IBM Canada watsonx User Group

 View Only

AI – A Russian Nesting Doll: Understanding the Difference and Choosing the Right Tool for the Job

By Cole Hickey posted Tue August 20, 2024 10:21 AM

  

The world of artificial intelligence (AI) is full of exciting possibilities, but the terminology can be confusing. Terms like AI, machine learning (ML), deep learning (DL), and generative AI (GenAI) are often used interchangeably, leaving many wondering what the true distinctions are and how they apply to real-world problems. 

GenAI surely has the most star power at the moment, but does that mean that traditional ML and DL techniques are no longer needed? Not necessarily. Today, I aim to clear up some of the confusion. We'll explore the unique characteristics of AI, ML, DL, and GenAI, and which technology is right for specific jobs in the enterprise.  

I like to think of the different terms as a Russian nesting doll, starting with AI as the all-encompassing term, you can think of AI, ML, DL and GenAI as a set of Russian dolls nested within each other. The largest, outermost doll being AI and the smallest, innermost doll being Foundational Models (GenAI), leaving ML & DL in the middle. 

 

 

AI: The Encompassing Vision 

Artificial intelligence is by no means a new concept. The birth of the AI conversation was initiated by Alan Turing in 1950 when he published a paper with the famous question,Can machines think?”. Think of AI as the outermost doll, representing the ambition of creating intelligent machines capable of mimicking human cognitive functions. This includes learning, reasoning, problem-solving, and decision-making. AI encompasses a wide range of techniques, from traditional rule-based systems to the more advanced algorithms inspired by the human brain. 

 

Machine Learning: Empowering Systems to Learn 

Moving to the next doll, machine learning (ML) resides within AI. ML algorithms equip systems with the ability to learn from data without explicit programming. The “learning” part of ML means that algorithms (or models) attempt to optimize along a certain dimension; typically, to minimize error or to maximize the likelihood their predictions will be correct. They can identify patterns and relationships within data, enabling them to make predictions or decisions on new, unseen data.  

 

Deep Learning: Diving Deeper with Neural Networks 

Deep learning (DL) is the next doll nestled within ML. Inspired by the structure and function of the human brain, DL algorithms rely on artificial neural networks. These networks consist of interconnected layers of nodes that process information in a similar way to biological neurons. Deep learning models excel at handling complex, high-dimensional data such as images, text, and audio. Prior to 2010, very little work was being done in DL because even shallow neural networks are extremely compute-intensive. It was only when three forces came together – the affordability of immense stores of labeled data; the invention of new deep learning algorithms; and cheaper, more powerful CPUs and GPUs – that deep learning moved from research to real world applications. Now, the way in which deep learning can be adopted and used has become a critical strategic consideration for every cognitive enterprise. 

 

Foundation Models & Generative AI: The Art of Creating New Data 

The innermost doll in our analogy represents generative AI (GenAI). Although not the only type, foundation models are what people typically think of when we think of generative AI (think ChatGPT, Google Gemini, IBM watsonx).  This branch of AI focuses on creating new and original content, such as images, text, or music. GenAI algorithms learn the underlying patterns and distributions within a data set. They then use this knowledge to generate entirely new data that closely resembles the training data. The major difference between GenAI and traditional AI (ML & DL mentioned above), is that foundational models are built on vast amounts of unlabeled data and that can be used for different tasks with minimal tuning. Whereas traditional AI methods are typically used to develop task specific models that are great at solving a single problem. 

 

Choosing the Right Tool 

Now that we have the terminology down, it’s time to classify the different terms. AI itself is simply the umbrella term for all of these technologies. ML & DL both flow into what we typically consider traditional AI, and then foundation models can be classified as GenAI. This brings us to a more popular conversation of when to use traditional vs GenAI. Both have their benefits, and one is not necessarily here to replace the other 

Traditional AI is very good at learning from specific tasks and executing those decision-making capabilities on new data for said tasks. Traditional AI typically lends itself to the following use cases: 

  • Predictive Analytics: Forecasting and predicting prices, maintenance schedules, market trends. 

  • Fraud Detection: Rule-based systems can identify fraudulent transactions by flagging outliers based on predefined criteria.  

  • Robotics: Industrial robots follow precise programmed movements for tasks like assembly line operations.  

Generative AI, being trained on MASSIVE amounts of unlabeled data from many sources typically excels at: 

  • Content Creation: Generating articles, blog posts, marketing copy, and social media content.  

  • Image and Video Generation: Creating realistic images, videos, and art.  

  • Summarization of Text: Summarizing large amounts of texts using NLP.  

  • Code Co-creation: Generating code based on a prompt in natural language. 

 

Conclusion 

GenAI is exciting and extremely powerful, but traditional AI is not going anywhere. Both can be revolutionary for enterprises. They can create time and cost savings in nearly every part of the business. Regardless of which acronym you choose to use in your business, what is most important is that we are constantly focused on trustworthy, transparent, and explainable AI projects. 

This is where a platform like IBM’s watsonx can inspire confidence in keeping AI initiatives on track. watsonx provides a studio where you can train, tune and deploy BOTH traditional and generative AI models. It also provides a fit-for-purpose data lake house built for AI workloads. Lastly, watsonx provides a governance layer that ensures explainability and transparency over both the data and AI pieces.  

AI can sometimes be overwhelming, but it can be a hugely useful tool in the enterprise. It is important to implement and deploy AI projects with a trusted partner that has experience in the world of AI, on a platform that can be trusted. 

1 comment
8 views

Permalink

Comments

Tue August 20, 2024 12:19 PM

Love this! Great analogy!