There have been many approaches one can use in solving data problems, and Neuro-Symbolic AI is the most recent addition to this arsenal. Deep Learning, as well as Symbolic AI, have been mainstays of data science technology, but Neuro-symbolic AI promises to transcend the limitations of both techniques. Neuro-Symbolic AI inherits the architecture of Deep Learning techniques and uses them with symbolic reasoning methods.
Symbolic AI and Deep Learning are both AI techniques based on the functioning of the human brain. Symbolic AI understands the data by forming internal symbolic representations which are very similar to how the human brain associates in thought and reasoning processes. Objects and even more abstract concepts are captured and then rules are created for them. These symbols are manipulated and human knowledge is represented in the form of symbols understood by humans. Because of this symbolic representation of the world, Symbolic AI understands the world through logic like humans and gain its reasoning ability
Deep Learning is more akin to the human brain in the way they are developed, with multiple neurons that form a neural network. Combining various weights, biases, and data inputs with each neuron serving a specific purpose, these elements come together and are used in solving machine learning problems including accurately classifying and describing objects within their data. These artificial neural networks can be extremely complex. Both technologies do, however, possess their own list of shortcomings.
In terms of Deep Learning, some of the issues faced are as follows:³
- Machine learning requires a massive amount of data to train neural networks, which is not easy to get every time.
- Selecting the right algorithm is crucial as the results may be biased and lead to a bad prediction.
- They lack the ability to generalize and are bound by their training data i.e. there is a lack of creativity and they are only efficient at what they already know.
- Even though many tried to explain how deep learning learns, deep learning is still a black box algorithm that can’t be explained.
With regard to Symbolic AI, the following limitations are observed:²
- Transferability-Symbolic AI made to tackle one problem may not be used to approach another problem due to the fixed set of rules.
- In the event of one assumption failing, there is a likelihood that all assumptions will fail, so Symbolic AI has an inherently brittle nature.
- Symbolic AI cannot process messy real-world data such as images, sounds. As you can’t define specific rules for ambiguous data, symbolic AI struggles in solving the problem
- Due to the fixed nature of the environmental variables, techniques employing Symbolic AI are not very robust to change.
Neuro-Symbolic AI uses Deep Learning to boost the Symbolic AI approach, and by combining logic and learning both limitations are transcended. Deep learning uses correlation but cannot use logic and this is where Symbolic AI comes in, it also adds value by filtering out irrelevant data. Neuro-Symbolic AI also learns with a much smaller training dataset, making data acquisition a lot easier ¹. Neuro-Symbolic AI is proven to solve much harder problems and is inherently more comprehensive in terms of decisions and actions. These are also easier to control.
This paradigm shift in AI technology is a step closer to emulating common sense present in humans. Thus Neuro-Symbolic AI is the latest stride in the advancement towards human-like intelligence in AI.
Communicating Knowledge. Saltlux
[1] Lawton, G. (2020, May 4). Neuro-symbolic AI emerges as powerful new approach. SearchEnterpriseAI. Retrieved September 29, 2021, from https://searchenterpriseai.techtarget.com/feature/Neuro-symbolic-AI-seen-as-evolution-of-artificial-intelligence
[2] Moskvitch, K. (2020, July 30). Neurosymbolic AI to give us machines with true common sense. Medium. Retrieved September 29, 2021, from https://medium.com/swlh/neurosymbolic-ai-to-give-us-machines-with-true-common-sense-9c133b78ab13
[3] What deep learning is and isn’t. The Data Scientist. (2020, November 26). Retrieved September 29, 2021, from https://thedatascientist.com/what-deep-learning-is-and-isnt/
