domain areas.
According to IBM, the following three key sub-areas of AI will result in more innovations:
- Creating reusable AI through foundation models
While many new AI systems are helping solve all sorts of real-world problems, creating and deploying each new system often requires a considerable amount of time and resources.
For each new application, you need to ensure that there’s a large, well-labeled dataset for the specific task you want to tackle. Not to mention, training one large natural-language processing model, for example, has roughly the same carbon footprint as running five cars over their lifetime.
The next wave in AI, called foundation models, will replace the task-specific models that have dominated the AI landscape to date by introducing systems that are trained on a broad set of unlabeled data and can be used and re-used for different tasks, with minimal fine-tuning.
IBM Research is well-underway in creating models and techniques that will allow foundation models to not only operate more efficiently but allow them to address a variety of domains beyond language. These include: automating code and chemistry, and making it easier to set up and run foundation models for AI workloads.
- Synthetic Data
Trained on words and videos from sites like Wikipedia and YouTube, deep learning models learn to make predictions and decisions based on patterns extracted from billions of real-world examples.
Yet, despite the progress, real data (e.g., financial information, healthcare records and consumer analysis) come with significant hurdles such as privacy, ethics and copyright laws. Data these days is also often expensive and comes with baked-in vulnerabilities like bias.
Synthetic data offer a work-around because they are computer-generated examples that can augment or replace real data to speed up the training of AI models, protect sensitive data, improve accuracy, or find and mitigate bias and security weaknesses.
IBM Research is exploring a number of ways in which we can use synthetic data to create more trustworthy and cost-effective AI systems: from using the ThreeDWorld virtual environment to create fake images for pretraining a vision model called Task2Sim, which notably outperformed a model trained on real images, to
- Commonsense AI
Today’s AI systems are quickly evolving to do more across business and society. Yet, challenges remain to create systems that can demonstrate agility, flexibility and real understanding of the topics and problems they’re asked to solve.
In 2023, the field of AI WILL BE flooded with techniques designed to take this problem head-on and create systems that exhibit increased commonsense and understanding, the way humans know it.
IBM IS continuing to explore new ways that can help AI exhibit human-like commonsense. For example, the MIT-IBM Watson AI Lab in collaboration with MIT and Harvard researchers created AGENT: Action-Goal-Efficiency-coNstraint-uTility. AGENT is a large-scale dataset of 3D animations inspired by experiments that study cognitive development in kids.
The animations depict someone interacting with different objects under different physical constraints. After testing and validation, this benchmark is shown to be able to evaluate the core psychological reasoning ability of an AI model. This means it can actually give a sense of social awareness and could interact with humans in real-world settings.
