Why Neuro-symbolic AI is the future of AI: Here’s how it works
You thought AI is intelligent? Well, it has a long road ahead.
Without a doubt, artificial intelligence is one of the most revolutionary technologies being developed. Though, as powerful as it is, there still are a lot of basic problems it is yet to become competent to solve.
David Cox, Director of MIT-IBM Watson AI lab says, “It’s time to reinvent artificial intelligence.” And, according to him, neuro-symbolic AI is the answer.
We take a quick look into what ails present AI, and how AI engineers can revolutionize the discipline with neuro-symbolic AI.
Say, an AI program is asked to “Look at the picture below and tell if there are an equal number of large things and metal spheres?”
It will be impossible for a state-of-the-art AI neural network program to answer this simple question.
This may sound strange after an incredibly successful era of the 2010s, filled with breakthroughs and ostensibly no AI winter. But, that’s largely the reality.
Cox thinks it’s time to make AI smarter and more intelligent, and Neuro-Symbolic AI can accomplish this.
Strictly speaking, neuro-symbolic AI is not new. Essentially, it combines the two already existing approaches of AI, which once were pitted against each other. Those two are:
Let’s understand each.
👉Symbolic AI: Rule-based
This approach was the first official attempt at creating artificial intelligence.
It reigned supreme between the 1950s and 1980s.
Symbolic AI is based on humans’ ability to understand the world by forming symbolic interconnections and representations. The symbolic representations help us create the rules to define concepts and capture everyday knowledge.
This means, to explain something to a symbolic AI system, a symbolic AI engineer and researcher will have to explicitly provide every single information and rule that the AI can use to make a correct identification.
✔️Here’s how a symbolic AI would define an apple.
👉Neural Networks AI: Data-based
Neural network AI works differently from symbolic, as it is data-driven, instead of rule-based.
The ‘neuro’ aspect refers to deep learning neural networks. They are inspired by the human brain’s ability to compete. This is the latest tech in AI through which AI experts have inspired many AI breakthroughs. The ideation of self-driving cars — neural net. The concept of personal assistants like Alexa — neural nets.
How do they learn? Through data.
To train a neural network AI, you will have to show it numerous pictures of the subject in question. Once it is smart enough, it can not only identify the object for which it was trained but can also make similar objects that may not even exist in the real world. Think, the faces of people who never existed created by an AI.
✔️Here’s how a neural network AI would define an apple.
Divided AI approaches: Problems
Isolated approaches to AI are proving inadequate. Many cracks are starting to show in each of the AI approaches.
⇨ Problem with classical, symbolic AI is that it is limited to highly restricted domains. Symbolic AI breaks down when it is not explicitly programmed for something.
⇨For Neural Networks, its biggest limitation also powers it, data. Neural networks learn from examples, just like humans. However, while a human may require a couple of training examples to learn about an object, Neural AI requires a lot many more. Unless AI engineers are able to feed in huge amounts of annotated data, the accuracy of the algorithm remains weak. Due to this, while the neural AI can get 80% cases correct, it falls short on the remaining 20%; especially outlier or corner cases.
Learning from small data and fewer examples are what AI experts dub to be the future of artificial intelligence or advanced AI. Let’s see how a combination of both symbolic and neural network AI can achieve this.
Confluence of Symbolic with Neural Network AI
Coming together of neural AI with symbolic AI leads to an ecstatic combination of — learning and logic.
It makes smarter AI systems. Deep learning helps symbolic AI in breaking down the world into symbols, not by relying on human programmers, but data. Symbolic AI incorporates common sense, reasoning, and domain know-how into deep learning.
Together, symbolic and neural network approaches of AI can lead to significant advances — from self-driving cars to NLP. All this while, requiring fraction of data as it does today for training.
Neural networks help in getting answers from the messiness of the real-world data to a symbolic representation of the world, constituting correlations in images. Together, they can do some pretty magical things in reasoning. — David Cox, Director of MIT-IBM Watson AI Lab
Advantages of Neuro-Symbolic AI
Advantage 1: Higher Accuracy
Some may wonder, when neural networks can answer 80% cases correctly, that’s a good enough number for a machine. Why the remaining 20% is so important?
Consider this outlier case. A burning traffic light. While a human driver would understand to respond appropriately to a burning traffic light, how do you tell a self-driving car to act accordingly when there is hardly any data on it to be fed into the system. Neuro-symbolic AI can handle not just these corner cases, but other situations as well with fewer data, and high accuracy.
Advantage 2: Data Efficiency
The data required to train today’s AI systems is huge. When a human brain can learn with a few examples, artificial intelligence engineers require to feed thousands into an AI algorithm. Neuro-symbolic AI systems can be trained with 1% of the data that other methods require.
Advantage 3: Transparency and Interpretability
Not much discussed, this aspect of AI systems also puzzles AI experts. It can be often difficult to explain the decisions and conclusions reached by AI systems. It’s like a ‘black box’.
While why a bot recommends a certain song over other on Spotify is a decision a user would hardly be bothered about, there are certain other situations where transparency in AI decisions becomes vital for users. For instance, if one’s job application gets rejected by an AI, or a loan application doesn’t go through. Neuro-symbolic AI can make the process transparent and interpretable by the artificial intelligence engineers, and explain why an AI program does what it does.
Originally published here:
