Some people argue that it’s becoming increasingly clear that our current approaches to building AI models are not sustainable, nor are they adequately addressing the ethical concerns raised by the technology. Deep learning, the dominant approach to AI research over the past decade, has revolutionized fields from image and speech recognition to natural language processing. But its limitations—such as the enormous amounts of data, time, and energy required to produce useful outcomes, and the lack of transparency and explainability—are becoming increasingly apparent.
To address these challenges, a new generation of AI researchers and startups, including Symbolica AI, are proposing a promising alternative: symbolic AI. This approach, which involves encoding the underlying structure of data instead of approximating insights from enormous data sets, has several significant advantages over deep learning.
Symbolic AI, also known as structured AI, has been used in applications like text editing software for decades. It solves tasks by defining symbol-manipulating rule sets dedicated to particular jobs, making it more efficient and cost-effective to train and run than deep learning models. Symbolic AI is also more reliable, transparent, and accountable. It can reason through complex scenarios using a set of rules and explain its reasoning in a clear and transparent way, which is crucial in safety-critical industries where explaining the reasoning behind AI decisions is not just a nice-to-have, but a regulatory requirement.
Symbolic AI models also offer a new way to approach AI ethics and sustainability. By focusing on symbolic AI, researchers and organizations can build AI systems that require less data and computational resources, reducing the environmental impact of AI research and development. Symbolic AI offers a more accessible approach to AI technology, allowing smaller organizations and individuals to develop and deploy AI systems without requiring vast resources.
Symbolica AI, a startup founded by ex-Tesla engineer George Morgan, is spearheading this movement. The company is developing a toolkit for creating symbolic AI models and pre-trained models for specific tasks, including generating code and proving mathematical theorems. Morgan claims that symbolic AI models can produce domain-tailored structured reasoning capabilities in much smaller models, while marrying a deep mathematical toolkit with breakthroughs in deep learning.
Symbolic AI models offer significant benefits over current deep learning methods. With Symbolica’s toolkit, organizations and researchers can build custom symbolic models tailored to their specific needs, improving performance and reducing the environmental impact of AI research and development. Moreover, Symbolica’s models offer formally verified and provably correct outputs, helping to ensure that AI systems are making decisions based on accurate and reliable knowledge.