Google recently announced a significant development of artificial intelligence (AI) capable of handling many actions, including complex machine learning and reasoning. This technology is called Google PaLM (Path to Next Generation Language Models) Algorithm.
Researchers mentioned that the primary purpose of this AI is to outperform existing search engine AI and process human languages and reasoning. It provides a glimpse into the next generation of search.
Still, we all know that technology is not always 100% perfect or practical. There is always room for error and restrictions with large-scale language models, which can unintentionally lead to unethical outcomes. Thus, in this article, let’s look at the features of the Google PaLM Algorithm and how it would change AI and search.
What is the Google PaLM algorithm?
Google published an essay outlining the goals of Pathways, a new AI architecture. Since Google released a statement on developing the PaLM Algorithm, many people and agencies have been interested in this technology as it incorporates few shots learning. Google PaLM builds language models by training on pairs of words to predict the next term.
These models are trained on billions of sentences from various sources, such as books, news articles, and other online data.
Why is Google PaLM Interesting?
In October 2021, Google published an article outlining the goals of PaLM. The article explained that AI technology aims to serve as the turning point in researching and developing Google’s AI system. They intend to move away from their traditional AI system that can only do limited tasks.
They built PaLM as a way to solve problems concurrently by learning ways to solve them instead of having thousands of AI perform tons of tasks. It makes information processing and tasks quicker, easier, and more efficient.
The PaLM algorithm can handle millions of tasks, including complex learning and reasoning. In addition, Google stated that it could outperform existing AI state the art and humans in the language and reasoning tests.
Likewise, there are other capabilities of the PaLM Algorithm, as stated by Google on their article:
1. Deep Learning Capabilities
The deep learning capabilities of PaLM aim to imitate how humans learn together with diverse knowledge on how to tackle issues that it has never seen before. Google Brain Research, Hugo Larochelle described its deep learning capabilities as beyond deep learning.
He explained that PaLM could use all of its knowledge to solve new challenges, unlike the usual deep learning that requires collecting vast amounts of data and human labor. Hugo stated that deep learning would likely not be the path toward an AI that can solve many tasks. It is because each job requires millions of examples.
A good example is its ability to understand how aerial pictures can forecast landscape elevation for another project. Users can use this information to predict floodwaters for specific terrain.
Another excellent example of its deep learning capabilities is how it can explain a joke that it has never heard before. Below is an example demonstrating its capabilities from Google.
Prompt:
Explain this joke
Joke:
Did you see that Google just hired an eloquent whale for their TPU team? It showed them how to communicate between two different pods!
Model Response:
Prediction: TPUs are a type of computer chip that Google uses for deep learning. A “pod” is a group of TPUs. A “pod” is also a group of whales. The joke is that the whale is able to communicate between two groups of whales, but the speaker is pretending that the will is able to communicate between two groups of TPUs.
2. Inference Chaining
Inference chaining is a method in which inference rules are applied to present data. The AI evaluates existing data, derivation, and conditions to extract new data or information. This ability of PaLM can help users construct new answers and concepts from current ones.
Here’s an example of inference chaining, where multiple concepts and entities must be correctly joined and considered to create an answer.
Input:
Guido was writing a Python program. He keeps getting confused by variable scope and looking it up on StackOverflow. Is it possible that the Guido in the prompt refers to Guido van Rossum?
Model Output:
Guido van Rossum is the creator of Python. He would not need to look up variable scope on StackOverflow. The answer is “no.””
However, Hugo mentioned that it has superior deep learning capability, but it has its limit. It cannot overcome the AI’s limitations, including finding a way to mitigate the harmful spread of misinformation, discrimination, exclusion, and toxicity. These would ultimately result in negative ethical results.
3. Translating Languages
Google PaLM algorithm is better at translating another language to English and vice versa. It surpasses the state-of-the-art AI Technology of Google known as BIG-Bench, including 150 tasks related to answering questions, solving language, and reasoning capability.
However, it is noteworthy that it still failed in some areas. The researchers noted that human performance outperformed PaLM by 35%, particularly in mathematics-related tasks. Still, the researcher mentions that it succeeded in translation, a common problem that more multilingual data could solve.
4. Reasoning Ability
Likewise, PaLm’s arithmetic and common sense performance is another unique capability. Below is an example of its Arithmetic Tasks:
Question:
Roger has 5 tennis balls. He buys 2 more cans of tennis balls. Each can has 3 tennis balls. How many tennis balls does he have now?
Answer:
The answer is 11.
Likewise, here’s an example of commonsense reasoning of Palm as released by Google:
Question:
Sean was in a rush to get home, but the light turned yellow, and he was forced to do what?
Answer Choices:
(a) take time (b) dawdle (c) go slowly (d) ocean (e) slow down
Answer:
The answer is (e) slow down.
Is it the Next Generation Search Engine?
The abilities of PaLM mentioned above demonstrate how next-generation search engines can answer complex questions by utilizing existing data on the internet. Google Pathways and PaLm is an excellent way to achieve an AI architecture that can reflect the data and world around us.
And while this AI displays an exemplary performance, Google researchers emphasized that this is not the final word on AI and search. It is simply the first step toward the following kind of search engine technology.
So, stay tuned for more updates on Google Blog, and let us know what you think!
